AI Search Behaviour Statistics UK 2026

Section 3 — Follow-Up Questions, Multi-Step Research & Expanding Query Behaviour Statistics

One of the clearest differences between traditional Search and AI Search is that the interaction does not necessarily end after the first query.

Conversational AI allows the user to continue researching within the same session.

A typical journey can become:

Initial Question
→ Clarification
→ Follow-Up
→ Comparison
→ Recommendation
→ Decision

The user does not have to reformulate each stage as an entirely new Search.

Instead, the AI system can retain conversational context while the user progressively refines what they want.

This section examines measurable evidence of that change, including:

• queries per session;
• growth in repeated interaction;
• live-web Search usage;
• changes in prompt length;
• planning behaviour;
• and brainstorming or decision-support queries.

Data classification: Statistics 21–27 use US clickstream evidence from Semrush covering more than one billion lines of ChatGPT activity between October 2024 and February 2026. Statistics 28–30 use first-party Google data describing AI Mode behaviour in the United States and global AI Mode growth.

21. ChatGPT users averaged only 1.16–1.21 queries per session through most of 2025

US Observed Clickstream Benchmark. Semrush found that the average number of prompts submitted during a ChatGPT session remained relatively stable through much of 2025.

Average queries per session generally ranged between 1.16 and 1.21.

This suggests that many early ChatGPT interactions remained relatively simple.

Users frequently:

• asked one question;
• received an answer;
• and ended the session.

That behavioural pattern is closer to conventional Search than a deeply conversational research process.

However, the pattern changed sharply toward the end of the study period.

Behaviour implication: Early AI Search adoption does not automatically imply highly conversational behaviour.

Users can initially treat an AI system like a more advanced Search box before gradually learning to use follow-up questions and context.

Source:

Semrush — ChatGPT Traffic Analysis: Insights from 17 Months of Clickstream Data

22. Average ChatGPT queries per session reached 1.75 by February 2026

US Observed Clickstream Benchmark. By February 2026, Semrush found that the average number of queries per ChatGPT session had increased to 1.75.

This represents a meaningful behavioural shift.

Users were increasingly continuing the interaction after the first response.

Additional prompts can include:

• clarification;
• asking for alternatives;
• adding constraints;
• requesting comparison;
• challenging an answer;
• or asking what to do next.

This is one of the defining characteristics of conversational Search.

Behaviour implication: The relevant unit of AI Search Behaviour is increasingly the session rather than the individual query.

A brand can therefore appear:

• in the first response;
• only after a follow-up;
• during a comparison;
• or at the final recommendation stage.

Source:

Semrush — ChatGPT Search Insights 2026

23. ChatGPT queries per session increased 50% during the final four months of Semrush’s study

US Observed Clickstream Benchmark. Semrush reported that after approximately twelve months of relatively stable engagement, average queries per session increased by 50% during the final four months of its study.

The speed of the change is important.

User behaviour did not gradually increase at the same rate throughout the entire period.

Instead, repeated interaction accelerated sharply.

This may reflect:

• greater user familiarity;
• improved conversational capabilities;
• more advanced Search functionality;
• or users becoming increasingly comfortable treating ChatGPT as a research partner.

Behaviour implication: AI Search Behaviour can evolve rapidly even after initial platform adoption has matured.

Businesses should therefore expect prompt patterns and customer journeys to continue changing rather than assuming current behaviour is stable.

Source:

Semrush — 17-Month ChatGPT Clickstream Study

24. ChatGPT used live web search on 34.5% of queries in February 2026

US Observed Clickstream Benchmark. Semrush found that ChatGPT activated web-search functionality on approximately 34.5% of queries in February 2026.

This means most ChatGPT prompts in the study did not require a live-web Search interaction.

The user can therefore move between two distinct research modes:

• model-based conversation using existing model knowledge and context;
• live-web research where ChatGPT retrieves current external information.

The distinction is behaviourally important.

Users may begin with a general question and only trigger live-web research when the task becomes:

• current;
• commercial;
• specific;
• or evidence-dependent.

Behaviour implication: AI Search journeys can move between conversation and active web retrieval without the user changing platform.

That is fundamentally different from traditional Search, where external retrieval is the starting point of almost every query.

Source:

Semrush — ChatGPT Search Insights

25. ChatGPT’s web-search activation rate fell from around 46% to 34.5%

US Observed Clickstream Benchmark. Semrush reported that ChatGPT’s Search feature was activated on approximately 46% of queries in late 2024.

By February 2026, that had fallen to 34.5%.

The change is important because AI Search Behaviour is not simply moving toward more live-web searching.

Users are also increasingly using conversational AI for tasks that can be completed without active external retrieval.

Those tasks can include:

• reasoning;
• explaining;
• summarising existing context;
• brainstorming;
• planning;
• and transforming information.

Behaviour implication: AI Search expands the meaning of Search beyond retrieving webpages.

The system can function as both:

Search Engine

and:

Research Interface.

Source:

Semrush — ChatGPT Clickstream Analysis

26. Search-enabled ChatGPT prompts nearly doubled in length from 4.7 to 8.7 words

US Observed Clickstream Benchmark. Semrush compared ChatGPT prompts from January–February 2025 with those from January–February 2026.

For prompts that triggered ChatGPT’s web-search functionality, average length increased from:

• 4.7 words in early 2025
• to 8.7 words in early 2026.

The average Search-enabled prompt therefore nearly doubled in length within one year.

This suggests users increasingly provide greater context when asking ChatGPT to search the web.

Rather than entering:

“best hotels Marbella”

a user may increasingly ask:

“What are the best quiet luxury hotels near Marbella for a three-night couple’s break?”

Behaviour implication: Even when AI systems retrieve from the live web, user behaviour is becoming more conversational and contextual than conventional keyword Search.

The relevant optimisation target increasingly includes:

• needs;
• constraints;
• use cases;
• audiences;
• and decision context.

Source:

Semrush — ChatGPT Prompt-Length Analysis

27. Non-search ChatGPT prompts fell from 24.9 to 13.5 words on average

US Observed Clickstream Benchmark. Over the same period, Semrush found the opposite trend among prompts that did not trigger web Search.

Average non-search prompt length fell from:

• 24.9 words in January–February 2025
• to 13.5 words in January–February 2026.

This narrowing difference suggests users are becoming more efficient in how they communicate with AI systems.

Earlier users may have provided lengthy instructions because they were uncertain about what the model required.

As familiarity grows, users can learn when:

• a short follow-up is enough;
• previous context already explains the task;
• or the model can infer the next stage of the conversation.

For example, a follow-up may be as simple as:

“Compare the first two.”

The system already has the earlier conversational context.

Behaviour implication: Longer individual prompts are not the only sign of conversational Search.

Context retention can make later prompts shorter because the conversation itself contains the additional information.

Source:

Semrush — 17 Months of ChatGPT Usage Data

28. Google AI Mode queries have more than doubled every quarter since launch

Global Platform Behaviour Data. In May 2026, Google reported that AI Mode queries had more than doubled every quarter since the product launched.

Google also reported that AI Mode had surpassed one billion monthly active users globally.

The rapid growth indicates increasing consumer willingness to use a conversational Search interface for tasks that conventional Search may previously have divided across multiple queries.

AI Mode is specifically designed to support:

• complex questions;
• follow-up questions;
• comparisons;
• research;
• planning;
• and continued conversation.

Behaviour implication: Conversational Search is scaling rapidly inside a mainstream Search ecosystem, not only within standalone chatbot products.

This significantly increases the likelihood that multi-step Search Behaviour becomes normalised among a broader population.

Source:

Google — How AI Mode Is Changing and Expanding the Way People Search

29. Planning-related AI Mode queries grew 80% faster than AI Mode queries overall

US AI Mode Behaviour Data. Google reported in May 2026 that AI Mode queries associated with planning had grown 80% faster than AI Mode queries overall during the previous six months.

Planning is a particularly important category of Search Behaviour because it normally requires multiple pieces of information to be combined.

Examples can include:

• planning a holiday;
• organising an event;
• building an itinerary;
• choosing products for a project;
• planning a purchase;
• or deciding between several alternatives.

Traditional Search may require a sequence of independent queries and website visits.

Conversational AI can bring much of that activity into one continuous interaction.

Behaviour implication: AI Search appears particularly suited to multi-stage tasks rather than simple fact retrieval alone.

This expands Search from information discovery toward decision support.

Source:

Google — AI Mode Search Behaviour Insights, May 2026

30. Brainstorming queries in AI Mode grew 30% faster than overall AI Mode usage

US AI Mode Behaviour Data. Google reported that brainstorming-related AI Mode queries had grown 30% faster than AI Mode queries overall since launch.

Google also observed increasing searches beginning with phrases such as:

• “where to”;
• “where should I”;
• and “ideas for”.

These query structures indicate a behavioural move away from users always knowing exactly what they want before they search.

Instead, consumers can increasingly ask AI to help:

• define the options;
• generate possibilities;
• narrow the field;
• and structure the decision.

This changes the role of Search.

Traditional Search often begins after the user has formulated the requirement.

AI Search can participate in formulating the requirement itself.

Behaviour implication: Organisations may need visibility earlier in the customer journey, when the user is still deciding:

“What should I consider?”

rather than only:

“Where can I buy it?”

Source:

Google — AI Mode Behaviour Research 2026

Section 4 — Trust, Verification & Source-Checking Behaviour Statistics

AI Search creates a different trust relationship from traditional Search.

With a conventional search engine, users normally see multiple links and decide which source to open.

With generative AI, the system often synthesises those sources into one direct answer.

This can make Search faster, but it also changes where trust is placed.

The user may be trusting:

• the AI platform;
• the sources selected by the AI;
• the accuracy of the synthesis;
• and the model’s interpretation of the question.

The following statistics examine how much UK consumers trust AI Search, whether frequent users behave differently, and what people do to verify or improve the answers they receive.

Data classification: Statistics 31–40 primarily use nationally representative UK consumer research conducted by Yonder for Which? in September 2025, based on 4,189 UK adults. The measures relate to reported consumer behaviour and attitudes rather than observed clickstream behaviour.

31. 47% of UK AI Search users trust AI results to a reasonable or great extent

UK Consumer Data. Which? found that 47% of AI Search users trust the information returned by AI Search tools to either a reasonable or great extent.

This means almost half of users place substantial trust in AI-generated answers.

That level of confidence is important because AI Search can present information in a fluent, authoritative and complete-looking format.

The behavioural risk is that presentation quality can sometimes be interpreted as evidence quality.

Behaviour implication: Organisations appearing within AI answers are operating inside an environment where many users place meaningful trust in the generated response itself.

This increases the importance of:

• accurate information;
• clear sourcing;
• current data;
• and authoritative digital evidence.

Source:

Which? — Consumer Use and Attitudes Towards AI Search Tools

32. A further 35% of UK AI Search users trust the results to some extent

UK Consumer Data. Which? found that another 35% of users said they trust AI Search results to some extent.

When combined with users reporting reasonable or great trust, the evidence shows that only a minority of users express very low trust.

However, “some trust” should not be confused with complete confidence.

It may indicate that users see AI as:

• a useful starting point;
• a convenient summary tool;
• or an initial research assistant

while still recognising the need for further checking.

Behaviour implication: AI Search is often being used within a conditional trust model.

Users may accept the answer provisionally while reserving judgment until additional research is completed.

Source:

Which? — UK AI Search Trust Research

33. Only 15% of UK AI Search users report little or no trust in AI Search

UK Consumer Data. Which? found that 15% of AI Search users said they had little or no trust in the results generated by AI Search tools.

This is substantially smaller than the population expressing at least some trust.

The figure is significant because AI Search systems are still capable of:

• factual mistakes;
• outdated information;
• misinterpretation;
• or incomplete sourcing.

Yet explicit distrust remains relatively uncommon among people who already use the technology.

Behaviour implication: Once a consumer adopts AI Search, the default behaviour may often be to engage with the answer rather than reject it outright.

This places greater importance on the accuracy of information the AI ecosystem can retrieve about organisations, products and services.

Source:

Which? — AI Search Consumer Survey

34. 64% of frequent AI Search users trust the tools to a reasonable or great extent

UK Consumer Data. Trust increases substantially among people who use AI Search frequently.

Which? found that 64% of frequent users trust AI Search tools to a reasonable or great extent.

Among less frequent users, the equivalent figure was only 31%.

The relationship may operate in both directions.

People who trust AI more may use it more frequently.

At the same time, repeated successful use may increase familiarity and confidence.

The survey does not establish which effect is dominant.

Behaviour implication: Trust and habitual usage appear strongly associated.

As consumers become frequent users, AI-generated answers may play a greater role in:

• research;
• comparison;
• shortlisting;
• and decision-making.

Source:

Which? — Consumer AI Search Trust Data

35. Around 70–72% of ChatGPT, Gemini and Copilot users trust those individual tools a fair amount or a lot

UK Consumer Data. When consumers were asked about individual AI platforms, Which? found broadly similar levels of trust among the three most popular tools.

Approximately 70% to 72% of ChatGPT, Gemini and Microsoft Copilot users said they trusted the individual tool a fair amount or a lot.

This suggests that trust is not concentrated solely around the market-leading platform.

Users can develop confidence in several different AI ecosystems.

Behaviour implication: Platform switching does not necessarily imply low trust.

Consumers may trust multiple AI systems while selecting between them according to:

• task;
• convenience;
• integration;
• or previous experience.

Source:

Which? — Trust Levels of Popular AI Search Tools

36. Only 60% of Meta AI users trusted the platform a fair amount or a lot

UK Consumer Data. Which? found that trust in Meta AI was lower than for ChatGPT, Gemini and Copilot.

Approximately 60% of Meta AI users said they trusted the platform a fair amount or a lot.

That still represents a majority of users, but it is around 10 percentage points below the trust levels recorded for several competing AI platforms.

This demonstrates that consumers do not necessarily assign identical levels of confidence to every AI system.

Behaviour implication: AI Search Behaviour is influenced not only by whether a platform is available, but also by how trustworthy users perceive it to be.

Platform-level trust could therefore affect:

• frequency of use;
• willingness to follow recommendations;
• and reliance on generated answers.

Source:

Which? — AI Search Platform Trust

37. Only 30% of UK AI Search users believe AI tools are independent or unbiased to a reasonable or great extent

UK Consumer Data. Which? found that only 30% of AI Search users believed the tools were independent or unbiased to a reasonable or great extent.

More than half — 55% — believed AI Search tools were either:

• not unbiased at all;
• or unbiased only to some extent.

A further 14% said they did not know whether AI Search tools were independent or unbiased.

This creates an interesting contrast.

Consumers may trust an AI answer enough to use it while remaining sceptical about whether the system is fully neutral.

Behaviour implication: Trust does not necessarily mean users perceive AI systems as independent arbiters.

Consumers can simultaneously find an AI answer useful while questioning:

• why particular sources were selected;
• why certain products were recommended;
• or whether commercial or platform factors influenced the response.

Source:

Which? — AI Search Independence and Bias Research

38. 55% of UK AI Search users ask follow-up questions to improve answer quality

UK Consumer Data. Which? found that 55% of AI Search users ask follow-up questions to a reasonable or great extent in order to improve the quality of the results they receive.

This provides direct evidence of active verification and refinement behaviour.

Users are not always treating the first AI-generated response as final.

They may instead ask:

• “Are you sure?”
• “What is the source?”
• “Explain that further.”
• “What are the alternatives?”
• “Compare these options.”
• or “What evidence supports that?”

The behaviour demonstrates one of the key advantages of conversational Search:

the answer itself can become the starting point for the next Search.

Behaviour implication: AI Search users can actively improve, challenge and refine generated answers without leaving the conversation.

This makes Search Behaviour increasingly iterative rather than linear.

Source:

Which? — Follow-Up Question Behaviour

39. 73% of frequent AI Search users use follow-up questions to improve results

UK Consumer Data. Follow-up behaviour is substantially stronger among frequent AI Search users.

Which? found that 73% of frequent users ask follow-up questions to a reasonable or great extent.

Among less frequent users, the figure was only 39%.

This difference suggests that people become more sophisticated in how they interact with AI as usage increases.

Frequent users may learn that the first response does not need to be the final response.

Instead, they can progressively improve the answer by:

• adding context;
• challenging assumptions;
• changing criteria;
• requesting evidence;
• or asking for alternatives.

Behaviour implication: Greater AI Search experience appears associated with more interactive Search Behaviour.

The mature AI Search user is increasingly:

Questioning + Refining + Comparing

rather than simply accepting one generated answer.

Source:

Which? — Frequent AI Search User Behaviour

40. 23% of UK AI Search users often or always cross-check one AI tool against another

UK Consumer Data. Which? found that 23% of AI Search users often or always cross-reference the results from one AI Search tool with another AI platform.

The behaviour becomes considerably more common among frequent users.

Among frequent AI Search users, 36% often or always cross-check one AI system against another.

Among less frequent users, the figure is only 11%.

This means a significant minority of advanced AI users are already behaving as multi-model researchers.

A consumer might:

Ask ChatGPT
↓
Check Gemini
↓
Compare the Answers
↓
Search Google
↓
Reach a Decision

This can make inconsistencies between platforms highly visible to the user.

Behaviour implication: Organisations increasingly need consistent digital evidence across the wider Search ecosystem.

If one AI platform describes a business differently from another, sophisticated users may encounter that inconsistency during the same research journey.

Which? separately found that 34% of surveyed consumers believed AI drew upon authoritative sources, reinforcing the importance of ensuring authoritative and accurate source information is available for AI systems to retrieve.

Source:

Which? — AI Search Cross-Checking Behaviour

Section 5 — AI Search vs Traditional Search & Professional Advice Behaviour Statistics

AI Search is not simply competing with Google.

It is increasingly competing with several different forms of information seeking, including:

• traditional Search;
• professional advice;
• comparison websites;
• publisher content;
• and direct brand research.

This makes one behavioural question particularly important:

When do people treat AI as an additional research tool, and when do they use it instead of another source?

The evidence suggests that the answer varies considerably according to:

• the subject;
• the user’s age;
• the perceived complexity of the decision;
• and the stage of the customer journey.

Data classification: Statistics 41–46 use nationally representative UK consumer data from Which?. Statistics 47–49 use observed US downstream behaviour from Similarweb. Statistic 50 uses US ChatGPT clickstream data from Semrush.

41. 19% of UK AI Search users often or always use AI instead of seeking healthcare advice

UK Consumer Data. Which? found that 19% of AI Search users often or always use AI-generated Search results as a substitute for seeking advice from a healthcare professional.

This means almost one in five AI Search users reported sometimes replacing professional healthcare advice with an AI-generated response.

Healthcare is a particularly important category because decisions can involve:

• symptoms;
• medications;
• treatment;
• diagnosis;
• and other high-stakes information.

The behaviour therefore goes substantially beyond casual information discovery.

Behaviour implication: Some consumers are beginning to treat AI systems not merely as Search tools, but as substitutes for traditional expert-information pathways.

This increases the importance of authoritative and clearly sourced health information being available to AI systems.

Source:

Which? — Consumer Use and Attitudes Towards AI Search Tools

42. 17% of UK AI Search users often or always use AI instead of seeking financial advice

UK Consumer Data. Which? found that 17% of AI Search users often or always use generative AI as a substitute for professional financial advice.

Consumers can ask AI systems about:

• investments;
• banking;
• budgeting;
• tax;
• insurance;
• credit;
• and personal financial decisions.

These interactions may occur before, alongside or instead of professional financial guidance.

Which? also reported that finance and money management was one of the limited areas where users said they placed greater reliance on AI than traditional Search.

Behaviour implication: AI Search can move beyond information retrieval into decision support in financially consequential areas.

For financial organisations, accurate digital evidence and current product information become particularly important because AI systems may help shape decisions before a user reaches a provider.

Source:

Which? — UK AI Search Consumer Survey

43. 13% of UK AI Search users often or always use AI instead of seeking legal advice

UK Consumer Data. Which? found that 13% of AI Search users often or always use AI Search instead of seeking professional legal advice.

The percentage is lower than for healthcare and finance, but still represents more than one in ten AI Search users.

Legal research often involves complex context, jurisdiction and individual circumstances.

Nevertheless, users are already asking generative AI systems to help explain:

• rights;
• contracts;
• disputes;
• legal terminology;
• and procedural questions.

Behaviour implication: Consumers increasingly use AI within high-consideration research journeys where inaccurate or incomplete information can have meaningful consequences.

This reinforces the importance of trusted primary and professional sources within the AI information ecosystem.

Source:

Which? — Professional Advice Substitution Research

44. 30% of UK adults aged 18–34 often or always use AI instead of healthcare advice

UK Consumer Data. Professional-advice substitution is substantially more common among younger users.

Which? found that 30% of adults aged 18–34 said they often or always use AI instead of seeking advice from a healthcare professional.

That compares with 19% among AI Search users overall.

The age difference suggests that younger consumers may be more willing to treat AI as an initial decision-support environment for important subjects.

Behaviour implication: Age affects not only whether consumers use AI Search, but also how much responsibility they are willing to give it within the decision journey.

For younger audiences, AI may increasingly function as:

Initial Research → Explanation → Decision Support

before any professional contact occurs.

Source:

Which? — AI Search Behaviour by Age

45. 28% of UK adults aged 18–34 often or always use AI instead of professional financial advice

UK Consumer Data. Which? found that 28% of adults aged 18–34 often or always use AI Search instead of seeking professional financial advice.

That is more than one quarter of younger adults within the relevant survey group.

The behaviour may be encouraged by the ease with which AI systems can explain:

• financial terminology;
• product differences;
• budget scenarios;
• investment concepts;
• and potential options.

The user can also ask unlimited follow-up questions without the friction involved in arranging a professional consultation.

Behaviour implication: Convenience can make AI an increasingly important early-stage financial research environment.

Professional providers may therefore encounter prospective customers later in the journey, after AI has already helped shape their understanding and expectations.

Source:

Which? — Consumer AI Search Research

46. 22% of UK adults aged 18–34 often or always use AI instead of professional legal advice

UK Consumer Data. Which? found that 22% of adults aged 18–34 reported often or always using AI instead of professional legal advice.

This compares with 13% across AI Search users overall.

Across all three professional categories measured by Which?, younger adults showed greater willingness to substitute AI for professional guidance:

• Healthcare — 30%
• Financial Advice — 28%
• Legal Advice — 22%

Behaviour implication: The transition toward AI-assisted decision-making is happening more quickly among younger consumers.

For organisations in regulated or high-trust sectors, visibility within AI Search may therefore increasingly influence the customer’s understanding before direct professional interaction begins.

Source:

Which? — Professional Advice and AI Search

47. Consumers recommended a brand by ChatGPT were 2.5 times more likely to visit that brand within seven days

US Observed Behaviour Benchmark. Similarweb tracked thousands of real user journeys across Finance, Travel and Beauty.

The study followed users who:

• asked ChatGPT an industry-related question;
• received a specific brand recommendation;
• did not click directly;
• had not previously visited the brand;
• and had not named the brand in their original prompt.

Users receiving the recommendation were 2.5 times more likely to visit the recommended brand’s website within the following seven days than users who received a competing recommendation.

This is important because it demonstrates that AI can change later Search Behaviour even when no immediate click occurs.

Behaviour implication: A recommendation can create:

AI Awareness → Memory → Later Search → Website Visit

rather than an immediately attributable referral.

Source:

Similarweb — Being Recommended in AI Makes Users 2.5x More Likely to Visit Your Site

48. 55.9% of AI-influenced brand visits arrived through Search

US Observed Behaviour Benchmark. Similarweb found that 55.9% of visits generated after an AI recommendation ultimately arrived at the brand’s website through Search.

This is one of the clearest demonstrations of the relationship between AI Search and traditional Search.

The user can first discover the organisation through ChatGPT but later use Google or another Search engine to navigate to it.

The AI system influences the decision.

The Search engine receives the final navigation query.

Behaviour implication: AI and traditional Search increasingly operate sequentially rather than competitively.

A common customer journey can become:

AI Recommendation
↓
Brand Recognition
↓
Traditional Search
↓
Website

Source:

Similarweb — The Downstream Impact of AI Visibility

49. Only 40.4% of comparable non-AI-influenced visits arrived through Search

US Comparative Behaviour Benchmark. Similarweb found that Search accounted for approximately 40.4% of visits among users without the measured AI recommendation influence.

That compares with:

• 55.9% among AI-influenced visitors;
• 40.4% among non-AI-influenced visitors.

The difference suggests that users exposed to an AI recommendation become particularly likely to conduct a subsequent Search.

This may happen because the AI interaction introduces a brand but the user still wants to:

• verify it;
• find the official website;
• read reviews;
• check prices;
• or conduct additional research.

Behaviour implication: Traditional Search can become a verification and navigation layer following AI discovery.

This is another reason AI Search should not be viewed simply as a replacement for Google.

Source:

Similarweb — AI-Influenced Search and Homepage Traffic

50. 21.6% of ChatGPT outbound referral traffic went directly to Google in February 2026

US Observed Clickstream Benchmark. Semrush found that by February 2026, Google accounted for 21.6% of all outbound referral traffic from ChatGPT.

Google was the largest individual destination for ChatGPT outbound referrals in the study.

This provides another form of behavioural evidence that consumers frequently move between AI Search and conventional Search.

ChatGPT users may use Google to:

• verify a recommendation;
• find a business website;
• research a product;
• look for reviews;
• perform a local Search;
• or continue research through a familiar interface.

Behaviour implication: The emerging Search Journey is not:

Google → Replaced by ChatGPT

but increasingly:

ChatGPT ↔ Google

with users moving between both environments according to what they are trying to achieve.

Source:

Semrush — ChatGPT Traffic Analysis: Insights from 17 Months of Clickstream Data

Section 6 — Product, Brand & Shopping Research Behaviour Statistics

Commercial Search Behaviour is one of the areas where generative AI is beginning to change the customer journey most visibly.

Consumers can now ask AI systems to:

• research products;
• compare alternatives;
• identify suitable brands;
• find deals;
• suggest gifts;
• build shopping lists;
• and narrow large numbers of choices into a manageable shortlist.

This changes the traditional ecommerce discovery model.

Instead of starting with a retailer, marketplace or Search engine, the customer can increasingly begin with a conversational research request.

Data classification: Statistics 51–52 use UK consumer research from Adobe Digital Insights. Statistics 53–60 use US consumer benchmarks from Adobe and are clearly identified as such. These benchmarks are included to illustrate emerging shopping behaviour rather than being presented as direct measurements of the UK population.

51. 35% of UK consumers had already used generative AI to assist with shopping by March 2025

UK Consumer Data. Adobe Digital Insights reported that 35% of UK consumers surveyed in March 2025 had already used generative AI tools to help with shopping.

This means more than one third of surveyed UK consumers had incorporated generative AI somewhere within the retail journey.

AI-assisted shopping can include:

• product research;
• comparison;
• gift inspiration;
• price research;
• and identifying suitable products.

The significance is not that AI has replaced ecommerce Search.

It is that a meaningful proportion of consumers are now introducing an AI research layer before visiting the retailer.

Behaviour implication: Product discovery can increasingly begin outside the retailer’s own website, marketplace or conventional Search results.

The retailer may therefore enter the customer journey only after an AI system has already helped define the shortlist.

Source:

Adobe Digital Insights UK — Retail in Flux

52. 47% of UK consumers said they planned to use generative AI for shopping during 2025

UK Consumer Data. The same Adobe research found that 47% of UK consumers planned to use generative AI to assist with shopping during 2025.

That was substantially higher than the 35% who had already used it when surveyed in March.

The difference suggested significant additional adoption intent.

Consumers were increasingly considering AI for:

• finding products;
• comparing choices;
• generating ideas;
• and making purchase research more efficient.

Behaviour implication: AI-assisted shopping was already moving beyond early adopters and into broader consumer consideration.

For retailers and brands, the emerging visibility question becomes:

“Does the organisation appear when an AI system helps a consumer decide what to consider?”

Source:

Adobe Digital Insights UK — Retail in Flux

53. 39% of US consumers were already using AI for online shopping by February 2025

US Consumer Benchmark. Adobe reported that 39% of consumers surveyed in February 2025 were already using AI for online shopping.

A further 14% said they expected to adopt AI-assisted shopping soon.

The result suggests that commercial use of AI was moving rapidly from experimentation toward mainstream research behaviour.

The shopping journey can increasingly begin with:

Need
↓
AI Research
↓
Product Shortlist
↓
Retailer or Brand

rather than beginning directly on an ecommerce site.

Behaviour implication: AI increasingly participates at the earliest stages of commercial discovery.

Source:

Adobe Digital Insights — The Explosive Rise of Generative AI Referral Traffic

54. Millennials recorded 46% adoption of AI-assisted online shopping

US Consumer Benchmark. Adobe found that 46% of Millennials surveyed were already using AI for online shopping.

A further 12% expected to use it by the end of 2025.

This meant 58% of surveyed Millennials had either:

• already adopted AI-assisted shopping;
• or expected to do so within the year.

Among higher-income Millennials earning at least $70,000, Adobe reported AI-assisted shopping adoption above 50%.

Behaviour implication: AI-assisted commercial Search Behaviour is particularly established among digitally experienced consumers with significant purchasing power.

For businesses targeting these groups, AI visibility can influence the customer journey before traditional ecommerce channels receive the visit.

Source:

Adobe Digital Insights — Generative AI Shopping Adoption

55. 72% of AI-shopping users rely on AI as a primary tool for researching products and brands

US Consumer Benchmark. Among consumers already using AI platforms for shopping, Adobe found that 72% relied on AI as a primary tool for researching products and brands.

This is one of the most significant behavioural findings in the shopping dataset.

AI is not simply being used at the edge of the purchase journey.

For many existing users, it is becoming one of the main environments in which research takes place.

Users can ask:

• which brands fit a requirement;
• which products have particular features;
• what alternatives exist;
• which option fits a budget;
• and what should be considered before purchase.

Behaviour implication: Product Search increasingly involves AI-mediated consideration sets.

The brands an AI system includes in its initial answer may gain an important advantage because they become part of the consumer’s shortlist before the website visit occurs.

Source:

Adobe — Generative AI Consumer Shopping Research

56. 47% of AI shoppers use generative AI to get product recommendations

US Consumer Benchmark. Adobe found that 47% of consumers using AI for shopping use the technology to receive product recommendations.

Recommendation behaviour is fundamentally different from conventional keyword Search.

A traditional query may ask:

“running shoes men”

An AI recommendation request can ask:

“What running shoes would you recommend for a heavier runner doing mostly road running with occasional knee pain?”

The second query contains far more information about:

• the consumer;
• the use case;
• the problem;
• and the desired outcome.

Behaviour implication: AI Shopping Search increasingly moves from:

Find Products

toward:

Recommend the Most Suitable Product.

That is a significant shift for brands because visibility depends not only on relevance to a category but on relevance to a specific user context.

Source:

Adobe Digital Insights — AI Shopping Use Cases

57. 43% of AI shoppers use generative AI to find deals

US Consumer Benchmark. Adobe found that 43% of AI-shopping users use generative AI to find deals.

This behaviour moves AI Search directly into price and value discovery.

Consumers can ask systems to help identify:

• discounts;
• lower-priced alternatives;
• better-value products;
• promotional offers;
• or options within a specific budget.

This creates a more explicit commercial research role for AI.

The customer is no longer merely asking:

“What exists?”

They may be asking:

“Which option gives me the best value?”

Behaviour implication: AI Search can participate in both discovery and price evaluation before the customer reaches a retailer.

For businesses, accurate pricing, product data and offer information therefore become increasingly important inputs into AI-assisted discovery.

Source:

Adobe — AI Shopping and Deal Research

58. 35% of AI shoppers use generative AI for gift ideas

US Consumer Benchmark. Adobe found that 35% of AI-shopping users use generative AI to generate gift ideas.

Gift research is particularly suited to conversational Search because the consumer can describe the recipient rather than specify the product.

A prompt can include:

• age;
• relationship;
• interests;
• occasion;
• budget;
• and previous purchases.

The AI system then helps transform an undefined requirement into potential products.

Behaviour implication: AI Search can influence the customer journey before a product category has even been chosen.

This is materially different from traditional product Search, where the user often begins with some idea of what they intend to purchase.

The journey becomes:

Person or Occasion → AI Ideas → Product Category → Product → Brand

Source:

Adobe Digital Insights — AI-Assisted Gift Discovery

59. 33% of AI shoppers use generative AI to create shopping lists

US Consumer Benchmark. Adobe found that 33% of AI-shopping users use generative AI to create shopping lists.

This illustrates how AI can influence multiple items within one commercial interaction.

A consumer might ask an AI system to build:

• a home-office setup;
• a holiday packing list;
• a new-baby checklist;
• a kitchen equipment list;
• a skincare routine;
• or products needed for a particular project.

Instead of searching individually for every product, the AI system can create an entire category of potential purchases.

Behaviour implication: AI can participate in basket formation, not simply individual product discovery.

This creates new opportunities for brands to appear within broader use-case and task-based queries.

The Search opportunity therefore extends beyond:

Product Keyword

to:

Problem + Project + Occasion + Intended Outcome.

Source:

Adobe — Generative AI Shopping Behaviour

60. 44% of consumers said they were likely to use AI for entertainment and media shopping research

US Consumer Benchmark. Adobe asked consumers which retail categories they would be likely to use AI assistants for when seeking inspiration, conducting research or shopping.

The highest response was entertainment and media at 44%.

This was followed by:

• clothing — 41%;
• health and beauty — 34%.

The results demonstrate that AI-assisted shopping is not confined to one type of purchase.

Consumers are considering the technology across both:

• functional purchases;
• and more subjective preference-driven purchases.

This is important because recommendation systems must increasingly interpret qualities such as:

• taste;
• style;
• suitability;
• use case;
• and personal preference.

Behaviour implication: AI Shopping Search is expanding from simple factual product lookup into categories where discovery and recommendation have historically depended heavily on publishers, retailers, reviews and personal judgment.

Source:

Adobe Digital Insights — Consumer AI Shopping Research

Section 7 — AI Recommendations, Purchase Confidence & Decision Behaviour Statistics

AI Search becomes commercially important when it begins to influence not only what consumers research, but what they eventually choose.

The transition can move through several stages:

Discovery
↓
Research
↓
Recommendation
↓
Comparison
↓
Confidence
↓
Purchase

At each stage, an AI system can shape:

• which brands are considered;
• which products are shortlisted;
• which features matter;
• how alternatives are compared;
• and whether the customer feels sufficiently confident to buy.

The following statistics examine satisfaction with AI recommendations, click behaviour, purchase confidence, complex-purchase research and consumers’ growing willingness to rely on AI during commercial decision-making.

Data classification: Statistics 61–64 use Adobe’s 2025 US Holiday Consumer Survey of more than 1,000 respondents alongside Adobe Analytics. Statistics 65–66 use Adobe’s earlier US consumer research involving 5,000 respondents. Statistics 67–70 use Adobe and Oxford Economics’ global 2026 customer survey of 4,000 consumers across North America, Latin America, Europe, APAC and the Middle East, including UK respondents.

61. 64% of AI-shopping consumers are satisfied with the links AI assistants provide

US Consumer Benchmark. Adobe found that 64% of consumers using AI for online shopping reported being satisfied with the links provided by AI assistants.

This matters because the quality of the link is part of the recommendation experience.

Consumers are not only judging:

• the wording of the AI answer;
• but also where the AI sends them next.

A useful recommendation therefore combines:

• relevant advice;
• a suitable brand or product;
• and a useful destination for further research or purchase.

Behaviour implication: AI Search increasingly acts as a filter between the consumer and the open web.

When users are satisfied with the sources and links selected for them, they have less reason to restart the entire research process elsewhere.

Source:

Adobe Digital Insights — AI-Driven Traffic Surges Across Industries

62. More than 55% of AI-shopping consumers actively click links recommended by AI assistants

US Consumer Benchmark. Adobe reported that more than 55% of consumers using AI for online shopping actively click the links AI assistants provide.

This is important because AI Search is often described as a purely zero-click environment.

The evidence shows a more nuanced pattern.

Some AI interactions end within the answer.

Others create downstream activity when the user wants to:

• inspect the product;
• verify the recommendation;
• check a price;
• read more detail;
• or complete the purchase.

Behaviour implication: AI-generated answers can operate as both:

Destination

and:

Gateway.

The commercial opportunity therefore extends beyond being mentioned in the AI answer to also being selected as the source or destination behind it.

Source:

Adobe Digital Insights — AI-Sourced Traffic Insights 2025

63. 65% of consumers using AI for online shopping say AI makes them more confident in their purchase

US Consumer Benchmark. Adobe found that 65% of consumers using AI for online shopping said they felt more confident in their eventual purchase after receiving help from AI.

This shows that AI is beginning to influence the psychological stage immediately before conversion.

Purchase confidence can improve when an AI system helps a consumer:

• understand the differences between products;
• identify relevant features;
• eliminate unsuitable alternatives;
• compare value;
• and answer remaining questions.

The AI interaction can therefore reduce uncertainty.

Behaviour implication: AI Search increasingly functions as a decision-confidence layer.

A consumer may reach a retailer website already feeling that much of the research has been completed.

Source:

Adobe Digital Insights — Holiday 2025 Consumer Research

64. 68% of AI-shopping consumers say they are less likely to return a product after AI has helped with the purchase

US Consumer Benchmark. Adobe found that 68% of consumers using AI for online shopping reported that they were less likely to return a product after using AI during the purchase process.

This may indicate that improved research helps consumers choose products that fit their requirements more closely.

AI can help narrow:

• size;
• features;
• budget;
• compatibility;
• use case;
• or product suitability

before the order takes place.

Adobe separately reported that overall online returns during the 2025 holiday season were down 1.2% year over year, although that broader movement should not be attributed solely to AI.

Behaviour implication: AI-assisted Search may influence not only whether a customer buys, but whether the final choice better matches the customer’s requirements.

Source:

Adobe Digital Insights — AI Recommendations and Purchase Confidence

65. 87% of AI-shopping users are more likely to use AI for large or complex purchases than for inexpensive purchases

US Consumer Benchmark. Adobe’s consumer research found that 87% of users were more likely to use AI for large or complex purchases, such as electronics, than for less expensive purchases.

This behaviour is particularly significant.

It suggests that AI adoption is not limited to convenience-driven low-value transactions.

Consumers may find AI especially useful when the purchase involves:

• many technical specifications;
• multiple competing products;
• significant financial commitment;
• or extensive comparison.

The more complex the decision, the greater the potential value of having an AI system summarise and organise the available information.

Behaviour implication: AI Search may become especially influential in high-consideration purchasing journeys.

This is strategically important for sectors including:

• technology;
• automotive;
• financial services;
• travel;
• property;
• and high-value ecommerce.

Source:

Adobe Digital Insights — Generative AI Shopping Research

66. 34% of consumers use AI assistants for product research before searching online for the best deal

US Consumer Benchmark. Adobe found that 34% of consumers reported using AI assistants for product research before searching online for the best deal.

This reveals an important sequence in the modern purchase journey.

AI does not necessarily complete the whole transaction.

Instead, it can perform the research stage first.

The consumer may then use a traditional Search engine, retailer or marketplace to find:

• price;
• availability;
• discounts;
• delivery options;
• or the final seller.

The behavioural sequence becomes:

AI Product Research
↓
Product Decision
↓
Traditional Search for Best Deal
↓
Transaction

Behaviour implication: The channel receiving the purchase may not be the channel that shaped the decision.

This makes conventional last-click attribution increasingly incomplete.

Source:

Adobe — Generative AI Referral and Shopping Behaviour

67. Around one quarter of consumers now cite AI-powered platforms as a top source for research, purchase decisions and recommendations

Global Consumer Benchmark. Adobe and Oxford Economics found that approximately 25% of surveyed consumers cited AI-powered platforms such as ChatGPT as one of their top sources when searching for information, making purchase decisions or finding recommendations.

The global survey included 4,000 consumers across multiple regions, including the United Kingdom.

AI-powered platforms were already reported as a more common top source than:

• brand websites;
• and online reviews and ratings.

Traditional Search engines still remained the most widely used source overall.

Behaviour implication: AI platforms are becoming a substantial new layer of commercial discovery without eliminating traditional Search.

The consumer information ecosystem increasingly contains:

Search Engines + AI Platforms + Publishers + Reviews + Brand Websites

with users moving between them as required.

Source:

Adobe & Oxford Economics — AI and Digital Trends 2026: Customer Behaviours and AI

68. 42% of consumers who already use AI-powered platforms frequently or always rely on them as a primary source

Global Consumer Benchmark. Among consumers already using AI-powered platforms for research, Adobe found that 42% said they always or frequently rely on AI-powered assistants, conversational platforms or AI Search as their primary source for advice, shopping or troubleshooting.

This is different from merely experimenting with the technology.

For these consumers, AI is becoming one of the first places they turn when they have:

• a question;
• a shopping requirement;
• a problem;
• or a decision to make.

Behaviour implication: The strongest AI adopters are moving from:

AI as an Additional Tool

toward:

AI as a Primary Research Environment.

This makes visibility within generated answers increasingly comparable to visibility within traditional Search results for those user segments.

Source:

Adobe — AI and Digital Trends 2026 Consumer Report

69. 49% of consumers say they would use AI to find personalised product recommendations

Global Consumer Benchmark. Adobe’s 2026 customer research found that 49% of consumers would use AI to search for personalised product recommendations.

Personalisation is important because it moves recommendation behaviour beyond generic category rankings.

The user increasingly expects AI to understand:

• their requirements;
• preferences;
• budget;
• previous behaviour;
• and intended use.

A recommendation may therefore differ significantly between two users asking about the same broad category.

Behaviour implication: Commercial Search is becoming more contextual and individualised.

The future discovery question is less likely to be simply:

“What is the best product?”

and increasingly:

“What is the best product for me, given my particular situation?”

Source:

Adobe — AI and Digital Trends 2026

70. 44% of consumers say they would use AI for instant customer service

Global Consumer Benchmark. Adobe found that 44% of surveyed consumers would use AI to access instant customer service.

This extends AI participation beyond pre-purchase research.

The AI-assisted customer journey can increasingly include:

• discovery;
• product research;
• recommendations;
• purchase questions;
• customer service;
• and post-purchase support.

This creates a much wider role for conversational systems.

The same consumer who asks an AI assistant what product to buy may later use an AI-enabled interaction to:

• check delivery;
• understand a feature;
• solve a problem;
• or obtain support.

Behaviour implication: AI is developing from a Search interface into a broader decision and service interface.

That means Search Behaviour increasingly overlaps with:

Discovery + Commerce + Customer Experience

rather than stopping once the user clicks through to a website.

Source:

Adobe & Oxford Economics — Customer Behaviours and AI 2026

Section 8 — Demographics, Device Use & Multimodal Search Behaviour Statistics

The next phase of AI Search is not only conversational.

It is increasingly multimodal.

Consumers can now search by:

• typing;
• speaking;
• uploading an image;
• taking a photograph;
• sharing a camera view;
• or circling something already visible on a device.

This removes one of the longstanding limitations of traditional Search: the user no longer always needs to know the words required to describe what they want.

Demographic differences remain important as well, with younger adults adopting AI considerably faster than older groups.

Data classification: Statistics 71–75 use Ofcom’s 2026 Adults’ Media Use and Attitudes research, based on 7,533 UK adults aged 16+. Statistics 76–80 use Google platform data covering AI Mode, Google Lens and Circle to Search and are identified as global or platform-level behavioural evidence rather than UK population statistics.

71. 54% of UK adults now use AI tools such as ChatGPT, Copilot or Gemini

UK Adult Data. Ofcom reported in April 2026 that 54% of UK adults now use AI tools such as ChatGPT, Microsoft Copilot or Gemini.

This represents a substantial expansion in AI adoption.

Ofcom’s earlier 2025 research had recorded considerably lower general AI usage, indicating that adoption continued to accelerate during 2025.

It is important to distinguish this measure from AI Search specifically.

The 54% figure covers AI-tool use more broadly, including:

• Search;
• work;
• education;
• creative tasks;
• and conversational uses.

Nevertheless, broader AI adoption reduces the behavioural barrier to AI-assisted Search.

Consumers already comfortable using an AI interface are more likely to consider the same type of interface when they need information.

Behaviour implication: Conversational interfaces are becoming familiar to a majority of UK adults, creating a larger population capable of moving between conventional Search and AI-assisted discovery.

Source:

Ofcom — Adults’ Media Use and Attitudes 2026

72. 79% of UK adults aged 16–24 use AI tools

UK Adult Data. Ofcom found that 79% of UK adults aged 16–24 use AI tools.

That means almost four in five people within this age group had adopted AI by late 2025.

Younger consumers are therefore operating within a significantly different information environment from many older adults.

For these users, AI can increasingly sit alongside:

• Google Search;
• YouTube;
• social platforms;
• messaging;
• and specialist websites

as another routine information interface.

Behaviour implication: AI-assisted discovery is particularly important for organisations whose customers, students, employees or users are concentrated within younger demographics.

For these audiences, AI should increasingly be treated as an established digital behaviour rather than an experimental technology.

Source:

Ofcom — UK AI Adoption by Age, 2026

73. 74% of UK adults aged 25–34 use AI tools

UK Adult Data. Ofcom found that AI adoption was also very high among the next age group, with 74% of UK adults aged 25–34 using AI tools.

The difference between 16–24-year-olds and 25–34-year-olds was therefore relatively small:

79% — Age 16–24
74% — Age 25–34

Together, these groups represent a large proportion of the population most likely to:

• adopt new digital interfaces;
• conduct extensive online research;
• shop digitally;
• and use multiple platforms within one decision journey.

Behaviour implication: High AI adoption is no longer confined to the youngest adult consumers.

It is now widespread across adults into their early thirties, broadening the commercial relevance of AI Search.

Source:

Ofcom — Adults’ Media Use and Attitudes 2026

74. 12% of UK AI users use AI tools for conversational purposes

UK Adult Data. Ofcom found that 12% of UK AI users said they use AI tools for conversational purposes.

This behaviour goes beyond conventional information retrieval.

Ofcom’s qualitative research identified examples of people using AI for:

• relationship advice;
• companionship while working from home;
• creative planning;
• and other continuing conversations.

This is significant for Search because conversational familiarity changes user expectations.

Once people become accustomed to talking with AI, they may increasingly expect Search to support:

• follow-up questions;
• memory of prior context;
• clarification;
• and personalised interaction.

Behaviour implication: Search is increasingly being influenced by interaction patterns developed outside conventional Search engines.

Users may expect the information experience to feel like a dialogue rather than a sequence of disconnected queries.

Source:

Ofcom — AI Companionship and Conversational Usage

75. 19% of UK AI users aged 25–34 use AI for conversational purposes

UK Adult Data. Conversational usage is particularly strong among younger adults.

Ofcom found that 19% of AI users aged 25–34 use AI tools for conversational purposes.

That compares with 12% of AI users overall.

The difference suggests that younger AI adopters may be more comfortable maintaining an extended interaction with a machine rather than treating the system as a simple question-and-answer tool.

This matters for Search because repeated conversation can lead to:

• more detailed context;
• greater personalisation;
• continued refinement;
• and more complex decision support.

Behaviour implication: The future of Search is likely to involve increasing overlap between:

Information Retrieval + Conversation + Personal Context

particularly among younger users.

Source:

Ofcom — UK Conversational AI Usage, 2026

76. Nearly one in six Google AI Mode queries now use voice or images rather than text

Global Platform Behaviour Data. Google reported in February 2026 that nearly one in six AI Mode queries were non-text queries using voice or images.

This is one of the clearest indicators that AI Search Behaviour is expanding beyond typing.

Users can increasingly begin a Search by:

• speaking naturally;
• showing the system an object;
• uploading an image;
• or combining visual information with a verbal question.

For example, the user no longer needs to know how to describe a particular object.

They can show it to the Search system and ask:

“What is this and where can I buy one?”

Behaviour implication: The future Search query is increasingly multimodal.

Search optimisation therefore needs to consider how organisations, products and places are represented through:

• text;
• images;
• product feeds;
• structured information;
• and other machine-readable evidence.

Source:

Google — Alphabet Q4 2025 Search and AI Usage Update

77. Google Lens processes more than 25 billion visual searches per month

Global Visual Search Data. Google reported in 2025 that Google Lens was handling more than 25 billion visual Search queries every month.

This had increased from approximately 20 billion monthly queries in October 2024.

Visual Search removes the requirement for a user to convert what they see into words.

Instead, the behaviour becomes:

See Something
↓
Photograph or Select It
↓
Search

That can apply to:

• products;
• plants;
• buildings;
• clothing;
• food;
• text;
• landmarks;
• and objects whose names the consumer does not know.

Behaviour implication: Search demand is no longer limited to questions users are capable of expressing in words.

Visual interfaces can unlock searches that previously might never have occurred.

Source:

Google — Visual Search and AI Behaviour

78. One in four Google Lens searches carries commercial intent

Global Visual Search Data. Google reported that one in four visual Search queries performed through Google Lens had commercial intent.

This makes visual Search particularly important for retail and product discovery.

A consumer might see:

• a pair of shoes;
• a chair;
• a watch;
• a kitchen appliance;
• or an item of clothing

without knowing the product name or brand.

They can photograph or select the object and move directly into:

• product identification;
• price comparison;
• reviews;
• retailer discovery;
• and purchase research.

Behaviour implication: Commercial Search can now begin with an object rather than a keyword.

This makes product imagery and accurate product data increasingly important components of Search visibility.

Source:

Google — Visual Search Consumer Behaviour

79. More than 1.5 billion people use Google Lens each month

Global Platform Data. Google reported in June 2025 that more than 1.5 billion people use Google Lens each month to Search what they see.

Google also reported that Lens had grown 65% year over year and had already generated more than 100 billion visual Searches during the year at the time of publication.

The scale demonstrates that visual Search is no longer a specialist Search behaviour.

It is becoming a major route into information and product discovery.

Visual Search is particularly useful when:

• the user does not know the name of an object;
• language is difficult;
• the visual characteristics matter;
• or describing the requirement would take longer than showing it.

Behaviour implication: Search increasingly begins with recognition rather than description.

This changes the traditional assumption that every Search Journey starts with text.

Source:

Think with Google — Google Lens Visual Search Trends

80. Younger users who adopt Circle to Search use it to start around 10% of their searches

Global Younger-User Behaviour Data. Google reported that younger users who try Circle to Search subsequently use it to begin around 10% of their searches.

Circle to Search allows Android users to:

• circle;
• highlight;
• tap;
• or select

something already visible on the device screen and immediately Search for it.

Google has repeatedly reported particularly strong engagement with Circle to Search among people aged 18–24.

The company also reported that Circle to Search usage increased by nearly 40% in one quarter during 2025.

By early 2026, the feature was available across more than 580 million Android devices.

The behavioural significance is that Search no longer needs to interrupt the activity the user is already performing.

A consumer can encounter something in:

• social media;
• a video;
• a website;
• a message;
• or another application

and immediately Search it without leaving the current screen.

Behaviour implication: Search is becoming increasingly embedded within other digital experiences.

Instead of consciously deciding to “go to a Search engine,” the consumer can Search at the exact moment curiosity occurs.

Sources:

Think with Google — The Future of AI-Powered Search
;

Google — Q4 2025 Search Update

Section 9 — Sector-Specific AI Search Behaviour Statistics

AI Search adoption does not develop at the same rate across every industry.

The value of conversational Search depends partly on the type of decision being made.

AI can be particularly useful when consumers need to:

• combine large amounts of information;
• compare several alternatives;
• understand technical terminology;
• personalise a recommendation;
• or reduce the complexity of a high-consideration decision.

This creates different Search Behaviour patterns across sectors including Travel, Financial Services, Healthcare and Technology.

Data classification: Statistics 81–82 use UK travel research from YouGov. Statistics 83–84 use Adobe US travel research. Statistics 85–86 use Adobe US financial-services consumer research. Statistic 87 uses EY global financial-services consumer research. Statistics 88–89 use UK healthcare evidence from Ipsos and Deloitte. Statistic 90 uses Adobe consumer technology and software purchase research.

81. 20% of UK travellers were comfortable using AI for trip planning in 2025

UK Travel Data. YouGov found that 20% of UK travellers said they were comfortable using AI-driven platforms for trip planning in 2025.

That represented an increase from 15% in 2024.

Travel is particularly suited to conversational research because even a relatively simple holiday decision can involve:

• destination;
• dates;
• budget;
• accommodation;
• transport;
• restaurants;
• activities;
• and personal preferences.

AI can combine these factors inside one conversation.

Behaviour implication: UK travellers are becoming progressively more comfortable allowing AI to participate in a complex planning journey rather than using it only for isolated factual questions.

Source:

YouGov — Brits More Comfortable Using AI for Trip Planning

82. 35% of UK travellers aged 25–34 were comfortable using AI for trip planning

UK Travel Data. YouGov found that comfort with AI travel planning was substantially higher among younger travellers.

Among UK travellers aged 25–34, 35% said they were comfortable using AI to help organise a trip.

That was up from 25% in 2024.

At the other end of the age range, only 10% of travellers aged 55 and over reported being comfortable with AI-assisted trip planning.

The difference demonstrates again that sector-level AI adoption is strongly influenced by demographic behaviour.

Behaviour implication: Travel brands targeting younger adults are likely to encounter materially greater AI-assisted discovery than businesses whose customers are concentrated among older travellers.

Source:

YouGov — UK AI Travel Planning Research

83. 29% of US consumers had used generative AI to plan trips

US Travel Consumer Benchmark. Adobe Digital Insights found that 29% of surveyed US consumers had used generative AI services for trip planning.

Travel planning is a strong example of AI Search moving beyond one-question information retrieval.

Users can ask one system to help with:

• where to travel;
• when to go;
• where to stay;
• how to travel locally;
• what to eat;
• what to see;
• and how much the overall trip may cost.

The Search session can therefore operate more like a personalised travel-research process.

Behaviour implication: Travel demonstrates how generative AI can compress what historically required many separate Search queries into one continuing research conversation.

Source:

Adobe Digital Insights — Consumers Embrace Generative AI for Trip Planning

84. 84% of consumers using AI for travel planning said it improved their experience

US Travel Consumer Benchmark. Adobe’s 2026 travel research found that 84% of respondents using AI for travel planning said it improved their travel-planning experience.

Adobe found that the most common travel uses included:

• research — 43%;
• travel inspiration — 43%;
• transportation planning — 41%;
• itinerary creation — 34%.

The research also found very high confidence in AI-generated travel information.

Among respondents using AI tools for travel, 95% said they considered AI responses as trustworthy as traditional Search.

Behaviour implication: Once consumers adopt AI for Travel Search, satisfaction can be high enough for the tool to become part of the normal planning process rather than merely an experimental alternative.

Source:

Adobe Digital Insights — How AI Is Reshaping Travel Planning

85. 27% of US consumers were using generative AI for banking and financial needs

US Financial Services Benchmark. Adobe found that 27% of surveyed consumers were already using generative AI for banking and financial needs.

Adoption was higher among younger groups:

• 34% among Gen Z;
• 35% among Millennials.

Financial Search can involve complex terminology and large numbers of competing products.

AI can reduce that complexity by helping users:

• understand financial concepts;
• compare accounts;
• build budgets;
• research investments;
• and examine tax implications.

Behaviour implication: Financial Search is beginning to move from simple provider discovery toward AI-assisted interpretation and decision support.

Source:

Adobe Digital Insights — Generative AI and Financial Services

86. 42% of consumers using AI for financial needs use it for checking or savings account recommendations

US Financial Services Benchmark. Among consumers already using generative AI for financial needs, Adobe found that 42% used it to obtain recommendations for checking or savings accounts.

Other financial uses included:

• 40% — explanations of complex financial topics;
• 39% — personalised budgeting;
• 37% — investment advice;
• 35% — tax advice.

These are not simply informational queries.

Several involve helping the user select or evaluate a financial product.

Behaviour implication: AI Search can increasingly sit between:

Consumer Need

and:

Financial Provider Selection.

For banks and financial-services organisations, being absent from AI-generated consideration sets may therefore matter even when conventional organic rankings remain strong.

Source:

Adobe — AI Banking and Financial Research Behaviour

87. 49% of global consumers have used AI to support savings or investment decisions

Global Financial Services Benchmark. EY reported in 2026 that 49% of surveyed consumers had used AI to support savings and investment decisions during the previous six months.

The research also found:

• 21% had used AI agents for financial product recommendations;
• 18% had used AI for budgeting, household finance management or trading support;
• 14% had allowed AI to select a financial-services provider on their behalf;
• 11% had allowed AI to manage finances with little or no human intervention.

The final two findings are especially important.

They indicate the beginning of a transition from:

AI Giving Information

toward:

AI Participating in the Decision.

Behaviour implication: Financial Search may eventually move beyond recommendation into agent-assisted provider selection and transaction execution.

That would fundamentally alter how consumers discover and choose financial products.

Source:

EY — Nearly Half of Global Consumers Now Use AI to Guide Savings and Investment Decisions

88. 28% of Britons have used an AI chatbot for information about their own health

UK Healthcare Data. Ipsos reported in January 2026 that 28% of Britons had used AI chatbots to obtain information about their own health.

For comparison, 66% reported using conventional online health websites such as NHS.uk to investigate symptoms.

This again demonstrates the hybrid nature of modern Search Behaviour.

Consumers may move between:

• AI chatbots;
• Search engines;
• NHS resources;
• health websites;
• and professional care.

Behaviour implication: AI is becoming an additional health-information entry point rather than eliminating conventional health research.

For healthcare organisations, consistency between AI-accessible information and authoritative clinical sources becomes increasingly important.

Source:

Ipsos UK — Britons Turn to Online Sources and AI to Check Symptoms

89. 26% of UK generative AI users have asked AI for health advice

UK Healthcare Data. Deloitte’s 2026 Digital Consumer Trends research found that 26% of UK generative AI users had used the technology for health advice.

Among those users, 38% believed the health information produced by generative AI was always factually accurate.

Deloitte also found:

• 20% of generative AI users had sought diet or fitness advice;
• 10% had used generative AI as they might use a therapist.

The findings show that users are increasingly willing to ask AI about subjects involving personal context and potentially consequential decisions.

Behaviour implication: Healthcare AI Search is moving beyond basic factual lookup into personalised advice, interpretation and wellbeing support.

That makes evidence quality and clear professional boundaries especially important within this sector.

Source:

Deloitte UK — Digital Consumer Trends 2026

90. 28% of consumers reported making a technology or software purchase with the help of AI

US Technology Consumer Benchmark. Adobe’s August 2025 consumer survey found that 28% of consumers had made a technology or software purchase with the help of AI.

Among those AI-assisted technology purchases:

• 37% involved electronics;
• 14% involved IT or cloud software;
• 9% involved cybersecurity software;
• 6% involved analytics or infrastructure tools.

Technology purchasing is particularly compatible with AI-assisted research because users often need to compare:

• specifications;
• features;
• compatibility;
• performance;
• price;
• and suitability for a particular use case.

Adobe also found that technology and software had the highest AI-driven visit share among the major industries it measured during the 2025 holiday season.

Behaviour implication: Complex technical products are particularly well suited to AI-assisted Search because the AI system can translate specifications into a personalised recommendation.

Instead of asking:

“Best laptop”

the consumer can ask:

“Which laptop is best for video editing, occasional gaming and travel, with at least 16GB RAM and a budget of £1,500?”

The Search becomes a specification-matching exercise rather than a generic category query.

Source:

Adobe Digital Insights — AI-Sourced Traffic Insights

Section 10 — Cross-Platform, AI-Integrated & Future Search Behaviour Statistics

The final ten statistics show why AI Search Behaviour should not be understood as a simple competition between Google and ChatGPT.

Search is increasingly becoming a capability distributed across:

• traditional Search engines;
• AI assistants;
• mobile operating systems;
• news environments;
• voice interfaces;
• cameras;
• screens;
• files;
• and other applications.

The user may no longer consciously decide to “go and Search”.

Search can increasingly occur inside whatever digital environment the user is already using.

Data classification: Statistics 91–92 use current Ofcom UK evidence published in September 2026. Statistics 93–99 use Google platform-level behavioural and usage data published during 2026. Statistic 100 uses Adobe and Oxford Economics’ international 2026 consumer research.

91. One in five UK adults now use AI tools to keep up with the news

UK Adult Data. Ofcom’s News Consumption Report 2026 found that one in five UK adults now use AI tools such as ChatGPT to keep up to date with the latest news.

This is a significant behavioural development because news discovery has historically depended heavily on:

• news websites;
• television;
• radio;
• Search engines;
• social media;
• and news aggregators.

Generative AI now represents another information-discovery layer within that ecosystem.

A consumer can ask:

• “What happened today?”
• “What are the main developments?”
• “Explain this story.”
• “What has changed since yesterday?”
• or “Summarise the arguments.”

rather than visiting an individual publisher first.

Behaviour implication: AI Search is extending into highly current information behaviour, where source freshness, authority and attribution become especially important.

Source:

Ofcom — One in Five Brits Using AI to Keep Up With News

92. UK adults now encounter around 12 different news sources each month, rising to 18 among 25–34-year-olds

UK Cross-Platform Behaviour Data. Ofcom found that UK adults use or encounter approximately 12 different news sources each month.

Among adults aged 25–34, that rises to approximately 18 sources.

This is not an AI-only measurement, but it demonstrates the wider digital behaviour into which AI Search is entering.

Consumers increasingly operate across multiple information environments rather than relying permanently on one source.

Those environments can include:

• Search;
• AI;
• social platforms;
• publishers;
• video;
• apps;
• and direct brand sources.

Behaviour implication: Modern information discovery is increasingly multi-source and cross-platform.

Organisations should therefore avoid interpreting one Search channel as though it represents the complete customer information journey.

Source:

Ofcom — News Consumption Report 2026

93. Google AI Overviews now reaches more than 2.5 billion monthly users

Global Platform Data. Google reported in 2026 that AI Overviews had surpassed 2.5 billion monthly active users.

This is important because AI Search Behaviour is no longer confined to people who deliberately open a standalone AI chatbot.

A consumer can encounter an AI-generated answer while conducting what they regard as an ordinary Google Search.

The behavioural transition can therefore occur without an explicit decision to adopt a new Search platform.

The user can:

Search Google
↓
Receive an AI-Generated Answer
↓
Continue Research
↓
Open Sources or Ask More Questions

Behaviour implication: AI Search adoption should not be measured solely through standalone chatbot usage.

Generative answers are increasingly embedded inside mainstream Search itself.

Source:

Google — New Opportunities, Control and Insights for Website Owners

94. The Gemini app surpassed one billion monthly users in August 2026

Global Platform Data. Google reported in August 2026 that the Gemini app had surpassed one billion monthly active users.

This demonstrates the scale at which standalone conversational AI behaviour is developing alongside AI-enhanced traditional Search.

The significance is not simply the number of users.

AI assistants increasingly support:

• research;
• conversation;
• voice;
• images;
• screen sharing;
• files;
• and actions across other applications.

Behaviour implication: Search is increasingly merging with the broader AI-assistant category.

Consumers may no longer distinguish clearly between:

Searching

and:

Asking an Assistant.

Source:

Google — More Than One Billion People Are Using the Gemini App Every Month

95. 63% of Gemini users now speak directly to the AI

Global Platform Behaviour Data. Google reported that 63% of Gemini users use voice to talk directly to the AI.

This shows how quickly interaction behaviour can move beyond the keyboard.

Voice allows consumers to ask questions:

• while walking;
• while driving where permitted;
• while cooking;
• while looking at something;
• or when typing would be inconvenient.

Spoken language is also naturally more conversational than traditional typed keywords.

A user is more likely to say:

“I need somewhere quiet to stay near central London next weekend that isn’t too expensive and has parking.”

than reproduce that full sentence as an old-style keyword query.

Behaviour implication: As voice adoption increases, Search language may become even more contextual, natural and conversational.

Source:

Google — Gemini Usage Behaviour 2026

96. One in five Gemini Live interactions now goes beyond voice

Global Multimodal Behaviour Data. Google reported that one in five Gemini Live interactions now goes beyond voice.

Users increasingly incorporate:

• live camera feeds;
• or screen sharing

into their AI conversations.

This changes the information request fundamentally.

The user no longer always needs to explain the problem verbally.

They can simply show the AI:

• the object;
• the screen;
• the document;
• the problem;
• or the physical environment.

and ask a question about it.

Behaviour implication: Search increasingly shifts from:

Describe the Problem

toward:

Show the Problem.

This significantly expands the range of situations in which consumers can initiate information discovery.

Source:

Google — Gemini Multimodal Usage Data

97. 38% of school-related Gemini requests include an attachment

Global Multimodal Platform Data. Google reported that 38% of school-related requests made through Gemini include an attachment.

This provides another example of Search moving beyond the text query.

The user can provide:

• a document;
• an image;
• a worksheet;
• notes;
• or another file

and ask the AI to reason about the material.

The question therefore contains two elements:

User Prompt + Supplied Context

rather than the prompt alone.

Behaviour implication: Future Search Intent may increasingly be impossible to understand from keyword text alone because the critical context can be contained within:

• an uploaded file;
• an image;
• a screen;
• or an ongoing conversation.

This is another reason conventional keyword datasets represent only part of emerging Search Behaviour.

Source:

Google — Gemini Multimodal Interaction Data

98. Gemini has more than 100 million active users on iOS alone

Global Device Behaviour Data. Google reported that the Gemini app had surpassed 100 million active users on Apple’s iOS platform.

This is significant because Gemini is a Google product operating at substantial scale within a competing mobile ecosystem.

It demonstrates that AI-assistant usage can cross traditional platform boundaries.

Consumers are increasingly assembling their own information ecosystem from:

• different devices;
• different Search engines;
• different operating systems;
• and different AI assistants.

Behaviour implication: AI Search loyalty should not automatically be inferred from the device or operating system a consumer uses.

Cross-platform AI behaviour is becoming normal.

A consumer may use:

iPhone + Google Search + Gemini + ChatGPT + Safari

within the same broader digital journey.

Source:

Google — Gemini App Usage, August 2026

99. Gemini power users on macOS prompt approximately twice as frequently as users on other surfaces

Global Device Behaviour Data. Google reported that Gemini power users on macOS submit prompts at around twice the frequency seen across other surfaces.

This indicates that AI behaviour is not uniform across devices or user groups.

Heavy users may incorporate AI into a much larger number of daily activities.

For these users, AI can become a persistent layer across:

• work;
• research;
• writing;
• analysis;
• Search;
• planning;
• and decision-making.

Behaviour implication: Measuring average AI adoption alone can conceal much heavier behaviour among advanced user segments.

The future Search landscape is likely to contain both:

Occasional AI Search Users

and:

AI-First Information Users.

The commercial importance of the latter group may be much greater than their population share alone suggests.

Source:

Google — Gemini Cross-Device Usage Behaviour

100. 70% of consumers say personalised AI recommendations should feel human rather than automated

Global Consumer Benchmark. Adobe and Oxford Economics’ 2026 research found that 70% of consumers said it was very or moderately important that personalised offers and recommendations feel human rather than automated or robotic.

This final statistic highlights an important tension in AI Search Behaviour.

Consumers increasingly value:

• speed;
• convenience;
• personalisation;
• and instant answers.

But they do not necessarily want the experience to feel mechanical.

The most effective AI interaction may therefore need to combine:

Machine Intelligence
+
Human Relevance

This matters for organisations because AI systems increasingly mediate the relationship between brand information and the customer.

Content designed purely for machines may fail to persuade people.

Content designed purely for people but poorly structured for machines may fail to enter AI-generated answers.

Behaviour implication: Organisations increasingly need information that is simultaneously:

• machine understandable;
• factually reliable;
• contextually relevant;
• and genuinely useful to humans.

Source:

Adobe & Oxford Economics — AI and Digital Trends 2026: Customer Behaviours and AI

Overall Analysis — What the 100 AI Search Behaviour Statistics Tell Us

The 100 statistics in this research show that AI Search is not simply creating another destination for information.

It is changing the behaviour of Search itself.

The evidence across UK consumer surveys, observed clickstream datasets, platform usage data and international behavioural studies points toward a Search environment that is becoming:

• more conversational;
• more contextual;
• more iterative;
• more multimodal;
• more personalised;
• and more distributed across platforms.

Traditional Search remains deeply important.

But consumer behaviour increasingly combines:

Google Search
+
AI Assistants
+
AI-Generated Search Results
+
Social & Publisher Sources
+
Brand Websites
+
Professional Sources

The result is a more complex discovery journey than the traditional model of:

Keyword → Ranking → Click.

1. AI Search Has Entered Mainstream UK Behaviour

The first major finding is that AI-assisted information seeking is no longer restricted to early adopters.

The UK evidence shows:

• 51% of UK adults use AI tools to search for products, services or advice;
• 24% use AI Search frequently;
• 54% of UK adults use AI tools more broadly;
• 79% of 16–24-year-olds use AI;
• 74% of 25–34-year-olds use AI.

AI-assisted Search therefore already represents a significant part of the UK information ecosystem.

However, adoption remains highly uneven.

Younger consumers are significantly more likely to:

• use AI;
• use it frequently;
• continue conversations;
• use it for professional or high-consideration decisions;
• and incorporate AI into shopping, travel and health research.

This means businesses should not think of AI Search adoption as one national average.

The relevant question is:

“How much AI-assisted Search Behaviour exists within our own audience?”

2. AI Search Is Expanding Search Rather Than Simply Replacing Google

The evidence does not support a simple narrative in which consumers stop using traditional Search and replace it with ChatGPT.

Traditional Search remains deeply embedded in consumer behaviour.

Which? found that 82% of UK adults still use traditional Search engines frequently.

At the same time, AI use is rising rapidly.

This creates a hybrid behaviour:

AI Search
↕
Traditional Search
↕
Websites & Third-Party Sources

The Similarweb evidence is particularly important.

Users receiving a ChatGPT brand recommendation were 2.5 times more likely to visit the recommended brand within seven days.

Yet 55.9% of those AI-influenced visits subsequently arrived through Search.

Semrush separately found Google receiving 21.6% of ChatGPT outbound referral traffic.

AI can therefore create demand that later appears inside conventional Search analytics.

The Search platform receiving the final visit may not be the platform that originally shaped the decision.

3. The Search Query Is Becoming a Conversation

The second major behavioural change concerns the query itself.

Traditional Search encouraged users to reduce information needs into short keyword phrases.

AI interfaces allow users to express the full situation.

Semrush found Google AI Mode queries averaged:

7.22 Words
vs
4.0 Words for Traditional Google Search

Its ChatGPT research also found that for much of the study period, 65–85% of prompts did not resemble conventional Search keywords.

The behavioural shift is:

Keyword
↓
Question
↓
Context-Rich Prompt
↓
Conversation

This changes what Search Intent means.

The user’s intent may now include:

• their situation;
• budget;
• location;
• preferences;
• constraints;
• previous experience;
• and desired outcome.

Keyword research therefore remains useful, but no longer captures the full Search Behaviour landscape.

4. Search Is Becoming Multi-Step Rather Than Single-Query

The third major shift is the growth of follow-up behaviour.

Semrush found average ChatGPT queries per session increasing to 1.75 by February 2026, representing a 50% increase during the final four months of its study.

Which? found:

• 55% of UK AI Search users ask follow-up questions to improve results;
• 73% of frequent users do so.

This means the first answer is increasingly only the beginning.

The user can continue:

Ask
↓
Clarify
↓
Refine
↓
Compare
↓
Challenge
↓
Recommend
↓
Decide

This has major implications for AI Search visibility measurement.

An organisation can be absent from the first answer but appear later when the user:

• adds a budget;
• changes location;
• asks for an alternative;
• requests a comparison;
• or narrows the shortlist.

Measuring one prompt therefore does not necessarily measure the complete AI Search Journey.

5. Consumers Trust AI — But They Also Verify It

Trust behaviour is more sophisticated than simple acceptance or rejection.

Which? found:

• 47% of AI Search users trust results to a reasonable or great extent;
• 35% trust them to some extent;
• 15% report little or no trust.

Trust rises strongly with frequent usage.

Yet consumers continue to verify.

Almost one quarter of AI Search users often or always compare results between different AI tools.

Among frequent users, this rises to 36%.

Only 30% of users believe AI Search tools are independent or unbiased to a reasonable or great extent.

The emerging trust model is therefore:

Use
↓
Trust Provisionally
↓
Verify
↓
Cross-Check
↓
Decide

For organisations, consistent digital evidence across multiple sources becomes increasingly important.

6. AI Is Moving Into High-Consideration Decisions

AI Search is not limited to simple informational queries.

The research shows consumers using AI for:

• healthcare information;
• financial advice;
• legal questions;
• travel planning;
• technology purchases;
• and complex product research.

Among UK AI Search users:

• 19% often or always use AI instead of healthcare advice;
• 17% instead of financial advice;
• 13% instead of legal advice.

The figures are considerably higher among younger adults.

International evidence also shows AI increasingly influencing:

• travel itineraries;
• investment decisions;
• financial product selection;
• and complex technology purchases.

AI therefore increasingly participates before the consumer contacts the organisation or professional.

7. Product Discovery Is Moving Upstream

One of the strongest commercial findings is that AI can shape the buying journey before the consumer visits a retailer or brand website.

Adobe’s research found:

• 35% of UK consumers had already used generative AI for shopping;
• 47% planned to use it during 2025.

Among AI-shopping users in international research:

• 72% used AI as a primary product and brand research tool;
• 47% used it for recommendations;
• 43% for deals;
• 35% for gift ideas;
• 33% for shopping lists.

This means AI can influence:

What Category?
↓
Which Brands?
↓
Which Products?
↓
Which Features Matter?
↓
What Should I Buy?

before conventional commercial Search becomes visible.

8. Recommendation Visibility Can Influence Real-World Commercial Behaviour

AI recommendation visibility is not merely an abstract branding metric.

The behavioural evidence shows it can influence later action.

Adobe found that among AI-assisted shoppers:

• more than 55% actively click AI-provided links;
• 65% say AI increases purchase confidence;
• 68% say they are less likely to return the product;
• 87% are more likely to use AI for large or complex purchases.

Similarweb’s downstream research provides another strong signal by showing that ChatGPT recommendations significantly increase later brand visits.

The commercial path can therefore become:

AI Visibility
↓
Brand Mention
↓
Recommendation
↓
Consideration
↓
Traditional Search or Direct Visit
↓
Conversion

This creates attribution challenges because the channel receiving the final click may receive credit for demand partly created by AI.

9. Search Is Becoming Multimodal

Search Behaviour is also moving beyond text.

Google reports:

• nearly one in six AI Mode queries uses voice or images;
• Google Lens handles more than 25 billion visual Searches per month;
• one in four Lens searches carries commercial intent;
• more than 1.5 billion people use Lens each month;
• 63% of Gemini users use voice;
• one in five Gemini Live interactions uses camera or screen information.

The Search query can therefore begin with:

Words
or
Voice
or
Image
or
Object
or
Screen
or
File

The traditional keyword is no longer always required.

This expands the Search opportunity beyond written content into:

• images;
• product feeds;
• structured data;
• entities;
• locations;
• and other machine-readable information.

10. Search Is Becoming Embedded Rather Than Visited

Perhaps the most important long-term change is that consumers increasingly do not need to visit a dedicated Search environment at all.

Search capabilities are becoming embedded into:

• AI assistants;
• mobile devices;
• operating systems;
• news consumption;
• productivity software;
• camera interfaces;
• and other applications.

This creates a shift from:

“I need to Search.”

toward:

“I need an answer.”

The technology determines how that answer is found.

100 AI Search Behaviour Statistics at a Glance

The table below summarises the 100 statistics examined in this research.

Each statistic is classified according to the geography or type of evidence behind it. UK-specific figures are distinguished from international, US and global behavioural benchmarks.

CGO AI Search Behaviour Framework™

The 100 statistics show that AI Search Behaviour is no longer adequately described by a simple Search model based on:

Query → Ranking → Click

AI-assisted discovery increasingly involves a much longer sequence of behaviours.

The CGO AI Search Behaviour Framework provides a structured model for understanding how consumers move from an initial need through AI-assisted research, verification and eventual decision-making.

The framework contains seven behavioural stages:

1. Need Formation
2. Question & Prompt Formation
3. AI Discovery
4. Conversational Research
5. Verification & Cross-Checking
6. Recommendation & Consideration
7. Decision & Action

1. Need Formation

The Search Journey begins before a query is entered.

The user first develops a need, problem, objective or uncertainty.

Examples include:

• choosing a holiday;
• finding a financial product;
• understanding a health concern;
• selecting software;
• researching a professional service;
• or identifying the right product for a specific use case.

Traditional Search strategies often begin only when the user converts that need into a keyword.

AI Search makes the earlier stage more visible because consumers can describe the underlying problem directly.

For example:

“I need somewhere warm in Europe for a quiet five-day break in October that isn’t too expensive.”

The consumer has not yet selected:

• a destination;
• a hotel;
• an airline;
• or even a precise category of solution.

The AI system can participate in shaping the requirement itself.

Measurement question: What customer needs, problems and situations are relevant to the organisation before a brand or product is mentioned?

2. Question & Prompt Formation

The second stage is how the user expresses the need.

Traditional Search normally compresses intent into a short phrase.

AI Search allows considerably more context.

A prompt may contain:

• location;
• budget;
• preferences;
• constraints;
• use case;
• experience level;
• and desired outcome.

This changes Search Behaviour from:

“best CRM software”

toward:

“What CRM would suit a 20-person professional-services company that needs simple reporting, email integration and does not want an enterprise-level system?”

The second query contains significantly more commercial context.

Measurement question: What contextual prompts describe the customer’s actual situation rather than merely the keyword category?

3. AI Discovery

The third stage occurs when the AI system constructs the initial answer.

The system may:

• identify entities;
• retrieve current information;
• select sources;
• summarise options;
• mention brands;
• or recommend possible solutions.

At this stage, an organisation may be:

• mentioned;
• cited;
• recommended;
• included in a shortlist;
• or omitted entirely.

The distinction matters.

A company may rank strongly in traditional Search but still fail to appear in an AI-generated answer.

Conversely, a company may be surfaced by an AI system because the wider evidence ecosystem strongly supports its relevance.

Measurement question: Does the organisation appear in the initial AI-generated answer for relevant customer needs?

4. Conversational Research

The fourth stage is where AI Search differs most clearly from conventional Search.

The user can continue the research without beginning again.

Follow-up questions can include:

• “Which is cheapest?”
• “Which is best for a small company?”
• “Compare the first three.”
• “What are the disadvantages?”
• “What would you recommend for my budget?”
• “Are there any UK alternatives?”

Each follow-up can change the consideration set.

A company absent from the first answer may enter later.

A company initially recommended may disappear once the user introduces another requirement.

This means AI visibility should be measured across a prompt sequence, not only one isolated question.

Measurement question: Does the organisation remain visible as the consumer refines the original request?

5. Verification & Cross-Checking

The fifth stage occurs when the user tests the generated answer.

Verification can take several forms:

• asking the AI for sources;
• asking a follow-up question;
• checking another AI platform;
• searching Google;
• reading reviews;
• visiting the organisation’s website;
• or consulting an authoritative third party.

This stage is especially important for:

• healthcare;
• financial services;
• legal services;
• high-value purchases;
• and other consequential decisions.

The user is effectively testing whether the evidence behind the AI response is consistent.

An organisation with contradictory information across the web may lose confidence at this point.

Measurement question: Does the organisation’s information remain consistent across AI systems, Search engines, first-party content and credible third-party sources?

6. Recommendation & Consideration

The sixth stage is where discovery becomes commercially significant.

The user begins to narrow the available options.

An AI system may be asked to:

• rank alternatives;
• compare products;
• recommend providers;
• identify the best option for a particular use case;
• or explain which choice most closely matches the user’s requirements.

At this stage, visibility alone is not enough.

The organisation needs to be sufficiently relevant and credible to remain inside the final consideration set.

The distinction is:

Being Mentioned
≠
Being Recommended

Recommendation visibility represents a stronger form of influence than simple citation or brand presence.

Measurement question: When the user asks the AI to recommend, compare or shortlist, does the organisation remain one of the options?

7. Decision & Action

The final stage is the consumer action that follows the research process.

That action may include:

• visiting a website;
• searching the brand name;
• making a purchase;
• contacting the organisation;
• requesting a quotation;
• booking an appointment;
• downloading information;
• or continuing research elsewhere.

Importantly, the action may not generate an obvious AI referral.

A consumer can discover a company through ChatGPT and later:

• Search Google for the brand;
• visit directly;
• use a marketplace;
• or make an offline enquiry.

This means AI’s influence can be hidden within conventional attribution systems.

Measurement question: What downstream Search, direct, conversion and branded-demand signals increase when AI visibility improves?

How to Measure AI Search Behaviour

Traditional Search measurement is built around rankings, impressions, clicks and conversions.

AI Search requires a broader measurement model because much of the customer journey can happen before a website visit occurs.

An organisation may influence a customer when it is:

• mentioned in an AI answer;
• cited as a source;
• included in a comparison;
• recommended;
• retained after follow-up questions;
• or discovered again through branded Search.

The objective is therefore to measure the complete sequence:

Prompt
↓
Visibility
↓
Conversation
↓
Verification
↓
Recommendation
↓
Downstream Action

1. Prompt Coverage Rate

Prompt Coverage measures how much of the relevant customer-question universe an organisation is monitoring.

A strong prompt set should include:

• informational questions;
• problem-based questions;
• comparison queries;
• recommendation queries;
• location-specific queries;
• use-case queries;
• and follow-up questions.

The metric can be calculated as:

A low Prompt Coverage Rate means the organisation may be drawing conclusions from too narrow a sample.

2. AI Visibility Rate

AI Visibility Rate measures how often the organisation appears in relevant AI-generated answers.

An appearance may include:

• brand mention;
• product mention;
• source citation;
• provider listing;
• or recommendation.

This should be measured separately across platforms because visibility can vary substantially between:

• ChatGPT;
• Gemini;
• Google AI Overviews or AI Mode;
• Copilot;
• Perplexity;
• and other relevant AI systems.

3. Follow-Up Retention Rate

An organisation that appears in the initial response may disappear after the user adds more context.

Follow-Up Retention Rate measures whether the organisation remains visible as the conversation becomes more specific.

For example:

“Best accounting software”

may become:

“Which of those is best for a UK business with fewer than 20 employees?”

and then:

“Which has the easiest reporting and lowest monthly cost?”

This is one of the clearest ways to distinguish superficial visibility from durable relevance.

4. Recommendation Visibility Rate

Being mentioned is weaker than being recommended.

Recommendation Visibility should therefore be measured separately from general AI visibility.

Recommendation prompts can include:

• “Which company would you recommend?”
• “What are the best options?”
• “Which provider is most suitable?”
• “Which product should I choose?”
• “What is the best option for my circumstances?”

This metric is particularly relevant where AI is influencing provider, product or destination selection.

5. AI Citation Rate

Citation Rate measures whether the organisation’s content is being selected as supporting evidence.

Citation visibility should be tracked separately from Brand Mentions because an organisation can be:

• mentioned but not cited;
• cited but not explicitly named;
• or both mentioned and cited.

All three patterns provide different information about how the organisation is being used by AI systems.

6. Cross-Platform Consistency Rate

Consumers increasingly move between several AI and Search environments.

Cross-Platform Consistency measures whether major platforms provide broadly consistent information about the organisation.

The areas to compare can include:

• brand description;
• services;
• products;
• locations;
• pricing;
• expertise;
• leadership;
• and recommendations.

Large inconsistencies may indicate:

• weak entity clarity;
• conflicting third-party information;
• outdated source data;
• or insufficient authoritative evidence.

7. AI-to-Branded-Search Lift

AI recommendations may generate later branded Search rather than a direct referral.

One useful measurement approach is therefore to track changes in branded Search activity alongside improvements in AI visibility.

This does not prove that AI caused the increase.

However, when interpreted alongside AI recommendation visibility, referral data and campaign activity, it can help identify possible AI-influenced demand.

8. AI Referral Share

Direct AI referral traffic remains an important metric, even though it captures only part of AI’s influence.

Referral traffic should be segmented wherever possible by:

• AI platform;
• landing page;
• sector;
• campaign;
• and conversion outcome.

9. AI-Influenced Conversion Rate

A direct AI click is not always necessary for AI to influence a conversion.

Where first-party survey data, CRM notes, assisted-conversion modelling or customer interviews are available, organisations can begin estimating AI-influenced outcomes.

Possible evidence of AI influence can include customers reporting that they:

• found the organisation through ChatGPT;
• received an AI recommendation;
• used Gemini to compare providers;
• or used an AI assistant before subsequently finding the brand through Google.

10. Decision-Stage Visibility

The strongest AI visibility often occurs close to a decision.

Decision-stage prompt sets can include:

• best provider;
• recommended company;
• best product for a specific use case;
• comparison between named brands;
• best option within a budget;
• and alternatives to a competitor.

This metric helps separate broad informational visibility from commercially meaningful visibility.

Recommended AI Search Behaviour Dashboard

Measure by Search Journey Stage

The same prompt should not be used to represent every stage of the customer journey.

A practical measurement programme should separate:

• Discovery prompts — broad problems and category questions;
• Research prompts — detailed informational questions;
• Comparison prompts — product, company and provider comparisons;
• Recommendation prompts — requests for preferred options;
• Verification prompts — questions about evidence, reviews, credibility or alternatives;
• Decision prompts — highly specific questions immediately before action.

This provides a more accurate picture of where the organisation enters and leaves the AI-assisted journey.

Measure by Platform

AI Search visibility should also be segmented by platform.

An organisation may perform strongly in one environment and weakly in another because different systems can use different:

• retrieval systems;
• source preferences;
• ranking logic;
• freshness signals;
• and citation patterns.

A practical dashboard may therefore monitor:

• ChatGPT;
• Google AI Overviews;
• Google AI Mode;
• Gemini;
• Microsoft Copilot;
• Perplexity;
• and additional platforms relevant to the market.

Measure Over Time

AI answers are dynamic.

A single test provides only a snapshot.

Measurement should therefore be repeated on a consistent schedule.

For high-priority commercial prompts, organisations may monitor:

• weekly visibility;
• monthly trend movement;
• quarterly competitive change;
• and major shifts after important website, content or authority improvements.

The objective is to distinguish:

Temporary Answer Variation

from:

Sustained Visibility Change.

Research Methodology

AI Search Behaviour Statistics UK 2026 was developed to provide a structured evidence base for understanding how people are changing the way they search, research, compare, verify and make decisions in an AI-mediated Search environment.

The research combines UK-specific consumer evidence with carefully identified international behavioural benchmarks where equivalent UK datasets were not available.

The objective was not to present every available AI statistic.

The objective was to identify statistics that help explain measurable changes in Search Behaviour.

1. Research Scope

The research focuses specifically on behavioural change associated with:

• AI Search adoption;
• conversational Search;
• prompt construction;
• follow-up questions;
• multi-step research;
• trust and verification;
• cross-platform Search behaviour;
• product and provider research;
• AI recommendations;
• shopping behaviour;
• sector-specific information seeking;
• voice Search;
• visual Search;
• multimodal Search;
• and AI-influenced decision-making.

Statistics dealing primarily with website traffic, referral volumes, click-through rates and revenue were excluded where those subjects were more appropriately covered in the separate CGO Media research on AI Search Traffic Statistics UK 2026.

This separation was designed to distinguish:

What People Do
from
What Traffic AI Generates

2. Evidence Hierarchy

Sources were prioritised according to an evidence hierarchy.

Where suitable evidence was available, preference was given to:

1. UK regulator and public-interest research;
2. nationally representative UK consumer surveys;
3. large observed clickstream datasets;
4. first-party platform usage data;
5. large international consumer surveys;
6. sector-specific behavioural research.

Examples of organisations represented within the final evidence set include:

• Ofcom;
• Which?;
• YouGov;
• Ipsos;
• Deloitte;
• Semrush;
• Similarweb;
• Adobe Digital Insights;
• Oxford Economics;
• EY;
• and Google.

3. UK Evidence Was Prioritised

Because this research is titled AI Search Behaviour Statistics UK 2026, UK-specific evidence was prioritised wherever reliable data existed.

Examples include:

• UK AI Search adoption;
• usage frequency;
• age differences;
• AI platform adoption;
• trust;
• follow-up behaviour;
• professional-advice substitution;
• shopping behaviour;
• travel research;
• health-information behaviour;
• and AI-assisted news consumption.

The research does not present US or global figures as though they directly represent UK consumers.

Where international evidence is used, it is explicitly classified as:

• US Consumer Benchmark;
• US Observed Behaviour Benchmark;
• US Clickstream Benchmark;
• Global Consumer Benchmark;
• Global Platform Data;
• or another appropriate geographic classification.

4. Consumer Surveys

Consumer survey evidence was included where the methodology and population provided a useful indication of actual or reported behaviour.

Important UK datasets used within the research include:

• Which? research based on 4,189 UK adults;
• Ofcom’s Adults’ Media Use and Attitudes research involving 7,533 UK adults;
• YouGov travel research;
• Ipsos UK health research;
• and Deloitte UK Digital Consumer Trends research.

Survey data measures what respondents report about their behaviour or attitudes.

It should therefore be distinguished from observed behavioural data such as clickstream records.

5. Observed Behaviour and Clickstream Evidence

Where available, observed behavioural datasets were used to complement survey evidence.

These sources provide insight into what users actually do across digital environments rather than relying solely on self-reported behaviour.

Examples include:

• Semrush analysis of Google and ChatGPT clickstream activity;
• Similarweb analysis of downstream behaviour following AI recommendations;
• and Adobe Analytics data examining AI-assisted consumer journeys.

Observed data is particularly useful for examining:

• prompt length;
• queries per session;
• web-search activation;
• downstream Search activity;
• referral patterns;
• and subsequent brand visits.

6. Platform-Reported Usage Data

First-party platform data was included where it provided unique information about behaviour that was not available through independent datasets.

Examples include Google reporting on:

• AI Mode usage growth;
• planning and brainstorming behaviour;
• voice and image queries;
• Google Lens usage;
• Circle to Search;
• AI Overviews reach;
• Gemini usage;
• voice interaction;
• camera and screen sharing;
• and attachment-based interactions.

First-party platform data is valuable because the platform owner has direct access to usage data.

However, it must also be interpreted within the context of that specific product ecosystem.

A statistic about Gemini users should not automatically be interpreted as representing all AI users.

7. Definition of a Statistic

A major purpose of this rewrite was to distinguish genuine statistics from strategic observations.

For inclusion among the 100 statistics, an item required a measurable numerical finding such as:

• a percentage;
• a ratio;
• a usage level;
• a growth rate;
• a measured behavioural difference;
• or a quantified platform metric.

Statements such as:

“AI Search is becoming more conversational”

were not treated as statistics unless supported by a measurable finding.

For example:

“Google AI Mode queries averaged 7.22 words compared with four words for traditional Google Search”

qualifies as a statistic because it is a measurable behavioural result.

8. Duplicate Findings Were Consolidated

Where several publications repeated the same underlying statistic, the research attempted to retain the most appropriate original or primary source rather than presenting duplicated figures as separate evidence.

Closely related statistics were retained only where they measured meaningfully different aspects of behaviour.

For example:

• overall AI usage;
• frequent AI usage;
• usage by younger adults;
• and usage by older adults

are treated separately because they describe different behavioural dimensions.

9. Statistics Were Organised Into Ten Behavioural Themes

The 100 statistics were grouped into ten research areas:

1. UK AI Search Adoption & Usage Frequency;
2. Conversational Search, Search Intent & Prompt Behaviour;
3. Follow-Up Questions, Multi-Step Research & Expanding Query Behaviour;
4. Trust, Verification & Source Checking;
5. AI Search vs Traditional Search & Professional Advice;
6. Product, Brand & Shopping Research;
7. AI Recommendations, Purchase Confidence & Decision Behaviour;
8. Demographics, Device Use & Multimodal Search;
9. Sector-Specific AI Search Behaviour;
10. Cross-Platform, AI-Integrated & Future Search Behaviour.

The structure is intended to show how behavioural change develops across the full Search Journey rather than presenting 100 isolated numbers.

10. Behavioural Interpretation Was Kept Separate From the Statistic

Each statistic is followed by a Behaviour implication.

This is deliberately separated from the underlying numerical finding.

The statistic represents the reported evidence.

The Behaviour implication represents CGO Media’s interpretation of what that evidence may mean for Search, discovery and organisational visibility.

This distinction is important because:

Evidence

and:

Interpretation

are not the same thing.

11. Correlation Was Not Treated as Causation

Where evidence identifies an association, this research does not automatically interpret that relationship as proof of causation.

For example, frequent AI Search users report higher trust than less frequent users.

This does not establish whether:

• greater trust causes more frequent usage;
• greater usage creates greater trust;
• or both are influenced by another factor.

Similarly, increases in branded Search occurring alongside stronger AI visibility should not automatically be attributed solely to AI.

Other influences can include:

• advertising;
• PR;
• seasonality;
• social media;
• offline activity;
• and broader Brand Awareness.

12. International Benchmarks Were Used Selectively

Some areas of AI Search Behaviour currently have richer datasets in the United States or internationally than in the UK.

These include:

• AI shopping behaviour;
• AI referral journeys;
• prompt construction;
• AI-assisted travel research;
• financial AI behaviour;
• and multimodal Search usage.

International evidence was included where it helped identify an emerging behaviour likely to be relevant to the development of Search.

These statistics are not presented as direct UK population estimates.

They are clearly labelled as comparative benchmarks.

13. Rapidly Changing Platform Metrics

AI Search is developing quickly.

Usage levels, platform features and Search behaviour can change materially within months.

Statistics relating to:

• AI Mode;
• AI Overviews;
• ChatGPT;
• Gemini;
• visual Search;
• and other generative interfaces

should therefore be interpreted as measurements for the period in which the underlying research was conducted.

They should not automatically be treated as permanent platform characteristics.

14. Source Verification

Each statistic was checked against the underlying source material available at the time of compilation.

Where possible, the research records:

• the source organisation;
• the relevant geography;
• the measured population;
• the study period;
• and the nature of the evidence.

Readers using an individual statistic in:

• journalism;
• academic work;
• commercial research;
• or other published material

should consult the original source before publication, particularly where current platform usage may have changed.

15. CGO Media Analytical Framework

After the statistical evidence was assembled, the findings were mapped against the CGO AI Search Behaviour Framework.

The framework organises Search Behaviour into seven stages:

1. Need Formation;
2. Question & Prompt Formation;
3. AI Discovery;
4. Conversational Research;
5. Verification & Cross-Checking;
6. Recommendation & Consideration;
7. Decision & Action.

The framework is an analytical model developed by CGO Media.

It should not be interpreted as a statistical finding in itself.

Research Limitations

AI Search Behaviour is changing rapidly, and the evidence base is still developing.

This research therefore needs to be interpreted with several important limitations in mind.

1. Not All Statistics Are UK-Specific

The research prioritises UK evidence wherever suitable data is available.

However, some areas of AI Search Behaviour currently have much richer datasets in the United States or at global level.

These include:

• AI shopping behaviour;
• prompt and session behaviour;
• AI referral journeys;
• multimodal Search;
• financial AI usage;
• and platform-level usage patterns.

International statistics are therefore used as comparative behavioural benchmarks.

They should not be interpreted as direct estimates of the UK population.

2. Survey Data Measures Reported Behaviour

A significant proportion of the evidence comes from consumer surveys.

Survey research measures what respondents say they:

• do;
• believe;
• trust;
• or intend to do.

Reported behaviour may not always correspond perfectly with observed digital behaviour.

For example, a respondent may:

• overestimate how often they use AI;
• underestimate how often they verify information;
• or describe intentions that do not later result in action.

Where possible, survey evidence has therefore been complemented by observed clickstream and platform usage data.

3. Clickstream Data Has Coverage Limitations

Observed clickstream research provides valuable evidence of real digital behaviour, but it does not capture every user or every device.

Depending on the dataset, limitations can include:

• desktop-only measurement;
• specific geographic markets;
• users participating in measurement panels;
• limited visibility into mobile apps;
• and incomplete visibility into logged-in or private interactions.

Clickstream results should therefore be interpreted as observed behaviour within the measured population rather than complete internet-wide behaviour.

4. AI Conversations Are Often Private

A large proportion of AI Search activity takes place inside private conversational interfaces.

This makes the emerging Search environment considerably harder to observe than conventional Search.

Traditional Search can often be studied through:

• keyword databases;
• Search Console;
• ranking tools;
• and public Search result pages.

AI conversations are often not publicly visible.

Researchers therefore have limited access to the complete universe of prompts users submit.

This creates an unavoidable measurement gap between:

Observed AI Search Behaviour

and:

Total AI Search Behaviour.

5. Platform Data Is First-Party Evidence

Several statistics in this research come directly from AI and Search platform operators.

These sources can provide data unavailable elsewhere, including:

• active-user numbers;
• voice usage;
• image Search volume;
• multimodal interaction;
• and feature adoption.

However, platform-reported statistics are also produced by organisations with a commercial interest in demonstrating product adoption and growth.

They should therefore be interpreted alongside independent research wherever possible.

6. Platform Definitions May Differ

Terms such as:

• active user;
• AI Search user;
• query;
• session;
• AI-assisted shopper;
• and frequent user

may be defined differently across studies.

For example, one research organisation may define frequent AI use as several times per week, while another may measure monthly activity.

Statistics from different studies should therefore not automatically be compared as though the definitions are identical.

7. Search and AI Product Features Change Quickly

AI Search products are evolving unusually quickly.

Features can change within weeks or months.

Examples include changes to:

• AI Overviews;
• AI Mode;
• ChatGPT Search;
• Gemini;
• Copilot;
• visual Search;
• voice interfaces;
• and AI agents.

A usage statistic measured in early 2026 may therefore describe a materially different product experience from the same platform later in the year.

Readers should always consider the measurement period attached to rapidly changing platform statistics.

8. AI Search and Traditional Search Overlap

The distinction between AI Search and traditional Search is becoming increasingly difficult to define.

A Google user may receive:

• a conventional Search result;
• an AI Overview;
• an AI Mode response;
• or a mixture of generated and conventional results.

Similarly, an AI assistant may activate web Search during a conversation.

This means users can cross between:

Search Engine Behaviour

and:

AI Assistant Behaviour

inside the same platform or session.

Future research may therefore require a broader definition of Search than historical channel classifications provide.

9. AI Influence Is Difficult to Attribute

One of the largest measurement limitations concerns attribution.

A consumer may:

• discover a brand in ChatGPT;
• search the brand on Google later;
• visit the website through organic Search;
• and make a purchase.

Traditional analytics may attribute that conversion to Google.

Yet the original brand discovery may have occurred within AI.

This creates what can be described as:

Visible Conversion Attribution
vs
Hidden AI Influence

Current analytics systems do not consistently capture that distinction.

10. Recommendation Visibility Can Vary Between Tests

AI-generated answers are not always deterministic.

The same prompt may produce different:

• brands;
• sources;
• recommendations;
• or wording

across repeated tests.

Variation can result from:

• model updates;
• retrieval changes;
• location;
• personalisation;
• fresh web results;
• conversation history;
• or stochastic generation.

AI visibility should therefore be measured repeatedly rather than from one isolated answer.

11. Personalisation Can Change Results

AI systems increasingly incorporate user context.

The same question may generate different results according to:

• location;
• language;
• previous conversation;
• account history;
• stated preferences;
• device context;
• or supplied files and images.

This means there may be no single universal answer to a commercial AI Search query.

Visibility research therefore needs to consider multiple realistic customer contexts.

12. Sector Behaviour Is Not Uniform

Consumers do not use AI in the same way for every decision.

The willingness to rely on AI may differ substantially between:

• buying clothing;
• choosing accounting software;
• planning a holiday;
• researching symptoms;
• selecting a bank account;
• or seeking legal information.

Sector-specific evidence should therefore be prioritised over generic AI usage data wherever available.

13. High Adoption Does Not Equal High Influence

Using an AI tool does not automatically mean the tool materially influenced the final decision.

A consumer may:

• ask a general question;
• ignore the answer;
• perform additional research;
• or ultimately choose a different option.

Similarly, a consumer who does not click an AI-generated link may still be influenced by the brand mention.

Usage, visibility, influence and conversion are therefore separate concepts.

14. Trust Does Not Mean Accuracy

Several studies show meaningful consumer trust in AI-generated answers.

That should not be interpreted as evidence that the answers are always accurate.

Consumer trust is a behavioural measure.

Accuracy is an information-quality measure.

The two should remain distinct.

This is particularly important in areas including:

• health;
• finance;
• law;
• and other high-stakes decisions.

15. Published Statistics May Be Superseded

AI adoption and platform usage are increasing quickly.

The statistics in this paper represent the strongest available evidence identified during the 2026 research period.

Some figures will inevitably be superseded by newer studies.

CGO Media therefore recommends that journalists, researchers and businesses check the publication date and original source before using individual statistics in future work.

Recommended Related CGO Media Research

AI Search Behaviour should not be considered in isolation.

Changes in how people search are directly connected to:

• AI-generated answers;
• source selection;
• citation systems;
• brand authority;
• entity understanding;
• Search visibility;
• website traffic;
• and commercial outcomes.

The following CGO Media research and frameworks provide additional context for understanding the wider Search environment described in this study.

AI Search Traffic Statistics UK 2026

The companion statistical research examining what happens after AI-assisted discovery, including AI referrals, zero-click behaviour, engagement, conversion, attribution and commercial value.

Where this paper examines what people do, the AI Search Traffic study examines how those behaviours translate into measurable traffic and commercial activity.

Explore AI Search Traffic Statistics UK 2026 →

CGO AI Search Readiness Framework

The AI Search Readiness Framework examines whether an organisation has the technical, content, entity and authority signals required to compete across AI-powered discovery systems.

It helps connect consumer Search Behaviour with organisational readiness.

Explore the CGO AI Search Readiness Framework →

CGO AI Citation Framework

AI Search users increasingly encounter brands and organisations through sources selected by generative systems.

The CGO AI Citation Framework examines the conditions that can influence whether a source becomes eligible for citation within AI-generated answers.

Explore the CGO AI Citation Framework →

CGO AI Authority Model

The CGO AI Authority Model examines how multiple forms of authority can combine to influence visibility within AI-generated Search environments.

The model considers the relationship between:

• Brand Authority;
• Entity Authority;
• Content Authority;
• Citation Authority;
• and broader Digital Authority.

Explore the CGO AI Authority Model →

CGO Brand Signal Framework

AI Search systems increasingly need to identify not only relevant documents, but credible organisations and entities.

The CGO Brand Signal Framework examines the digital signals that can help strengthen machine understanding of a brand across the wider Search ecosystem.

Explore the CGO Brand Signal Framework →

CGO Content Authority Framework

AI-assisted research increases the importance of content that is clear, useful, evidence-led and capable of supporting both human decisions and machine-generated answers.

The CGO Content Authority Framework examines how organisations can build stronger information assets around subject expertise and customer needs.

Explore the CGO Content Authority Framework →

CGO Media Knowledge Architecture Map

The Knowledge Architecture Map connects the wider CGO Media research programme across:

• Search Behaviour;
• AI Search;
• GEO;
• authority;
• citations;
• entities;
• content;
• and organisational visibility.

It is designed to show how individual research papers and frameworks connect rather than treating each subject as an isolated discipline.

Explore the CGO Media Knowledge Architecture Map →

CGO Media Statistics Library

The Statistics Library contains quantitative resources covering the changing Search, AI and digital-discovery environment.

It is intended for:

• business leaders;
• journalists;
• researchers;
• marketers;
• and Search professionals.

Explore the CGO Media Statistics Library →

Latest CGO Media Research

AI Search develops quickly, so new evidence can supersede individual statistics or reveal new forms of Search Behaviour.

The Latest Research page provides access to newly published CGO Media studies, datasets, observations and framework updates.

Explore the Latest CGO Media Research →

Frequently Asked Questions About AI Search Behaviour

What is AI Search Behaviour?

AI Search Behaviour describes how people use AI-powered systems to find information, ask questions, compare options, verify answers, receive recommendations and make decisions.

It includes behaviour across platforms such as:

• ChatGPT;
• Google AI Overviews;
• Google AI Mode;
• Gemini;
• Microsoft Copilot;
• Perplexity;
• and other generative AI systems.

How is AI Search Behaviour different from traditional Search Behaviour?

Traditional Search usually begins with a relatively short keyword or phrase and returns a ranked list of webpages.

AI Search allows users to provide much more context and continue the research as a conversation.

The behavioural shift can be represented as:

Keyword → Results Page
becoming
Question → Answer → Follow-Up → Comparison → Decision

AI Search can therefore support a larger proportion of the research journey within one interface.

How many UK adults use AI Search?

Which? found that 51% of UK adults use generative AI tools to search for products, services or advice online.

It also found that 24% use AI Search frequently, meaning daily or several times each week.

Ofcom’s broader 2026 AI research found that 54% of UK adults use AI tools.

The two figures measure related but slightly different behaviours and should not be treated as identical measures.

Which age group uses AI Search most?

Younger adults currently show the highest adoption.

Which? found that 75% of UK adults aged 18–34 use AI tools to search the internet, while 42% use them frequently.

Ofcom separately found that 79% of UK adults aged 16–24 use AI tools and 74% of those aged 25–34 do so.

Is AI replacing Google Search?

The current evidence suggests a more complex relationship.

Traditional Search remains heavily used, while AI Search is being added to the customer journey.

Which? found that 82% of UK adults still use traditional Search engines frequently.

At the same time, users increasingly move between AI and traditional Search.

Similarweb found that 55.9% of measured AI-influenced website visits arrived through Search, while Semrush found Google receiving 21.6% of ChatGPT outbound referral traffic by February 2026.

The emerging behaviour is therefore better represented as:

AI Search ↔ Traditional Search

Do people trust AI Search results?

Yes, but trust is not absolute.

Which? found that:

• 47% of UK AI Search users trust results to a reasonable or great extent;
• 35% trust them to some extent;
• 15% report little or no trust.

Trust is higher among frequent AI Search users.

Do people verify AI-generated answers?

Yes.

Which? found that 55% of AI Search users ask follow-up questions to improve results.

Among frequent users, this rises to 73%.

It also found that 23% often or always cross-check the output of one AI platform against another, rising to 36% among frequent users.

Do consumers use AI Search for shopping?

Yes.

Adobe found that 35% of UK consumers had already used generative AI to assist with shopping by March 2025.

A further 47% said they planned to use generative AI for shopping during 2025.

International research shows AI being used for:

• product recommendations;
• brand research;
• deal finding;
• gift ideas;
• and shopping lists.

Can AI Search influence what consumers eventually buy?

Evidence increasingly suggests that it can.

Adobe found that among consumers using AI for online shopping:

• 65% said AI increased purchase confidence;
• more than 55% clicked AI-recommended links;
• 87% were more likely to use AI for large or complex purchases.

Similarweb also found that users receiving a ChatGPT brand recommendation were 2.5 times more likely to visit that brand within seven days.

Does AI Search generate website traffic?

Yes, but direct referral traffic represents only one part of AI influence.

A user may discover a brand through AI and later reach the website through:

• Google Search;
• direct navigation;
• a marketplace;
• a review site;
• or another platform.

This means AI can influence demand without receiving direct attribution for the final visit.

How are AI Search queries different from Google queries?

AI queries tend to contain more context and can become part of an ongoing conversation.

Semrush found Google AI Mode queries averaged 7.22 words, compared with approximately 4 words for traditional Google Search.

Its longer-term ChatGPT research also found that many AI prompts did not resemble conventional Search keyword language.

Do AI users ask follow-up questions?

Yes.

Follow-up behaviour is one of the defining differences between AI Search and conventional Search.

Semrush found average ChatGPT queries per session increasing to 1.75 by February 2026.

Which? separately found that 55% of UK AI Search users actively use follow-up questions to improve the answer.

What is multimodal Search?

Multimodal Search allows people to search using more than typed text.

A query can include:

• voice;
• images;
• camera input;
• screen content;
• documents;
• or other files.

Google reports that nearly one in six AI Mode queries now uses voice or images, while Google Lens processes more than 25 billion visual Searches each month.

Why is visual Search important for businesses?

Visual Search enables consumers to search for something they can see even when they do not know its name.

Google reports that one in four Lens searches has commercial intent.

For product-based businesses, this increases the importance of:

• high-quality product imagery;
• accurate product data;
• structured information;
• and clear entity relationships.

Do people use AI Search for healthcare information?

Yes.

Ipsos found that 28% of Britons had used an AI chatbot for information about their own health.

Deloitte separately found that 26% of UK generative AI users had sought health advice through AI.

Health-related AI use should be interpreted carefully because consumer trust in an AI answer does not guarantee medical accuracy.

Do people use AI Search for financial decisions?

Yes.

Adobe’s US research found that 27% of surveyed consumers used generative AI for banking and financial needs.

EY’s 2026 international research found that 49% of surveyed consumers had used AI to support savings or investment decisions.

These international figures should be treated as comparative behavioural benchmarks rather than direct UK population estimates.

Do consumers use AI for travel planning?

Yes.

YouGov found that 20% of UK travellers were comfortable using AI for trip planning in 2025.

Among UK travellers aged 25–34, the figure rose to 35%.

AI is particularly suited to travel because it can combine destinations, accommodation, transport, activities and personal preferences within one research conversation.

What does AI recommendation visibility mean?

AI recommendation visibility measures whether an organisation appears when a user asks an AI system to suggest, compare or recommend products, companies or providers.

It is a stronger signal than a simple Brand Mention.

An organisation may be visible in an informational answer but absent when the user asks:

“Which one would you recommend?”

How should businesses measure AI Search Behaviour?

CGO Media recommends measuring several layers rather than relying on one metric.

These include:

• Prompt Coverage Rate;
• AI Visibility Rate;
• Follow-Up Retention Rate;
• Recommendation Visibility Rate;
• AI Citation Rate;
• Cross-Platform Consistency;
• Branded Search Lift;
• AI Referral Share;
• AI-Influenced Conversion Rate;
• and Decision-Stage Visibility.

Why is AI Search difficult to measure?

Much of AI Search happens inside private conversational interfaces.

Users may also discover a brand through AI and later convert through a different channel.

This can make AI influence invisible to conventional analytics.

The challenge is therefore not simply measuring AI referral traffic.

It is measuring how AI influences:

• awareness;
• trust;
• comparison;
• branded Search;
• website visits;
• and eventual decisions.

What is the biggest change in Search Behaviour caused by AI?

The most important change is the movement from finding information toward interacting with information.

Traditional Search generally asks:

“Which page should I visit?”

AI Search increasingly allows the user to ask:

“Help me understand, compare and decide.”

That changes both the Search Journey and the way organisations need to think about visibility.

Conclusion — AI Search Behaviour in 2026

The 100 statistics examined in this research show that Search Behaviour is undergoing a structural change.

The shift is not simply from Google to ChatGPT.

It is from a Search model dominated by short queries and ranked links toward a broader discovery environment built around:

• questions;
• conversation;
• context;
• recommendations;
• verification;
• multimodal interaction;
• and AI-assisted decision-making.

Traditional Search remains central to the customer journey.

But AI increasingly participates before, during and after the conventional Search interaction.

AI Search Is Becoming Mainstream

The UK evidence already shows widespread adoption.

More than half of UK adults use AI tools for Search-related activity, while adoption among younger adults is substantially higher.

The most important implication is not simply that more people are using AI.

It is that frequent users are beginning to behave differently.

They are more likely to:

• ask follow-up questions;
• cross-check platforms;
• use AI for complex research;
• seek recommendations;
• and integrate AI into important decisions.

The Search Query Is Expanding

Search historically required people to translate a need into a short keyword phrase.

Generative AI removes much of that constraint.

The user can increasingly explain:

• who they are;
• what they need;
• what they can afford;
• what they prefer;
• what they want to avoid;
• and what outcome they want.

The Search system then helps structure the decision.

This means Search Intent is becoming richer and more contextual.

The Search Journey Is Becoming Conversational

The first response is increasingly not the end of the Search.

Users can continue by asking:

• “Which is best?”
• “What are the alternatives?”
• “What if my budget is lower?”
• “Which one would you recommend?”
• “Can you compare those?”
• “What evidence supports that?”

The customer journey therefore becomes:

Question
↓
Answer
↓
Follow-Up
↓
Refinement
↓
Comparison
↓
Recommendation
↓
Decision

Trust Is Becoming Conditional Rather Than Absolute

Consumers are increasingly willing to use and trust AI-generated information.

But many also:

• ask follow-up questions;
• cross-check another AI system;
• search Google;
• visit source websites;
• or consult professional information.

This means AI Search does not eliminate the need for trusted sources.

It increases their importance.

The stronger the underlying evidence ecosystem, the easier it becomes for users and AI systems to validate the same information.

AI Is Moving Into Commercial Decisions

AI is increasingly involved in:

• product research;
• travel planning;
• financial research;
• technology purchases;
• health information;
• professional-services research;
• and provider selection.

This matters because AI visibility can influence the customer before a measurable website visit occurs.

A consumer may first encounter a brand inside an AI answer and only later:

• Search the company name;
• visit the website;
• read reviews;
• compare alternatives;
• or make contact.

The influence therefore begins upstream of the final click.

AI Search and Google Search Are Becoming Interconnected

The evidence throughout this research repeatedly shows movement between AI and traditional Search.

The emerging journey is not:

AI or Google.

It is increasingly:

AI ↔ Google ↔ Websites ↔ Third-Party Sources

Different environments can perform different functions.

AI may help:

• understand;
• compare;
• or recommend.

Google may then help:

• verify;
• navigate;
• locate;
• or continue the research.

The customer may move between them several times before acting.

Search Is Becoming Multimodal

Typing is no longer the only way to Search.

Users can increasingly Search through:

• voice;
• photographs;
• screens;
• live camera views;
• documents;
• and other contextual inputs.

This creates Search opportunities that did not exist within a purely keyword-based model.

The user no longer always needs to know how to describe what they want.

They can increasingly:

Show It
Ask About It
Understand It
Act On It

The Measurement Model Must Change

Search performance can no longer be understood through rankings and clicks alone.

Organisations increasingly need to measure:

• Prompt Coverage;
• AI Visibility;
• Follow-Up Retention;
• Citation Visibility;
• Recommendation Visibility;
• Cross-Platform Consistency;
• Branded Search Demand;
• AI Referral Traffic;
• and AI-Influenced Commercial Outcomes.

This provides a fuller view of how visibility contributes to the customer journey.

References

The following sources provide the principal statistical and methodological evidence used throughout AI Search Behaviour Statistics UK 2026.

Where international research is included, it is identified throughout the paper as comparative evidence rather than being presented as directly representative of UK consumers.

1. Which? (2025). Consumer Use and Attitudes Towards AI Search Tools. Research based on 4,189 UK adults aged 18+. View source.
2. Ofcom. (2025). From Apps to AI Search: How the UK Goes Online in 2025. View source.
3. Ofcom. (2026). Adults’ Media Use and Attitudes 2026. Research examining UK adult media and AI usage. View source.
4. Ofcom. (2026). One in Five Brits Using AI to Keep Up With News. News Consumption Report 2026. View source.
5. Semrush. (2025). Google AI Mode’s Early Adoption and SEO Impact. Analysis of almost 69 million US desktop Google Search sessions. View source.
6. Semrush. (2026). ChatGPT Traffic Analysis: Insights from 17 Months of Clickstream Data. Analysis based on more than one billion lines of US clickstream data. View source.
7. Similarweb. (2026). The Downstream Impact of AI Visibility. Research examining behaviour following AI-generated brand recommendations. View source.
8. Adobe Digital Insights. (2025). The Explosive Rise of Generative AI Referral Traffic. Consumer research covering AI-assisted shopping, financial services and commercial discovery. View source.
9. Adobe Digital Insights UK. (2025). Retail in Flux. UK research examining generative AI adoption within shopping behaviour. View source.
10. Adobe Digital Insights. (2025). AI-Sourced Traffic Insights 2025. Research examining AI-assisted technology purchases, engagement and consumer behaviour. View source.
11. Adobe Digital Insights. (2025–2026). AI-Driven Traffic Surges Across Industries. Research examining AI shopping behaviour, purchase confidence and recommendation engagement. View source.
12. Adobe & Oxford Economics. (2026). AI and Digital Trends 2026: Customer Behaviours and AI. International consumer research covering AI recommendations, personalisation and decision behaviour. View source.
13. YouGov. (2025). Brits More Comfortable Using AI for Trip Planning Than Last Year. UK travel consumer research. View source.
14. Ipsos. (2026). Britons Turn to Online Sources and AI to Check Symptoms. UK research examining AI use for personal health information. View source.
15. Deloitte UK. (2026). Digital Consumer Trends: Generative AI and Health Advice. UK consumer research examining AI-assisted health and wellbeing behaviour. View source.
16. EY. (2026). Nearly Half of Global Consumers Now Use AI to Guide Savings and Investment Decisions. International financial-services consumer research. View source.
17. Google. (2026). How AI Mode Is Changing and Expanding the Way People Search. Platform research examining planning, brainstorming and conversational Search behaviour. View source.
18. Google. (2026). Alphabet Q4 2025 Search and AI Usage Update. Platform data covering AI Mode, multimodal queries and Circle to Search. View source.
19. Google. (2025). Visual Search and Google Lens Usage. Platform data covering monthly Lens queries and commercial-intent behaviour. View source.
20. Google. (2026). More Than One Billion People Are Using the Gemini App Every Month. Platform data covering voice, multimodal, attachment and device behaviour. View source.

CGO Media Research Ecosystem

AI Search Behaviour Statistics UK 2026 forms part of the wider CGO Media Search and AI Research Programme.

The programme examines how human Search Behaviour, generative systems, source selection, authority and digital visibility are converging into a broader Search ecosystem.

Rather than treating AI Search, traditional SEO and GEO as separate disciplines, CGO Media’s research architecture examines how they connect.

Research Library

The CGO Media Research Library contains longer-form research examining the evolution of Search, AI answer construction, recommendation systems, source selection, entity authority and Digital Authority.

Explore the CGO Media Research Library →

Framework Library

The Framework Library translates research findings into structured models covering areas including:

• AI Search Readiness;
• AI Citation Authority;
• Entity Authority;
• Content Authority;
• Brand Signals;
• Search Visibility;
• Technical SEO;
• and GEO.

Explore the CGO Media Framework Library →

Statistics Library

The CGO Media Statistics Library brings together quantitative research covering Search Behaviour, AI Search, traffic, visibility, authority and commercial performance.

Explore the CGO Media Statistics Library →

Latest Research

The Latest Research resource provides access to newly published CGO Media studies, datasets, research observations and framework updates.

Explore Latest Research →

Research Methodology

The wider CGO Media methodology explains the evidence classification, source-selection principles and analytical approach used across the research programme.

Read the CGO Media Research Methodology →

About the Research

Roger Wilkinson

Roger Wilkinson is an independent researcher, SEO practitioner and Founder of CGO Media.

His research focuses on the transition from conventional Search toward an environment increasingly shaped by generative AI, conversational discovery, source selection, citations, recommendation systems and Digital Authority.

Current research areas include:

• AI Search Behaviour;
• Generative Engine Optimisation;
• AI Answer Construction;
• Source Selection;
• AI Citation Authority;
• Entity Authority;
• Brand Authority;
• Knowledge Architecture;
• Search Visibility;
• and the future of Search.

ORCID:

0009-0004-3325-0740

CGO Media Research Team

The CGO Media Research Team develops independent research, statistical resources and practical frameworks examining how Search, AI Search and GEO are changing organisational visibility.

The research programme places particular emphasis on:

• source provenance;
• evidence classification;
• geographic relevance;
• clear distinction between statistics and interpretation;
• and recognition of methodological limitations.

Meet the CGO Media Research Team →

Research Usage & Citation

AI Search Behaviour Statistics UK 2026 may be referenced by journalists, researchers, academics, businesses and other organisations examining changes in Search Behaviour and AI-assisted discovery.

When quoting an individual third-party statistic, CGO Media recommends consulting and citing the original underlying source as well as this synthesis where appropriate.

Recommended Citation

DOI

The research record is associated with:

DOI: 10.5281/zenodo.22957969

Where a repository citation format is required, researchers should use the bibliographic details provided on the corresponding repository record.

Journalists & Media

Journalists and media organisations may reference the findings and CGO Media analysis with clear attribution and a link to the research page.

Where a statistic originates from an external survey, platform dataset or research organisation, the original source should also be reviewed before publication.

For media enquiries, methodology questions, expert commentary and access to CGO Media research resources, visit:

CGO Media Press & Media Resources →

Academic & Research Use

Researchers may cite the statistical synthesis, analytical interpretation and CGO AI Search Behaviour Framework while preserving attribution to the original underlying evidence where relevant.

The research is intended to support further investigation into:

• AI Search Behaviour;
• conversational Search;
• consumer trust;
• prompt behaviour;
• AI-assisted commercial research;
• multimodal discovery;
• AI recommendations;
• Search Attribution;
• and the changing structure of digital decision journeys.

Business Use

Businesses and organisations may use the CGO AI Search Behaviour Framework to structure internal research and measurement programmes around:

1. Need Formation;
2. Question & Prompt Formation;
3. AI Discovery;
4. Conversational Research;
5. Verification & Cross-Checking;
6. Recommendation & Consideration;
7. Decision & Action.