GEO Statistics UK 2026

Statistic 11

61.7% of measured AI citations were “ghost citations”

Research analysing 3,981 domain appearances across 115 prompts and 14 countries found that approximately 61.7% of citation appearances were “ghost citations” — situations where a domain contributed as a source without receiving an equivalent brand mention in the generated answer.

Statistic 12

74.9% of measured domain appearances included a citation

Across the tested generative platforms, approximately 74.9% of measured domain appearances involved the domain being cited as a source.

Statistic 13

Only 38.3% of measured appearances included a brand mention

While citations were recorded in almost three-quarters of measured appearances, the associated organisation or brand was explicitly mentioned in only 38.3%.

Statistic 14

Only 13.2% of appearances achieved both citation and brand mention

Approximately 13.2% of measured appearances combined a citation with an explicit brand mention, illustrating how difficult it can be to achieve both source visibility and brand visibility simultaneously.

Statistic 15

ChatGPT produced citations in approximately 87% of measured appearances

Within the study, ChatGPT demonstrated a particularly strong tendency to support generated answers with source citations, with a measured citation rate of approximately 87%.

Statistic 16

Only 20.7% of measured ChatGPT appearances included a brand mention

Despite ChatGPT’s high citation rate, the corresponding measured brand-mention rate was only 20.7%.

Statistic 17

Gemini mentioned brands in 83.7% of measured appearances

Google Gemini displayed a very different visibility pattern. The study recorded a brand-mention rate of approximately 83.7%.

Statistic 18

Gemini cited sources in only 21.4% of measured appearances

In contrast with its high brand-mention rate, Gemini’s measured citation rate was approximately 21.4%, demonstrating a markedly different relationship between brands and visible supporting sources.

Statistic 19 • CGO Media Calculation

ChatGPT’s measured citation rate was more than four times its brand-mention rate

Using the reported rates of 87% for citations and 20.7% for brand mentions, CGO Media calculates that ChatGPT’s citation rate was approximately 4.2 times its brand-mention rate within the tested dataset.

Statistic 20

Comparative content generated 2.4× more brand mentions

Within the analysed dataset, comparative content was associated with approximately 2.4 times more brand mentions, making comparison-oriented content particularly relevant to GEO and recommendation visibility.

GEO Visibility Should Be Measured in Layers

Source Retrieved
→
Source Cited
→
Brand Mentioned
→
Brand Compared
→
Brand Recommended

What Businesses Should Measure

A GEO reporting framework should not rely on a single visibility percentage. At minimum, organisations should distinguish between:

• AI citation frequency.
• Brand mention frequency.
• Citation and mention overlap.
• Recommendation inclusion.
• Comparison inclusion.
• First-party versus third-party citations.
• Platform-specific visibility.
• Prompt-category visibility.

Primary Evidence Source — Statistics 11–20

Semrush / Kevin Indig — AI Citation and Brand Mention Research
Analysis of 3,981 domain appearances across 115 prompts and 14 countries, examining visibility patterns across ChatGPT, Google AI Overviews, Google AI Mode and Gemini. Used for citation rates, brand-mention rates, ghost citations, platform differences and comparative-content visibility.

Statistic 21

Only around 12% of AI-cited URLs overlapped Google’s top ten

An Ahrefs analysis of approximately 15,000 prompts found average URL-level overlap of only around 12% between AI citations and Google’s top-ten organic results for the original prompt.

Statistic 22

AI citation overlap with Bing’s top ten was approximately 10%

The same research found average URL-level overlap of approximately 10% between AI citations and Bing’s top-ten results.

Statistic 23 • CGO Media Calculation

Approximately 88% of AI citation URLs did not overlap Google’s top ten

If average exact-URL overlap with Google’s top ten was approximately 12%, the complementary share is approximately 88%. This illustrates the extent to which the AI citation environment can extend beyond the exact URLs ranking on Google’s first page.

Statistic 24

37.1% of Google AI Overview citation URLs ranked in Google’s organic top ten

In separate large-scale Ahrefs research examining approximately 863,000 keyword SERPs and four million AI Overview citation URLs, around 37.1% of cited URLs ranked within Google’s organic top ten.

Statistic 25

26.2% of AI Overview citations ranked between positions 11 and 100

Approximately 26.2% of URLs cited by Google AI Overviews ranked somewhere between organic positions 11 and 100 for the analysed searches.

Statistic 26

36.7% of AI Overview citation URLs did not rank in Google’s top 100

The study found that approximately 36.7% of citation URLs were not present within the first 100 organic Google results for the analysed query.

Statistic 27 • CGO Media Calculation

62.9% of AI Overview citations came from outside Google’s top ten

Combining the reported 26.2% ranking between positions 11 and 100 with the 36.7% outside the top 100 gives approximately 62.9% of cited URLs originating outside Google’s organic top ten.

Statistic 28

ChatGPT exact-URL overlap with Google’s top ten was approximately 10%

A separate Ahrefs analysis of 3,311 short-tail terms found only approximately 10% exact-URL overlap between ChatGPT citations and Google’s organic top-ten results.

Statistic 29

ChatGPT domain-level overlap with Google’s top ten reached 31.8%

When the analysis moved from matching the exact URL to matching the underlying domain, overlap increased substantially to approximately 31.8%.

Statistic 30 • CGO Media Calculation

Domain overlap was approximately 3.2× higher than exact-URL overlap

Comparing the reported 31.8% domain overlap with 10% exact-URL overlap gives a ratio of approximately 3.2 to 1. This suggests that ChatGPT may recognise or retrieve a domain associated with a topic without necessarily selecting the exact page that Google ranks in its top ten.

Central GEO Finding

Ranking highly in Google can increase the potential evidence available to generative systems, but conventional first-page ranking is neither a guaranteed requirement nor a guarantee of being selected as an AI citation.

The Domain May Matter Beyond the Ranking URL

The difference between 10% exact-URL overlap and 31.8% domain overlap is particularly important.

Topic Authority
→
Domain Recognition
→
Relevant Page Retrieval
→
Citation Eligibility

SEO vs GEO Is the Wrong Question

The evidence does not support abandoning traditional SEO in favour of GEO. Nor does it support assuming that conventional SEO automatically produces generative visibility.

The stronger model is SEO + GEO: improve conventional discoverability while simultaneously strengthening the signals and evidence required for retrieval, citation, brand recognition and recommendation.

Primary Evidence Sources — Statistics 21–30

Ahrefs — AI Citations vs Google and Bing Search Results
Approximately 15,000 prompts. Used for average URL-level overlap between AI citations and conventional search top-ten results.

Ahrefs — Google AI Overview Citation Analysis
Approximately 863,000 keyword SERPs and four million AI Overview citation URLs. Used for top-ten, positions 11–100 and outside-top-100 citation distribution.

Ahrefs — ChatGPT Citations vs Google Rankings
3,311 short-tail terms. Used for exact-URL and domain-level overlap between ChatGPT citations and Google’s organic top ten.

Statistic 31

Citation rates increased from 50% to 68% with higher reasoning

Research comparing minimal- and high-reasoning AI configurations across 100 prompts and 20 buyer journeys found that the measured citation rate increased from approximately 50% under minimal reasoning to 68% under high reasoning.

Statistic 32 • CGO Media Calculation

Higher reasoning produced an 18-percentage-point increase in citation rate

The increase from 50% to 68% represents an 18-percentage-point difference between the minimal- and high-reasoning configurations.

Statistic 33

Average citations increased from 2.6 to 4.5

The average number of citations associated with the tested responses increased from approximately 2.6 under minimal reasoning to 4.5 under high reasoning.

Statistic 34 • CGO Media Calculation

Average citation volume increased by approximately 73%

Moving from an average of 2.6 citations to 4.5 represents an increase of approximately 73.1%.

Statistic 35

High reasoning produced 4.6× more query fan-out

The study found approximately 4.6 times greater query fan-out under the high-reasoning configuration, showing how one user request can expand into a substantially larger retrieval process.

Statistic 36

Minimal reasoning generated 245 web searches

Across the tested buyer journeys, the minimal-reasoning configuration generated approximately 245 web searches.

Statistic 37

High reasoning generated 1,130 web searches

Under high reasoning, the number of web searches increased to approximately 1,130 across the same research framework.

Statistic 38 • CGO Media Calculation

High reasoning generated approximately 885 additional searches

The difference between 1,130 and 245 represents approximately 885 additional web searches generated by the high-reasoning configuration across the tested journeys.

Statistic 39

Only 25.6% of cited domains overlapped between reasoning modes

Despite both configurations addressing the same underlying buyer journeys, only approximately 25.6% of cited domains overlapped between minimal and high reasoning.

Statistic 40

99 domains appeared only under high reasoning

The study identified 99 domains that appeared as cited sources only under the high-reasoning configuration, demonstrating how deeper retrieval can introduce entirely new sources into the generative answer environment.

Central GEO Finding

The user’s visible prompt may be only the beginning of the search process. A generative system can create multiple underlying searches, explore different aspects of the request and retrieve sources the user never explicitly searched for.

User Prompt
→
Query Decomposition
→
Multiple Searches
→
Source Retrieval
→
Evidence Comparison
→
Generated Answer

From Keyword Targeting to Knowledge Architecture

Query fan-out increases the importance of connected information. A strong GEO architecture may therefore include:

• A clear primary topic or service page.
• Supporting research and evidence pages.
• Definitions and entity clarification.
• Comparison and alternative-use-case content.
• Pricing and commercial information where appropriate.
• Methodology and evidence documentation.
• Case studies and demonstrable experience.
• FAQs addressing secondary questions.
• External corroboration and third-party authority signals.
• Strong internal relationships between relevant resources.

Primary Evidence Source — Statistics 31–40

Semrush / Kevin Indig — AI Reasoning, Query Fan-Out & Source Selection Research
Study covering 100 prompts across 20 buyer journeys and comparing minimal- and high-reasoning configurations. Used for citation rates, citation volume, query fan-out, web-search volume, cited-domain overlap and high-reasoning-only source discovery.

Statistic 41

AI-cited content was approximately 25.7% fresher than Google organic content

Ahrefs research found that content cited by AI platforms was approximately 25.7% fresher, on average, than comparable content appearing within Google organic search results.

Statistic 42

The freshness analysis examined approximately 17 million AI citations

The research analysed approximately 17 million citations, giving the study a large observational base for examining differences in source recency.

Statistic 43

Seven AI platforms were included in the analysis

The approximately 17 million citations were collected across seven AI platforms, allowing freshness patterns to be examined beyond a single generative search system.

Statistic 44

ChatGPT in-text reference URLs were approximately 393 days newer

Within the ChatGPT data examined by Ahrefs, URLs appearing as in-text references were approximately 393 days newer than the comparison Google organic content.

Statistic 45

ChatGPT citation URLs were approximately 458 days newer

URLs appearing within ChatGPT’s citation layer were approximately 458 days newer than the corresponding comparison content in Google organic search.

Statistic 46 • CGO Media Calculation

The ChatGPT citation freshness gap was 65 days greater than the in-text-reference gap

The difference between the reported 458-day citation gap and 393-day in-text-reference gap is approximately 65 days.

Statistic 47

36.7% of Google AI Overview citations came from URLs outside the organic top 100

Separate Ahrefs research found that approximately 36.7% of URLs cited in Google AI Overviews did not rank within Google’s first 100 organic results for the analysed query.

Statistic 48

26.2% of AI Overview citation URLs ranked between positions 11 and 100

A further 26.2% of cited URLs appeared between positions 11 and 100, reinforcing the finding that generative citation eligibility extends well beyond the conventional first page.

Statistic 49 • CGO Media Calculation

62.9% of AI Overview citations were sourced from outside the organic top ten

Combining the 26.2% ranking between positions 11 and 100 with the 36.7% outside the top 100 produces approximately 62.9% of citations originating beyond Google’s organic top ten.

Statistic 50

37.1% of AI Overview citation URLs ranked within Google’s organic top ten

Although many citations originated outside the first page, approximately 37.1% of Google AI Overview citation URLs did rank within Google’s organic top ten. This demonstrates that traditional organic visibility remains relevant even though it does not fully determine generative citation selection.

Central GEO Finding

Freshness appears to form part of the generative source-selection environment, but freshness should be treated as one eligibility signal among many — not as a shortcut that replaces relevance, evidence, authority or entity clarity.

A GEO Source Eligibility Model

Relevance
+
Evidence
+
Authority
+
Entity Clarity
+
Freshness
→
Citation Eligibility

A Better Content Update Cycle

For important research, service and commercial pages, a GEO-focused review process can examine:

• Whether statistics remain current.
• Whether claims still have supporting evidence.
• Whether newer primary sources are available.
• Whether examples reflect the current market.
• Whether products or services have changed.
• Whether entity information remains accurate.
• Whether external citations remain accessible.
• Whether conclusions need revision.
• Whether structured data reflects the current content.
• Whether the page’s review date represents a genuine substantive review.

Primary Evidence Sources — Statistics 41–50

Ahrefs — AI Citation Freshness Research
Analysis of approximately 17 million citations across seven AI platforms. Used for overall AI citation freshness and ChatGPT in-text-reference and citation URL recency comparisons.

Ahrefs — Google AI Overview Citation Analysis
Approximately 863,000 keyword SERPs and four million AI Overview citation URLs. Used here to provide context on how citation eligibility can extend beyond Google’s conventional organic top ten.

Statistic 51

ChatGPT outbound referral traffic increased by approximately 206% during 2025

Semrush analysis of more than one billion lines of US clickstream data found that outbound referral traffic from ChatGPT increased by approximately 206% during 2025.

Statistic 52

More than 30% of measured ChatGPT referral traffic went to just ten domains

The same clickstream research found substantial concentration in outbound traffic: more than 30% of measured ChatGPT referrals were directed to ten domains.

Statistic 53

More than 20% of measured ChatGPT outbound referral traffic went to Google

Within the Semrush dataset, more than one-fifth of measured ChatGPT outbound referral traffic was directed to Google. This is important because AI-assisted discovery and conventional search can form part of the same research journey rather than functioning as completely separate behaviours.

Statistic 54

ChatGPT used web search on approximately 34.5% of measured queries

By February 2026, Semrush observed ChatGPT’s web-search capability being used on approximately 34.5% of measured queries, demonstrating the growing role of live web retrieval within AI answer construction.

Statistic 55

Approximately 65% of measured ChatGPT activity was classified as search or search-like

Separate Ahrefs research classified approximately 65% of observed ChatGPT usage as search or search-like behaviour under its study methodology.

Statistic 56

ChatGPT accounted for approximately 0.21% of measured website traffic

In an Ahrefs analysis covering approximately 76,000 websites, ChatGPT was estimated to account for around 0.21% of measured website traffic.

Statistic 57

Google accounted for nearly 40% of traffic in the same website dataset

Within the same approximately 76,000-site dataset, Google accounted for close to 40% of measured website traffic, illustrating the continuing difference in traffic scale between conventional search and ChatGPT.

Statistic 58

Google delivered approximately 190× more measured website traffic than ChatGPT

Ahrefs estimated that Google delivered approximately 190 times more website traffic than ChatGPT across the analysed sites.

Statistic 59

Only 1% of users clicked a cited source when a Google AI summary appeared

Pew Research Center analysis of 68,879 Google searches made by 900 US adults found that users clicked a source cited within an AI summary in only approximately 1% of visits where an AI summary appeared.

Statistic 60

Traditional-result clicks fell from 15% to 8% when an AI summary appeared

Pew found that users clicked a traditional Google search result on approximately 8% of visits containing an AI summary, compared with approximately 15% of visits without one.

Central GEO Finding

AI referral traffic is growing quickly, but website clicks substantially understate the potential visibility occurring inside generative answers themselves.

AI Answer
→
Brand Exposure
→
Trust Formation
→
Comparison
→
Possible Click
→
Possible Conversion

The Emerging GEO Attribution Problem

AI Discovery
→
Brand Awareness
→
Google Search
→
Website Visit
→
Conversion

A last-click analytics system may attribute the conversion to Google even when AI initiated the discovery journey.

Do Not Use AI Referral Traffic as a GEO Score

A website receiving little AI referral traffic may still have substantial generative visibility. Equally, a website receiving AI traffic is not necessarily highly visible across the wider set of prompts relevant to its market.

Primary Evidence Sources — Statistics 51–60

Semrush — ChatGPT Search and Referral Behaviour Research, 2026
Analysis of more than one billion lines of US clickstream data. Used for ChatGPT referral growth, referral concentration, traffic to Google and web-search usage.

Ahrefs — ChatGPT vs Google Traffic Research, 2026
Analysis of approximately 76,000 websites. Used for estimated ChatGPT and Google website-traffic shares, relative traffic scale and search-like ChatGPT behaviour.

Pew Research Center — Google AI Summary Search Behaviour, 2025
Analysis of 68,879 searches from 900 US adults. Used for traditional-result click rates and clicks on sources cited within AI summaries.

Statistic 61

18% of analysed Google searches produced an AI summary

Pew Research Center found that approximately 18% of the 68,879 Google searches examined in its March 2025 browsing dataset produced an AI-generated summary.

Statistic 62

12,593 searches in the dataset produced an AI summary

Of the 68,879 unique Google searches included in the research, 12,593 generated an AI summary alongside the conventional search results.

Statistic 63

58% of participants encountered at least one Google AI summary

Approximately 58% of the 900 US adults included in the browsing study conducted at least one Google search during March 2025 that produced an AI-generated summary.

Statistic 64

26% of visits with an AI summary ended the browsing session

Users ended their browsing session after approximately 26% of Google search visits containing an AI summary.

Statistic 65

16% of visits without an AI summary ended the browsing session

Where only conventional search results were present, users ended their browsing session after approximately 16% of visits.

Statistic 66 • CGO Media Calculation

Session-ending behaviour was 62.5% higher when an AI summary appeared

Comparing Pew’s reported rates of 26% and 16%, CGO Media calculates that the relative rate of session-ending behaviour was approximately 62.5% higher on visits containing an AI summary.

Statistic 67

88% of Google AI summaries cited three or more sources

The vast majority of the AI summaries examined contained multiple supporting sources. Approximately 88% cited at least three sources.

Statistic 68

Only 1% of AI summaries cited a single source

At the opposite end of the distribution, only approximately 1% of the AI summaries analysed by Pew relied on a single cited source.

Statistic 69

53% of searches containing ten or more words produced an AI summary

Longer searches were substantially more likely to trigger generative results. Approximately 53% of searches containing ten words or more generated an AI summary.

Statistic 70

60% of question-word searches generated an AI summary

Approximately 60% of searches beginning with question words such as “who”, “what”, “when” or “why” generated an AI summary in the Pew dataset.

Central GEO Finding

Generative visibility becomes particularly important as searches become longer, more conversational and more question-based — precisely the types of queries where users are asking search systems to interpret, explain, compare and synthesise information.

“Which SEO agencies in London have published original research into AI search and can demonstrate expertise in both traditional SEO and GEO?”

First-Party Source
+
Independent Source
+
Research Evidence
+
Comparative Evidence
→
AI Answer

The GEO Implication

Organisations should identify the complex questions surrounding their products, services and expertise rather than relying exclusively on short commercial keywords.

The more a search requires explanation, synthesis or comparison, the greater the potential importance of being eligible for inclusion within the generative answer itself.

Primary Evidence Sources — Statistics 61–70

Pew Research Center — Google Users Are Less Likely to Click on Links When an AI Summary Appears, July 2025
Browsing analysis covering 900 US adults and 68,879 unique Google searches. Used for AI-summary prevalence, source counts, session-ending behaviour and query characteristics.

Pew Research Center — Americans and AI 2026
Survey conducted February 17–23, 2026. Used as additional current context showing that 60% of US adults reported reading AI summaries at the top of search-engine results.

Statistic 71

Wikipedia accounted for 7.8% of ChatGPT citations in one large-scale analysis

Ahrefs’ analysis of approximately 17 million citations across seven AI platforms found that Wikipedia represented approximately 7.8% of citations observed in ChatGPT.

Statistic 72

Reddit represented 1.8% of measured ChatGPT citations

Within the same ChatGPT citation dataset, approximately 1.8% of citations came from Reddit, demonstrating that community-generated discussion can form part of the source environment used by generative systems.

Statistic 73

YouTube accounted for approximately 1.4% of measured ChatGPT citations

Approximately 1.4% of the measured ChatGPT citations originated from YouTube, showing that AI source discovery is not restricted to conventional text publishers or corporate websites.

Statistic 74 • CGO Media Calculation

Wikipedia’s measured ChatGPT citation share was more than four times Reddit’s

Using the reported citation shares of 7.8% for Wikipedia and 1.8% for Reddit, Wikipedia’s share was approximately 4.3 times larger within this particular ChatGPT dataset.

Statistic 75 • CGO Media Calculation

Wikipedia’s citation share was approximately 5.6× YouTube’s

Comparing the reported 7.8% Wikipedia share with YouTube’s 1.4% gives a ratio of approximately 5.6 to 1.

Statistic 76

AI source-selection research analysed approximately 17 million citations

The Ahrefs dataset underpinning these source-distribution observations contained approximately 17 million citations, providing a broad base for examining which domains and source types repeatedly appeared across AI answers.

Statistic 77

The citation analysis covered seven AI platforms

The research compared citations across seven AI platforms, reinforcing an important GEO principle: source-selection behaviour should not be assumed to be identical across different generative systems.

Statistic 78

88% of Google AI summaries cited at least three sources

Pew Research Center found that approximately 88% of the Google AI summaries it analysed cited three or more sources. This provides further evidence that generative answers frequently operate within multi-source evidence environments.

Statistic 79

Only 1% of analysed Google AI summaries relied on a single source

Pew found that only approximately 1% of AI summaries cited a single source, meaning the overwhelming majority drew on multiple references rather than one isolated webpage.

Statistic 80 • CGO Media Calculation

AI summaries were at least 88× more likely to cite three or more sources than exactly one source

Using Pew’s reported 88% multi-source figure and 1% single-source figure, AI summaries were at least 88 times as likely to cite three or more sources as to cite exactly one source within the analysed dataset.

Central GEO Finding

Generative visibility is an ecosystem problem. An organisation’s own website can provide first-party evidence, but AI systems may combine that information with independent sources before forming an answer, comparison or recommendation.

The GEO Trust Layer

First-Party Evidence
+
Independent Evidence
+
Entity Consistency
+
Research
+
Reputation Signals
→
Recommendation Confidence

Organisation Claim
→
External Evidence
→
Evidence Convergence
→
Higher Confidence
→
Potential Recommendation

What Organisations Should Audit

A GEO authority audit should extend beyond the organisation’s own domain and examine:

• Which independent websites mention the organisation.
• Whether important third-party information is accurate.
• Whether the brand is associated consistently with its core services and expertise.
• Whether credible publications cite its research or experts.
• Whether review platforms contain sufficient current evidence.
• Whether relevant industry directories and databases contain accurate entity information.
• Whether authoritative sources corroborate important claims.
• Whether competitor comparison pages include the organisation.
• Whether important brand entities are consistently named across the web.
• Whether negative, outdated or contradictory information is creating entity ambiguity.

Primary Evidence Sources — Statistics 71–80

Ahrefs — AI Citation Source Research
Large-scale analysis covering approximately 17 million citations across seven AI platforms. Used for source-domain patterns and the measured ChatGPT citation shares associated with Wikipedia, Reddit and YouTube.

Pew Research Center — Google AI Summary Search Behaviour, 2025
Analysis of 68,879 Google searches from 900 US adults. Used for the distribution of single-source and multi-source citations within Google AI summaries.

Statistic 81

GEO methods improved generative-engine visibility by up to approximately 40%

The original GEO research reported that optimisation methods could improve source visibility within generative-engine responses by up to approximately 40% under the experimental conditions tested.

Statistic 82

The GEO benchmark contained 10,000 diverse queries

The researchers introduced GEO-bench, a benchmark containing 10,000 queries across multiple domains and sources, allowing optimisation methods to be evaluated across a much broader environment than a small hand-selected prompt set.

Statistic 83

Nine optimisation methods were evaluated

The experimental framework evaluated nine GEO optimisation methods, including approaches involving citations, quotations, statistics, authoritative language, fluency and readability.

Statistic 84

Citation Addition improved one visibility metric by approximately 40%

Among the tested methods, adding relevant citations produced one of the strongest results. The researchers reported an improvement of approximately 40% on the study’s position-adjusted word-count visibility metric.

Statistic 85

Quotation Addition improved the same visibility metric by approximately 30%

Adding relevant quotations from authoritative sources increased the study’s position-adjusted word-count visibility measure by approximately 30%.

Statistic 86

Statistics Addition improved visibility by approximately 30%

Adding quantitative statistics to relevant content produced an improvement of approximately 30% on the position-adjusted word-count metric used in the GEO experiment.

Statistic 87

Fluency optimisation improved visibility by approximately 15–30%

Improving the fluency and presentation of source material produced visibility improvements in the approximate 15% to 30% range across the experimental measures reported by the researchers.

Statistic 88

Simple keyword stuffing performed poorly in GEO experiments

The experiments found that simply increasing the frequency of query-related keywords did not produce the strong visibility gains observed for evidence-led methods such as citations, quotations and statistics.

Statistic 89

The effectiveness of GEO methods varied substantially by subject domain

The research found that no single optimisation technique performed equally well across every subject. Different methods produced different results depending on the query domain and information requirement.

Statistic 90 • CGO Media Interpretation

The strongest reported experimental uplift was roughly four times a 10% change

The reported maximum improvement of approximately 40% demonstrates that GEO interventions can produce material changes within an experimental generative-search environment. It should not, however, be interpreted as a promise that commercial websites will receive a 40% increase in ChatGPT, Gemini or Google AI visibility.

Central GEO Finding

The original GEO experiments provide evidence that generative visibility can be influenced by content characteristics — particularly evidence, citations, quotations, statistics and clear authoritative presentation — rather than being determined solely by conventional keyword matching.

From Content to Generative Visibility

Clear Information
+
Evidence
+
Citations
+
Statistics
+
Authority
→
Retrieval & Citation Potential

Practical GEO Content Principles

• Support important factual claims with credible sources.
• Use original statistics where genuine data exists.
• Cite primary research rather than repeatedly citing secondary summaries.
• Use quotations only where they contribute meaningful evidence.
• Make key facts easy to identify and understand.
• Separate evidence from unsupported promotional claims.
• Write clearly rather than artificially repeating keywords.
• Match the evidence format to the information need.
• Maintain source freshness where the subject changes quickly.
• Measure visibility across multiple prompts and platforms rather than assuming an optimisation worked.

Primary Evidence Source — Statistics 81–90

Aggarwal et al. — GEO: Generative Engine Optimization
Foundational research introducing Generative Engine Optimisation and GEO-bench, a benchmark of 10,000 queries. The study tested nine optimisation methods and reported visibility improvements of up to approximately 40% under its experimental framework.

Research reference: arXiv:2311.09735. The paper was subsequently presented in the KDD 2024 research programme.

Statistic 91

Google AI Overviews surpassed 2.5 billion monthly active users

Google reported in 2026 that AI Overviews had surpassed 2.5 billion monthly active users, making generative answers a mainstream component of global search behaviour.

Statistic 92

Google AI Mode surpassed one billion monthly active users

Only around one year after its debut, Google’s more conversational and reasoning-led AI Mode surpassed one billion monthly active users.

Statistic 93 • CGO Media Calculation

AI Overviews’ reported monthly user scale was approximately 2.5× that of AI Mode

Comparing Google’s reported figures of more than 2.5 billion monthly active AI Overview users and more than one billion AI Mode users gives a scale ratio of approximately 2.5 to 1.

Statistic 94

AI Mode queries more than doubled every quarter after launch

Google reported at I/O 2026 that AI Mode query volume had more than doubled every quarter since launch, indicating rapid growth in use of its conversational AI search environment.

Statistic 95

Google Search queries reached an all-time high in 2026

Google reported that overall Search queries had reached an all-time high, with the company attributing part of the increase to users asking new kinds of questions through AI-powered Search features.

Statistic 96

AI Overviews grew from more than 1.5 billion to more than 2.5 billion users

At Google I/O 2025, the company reported more than 1.5 billion AI Overview users. By I/O 2026, that figure had increased to more than 2.5 billion monthly active users.

Statistic 97 • CGO Media Calculation

Reported AI Overview user scale increased by at least one billion in roughly one year

The movement from more than 1.5 billion users in May 2025 to more than 2.5 billion monthly active users in May 2026 represents an increase of at least one billion users between the two reported milestones.

Statistic 98

Early AI Mode users submitted queries two to three times longer than traditional searches

Google reported in 2025 that early AI Mode testers were submitting queries approximately two to three times the length of traditional searches, providing further evidence of a move toward more detailed and conversational search behaviour.

Statistic 99

AI Overviews increased usage by more than 10% for relevant query types in major markets

Google reported in 2025 that in major markets including the United States and India, AI Overviews were driving more than a 10% increase in Google usage for the types of queries where AI Overviews appeared.

Statistic 100

Google says its AI Search features now send billions of clicks to websites every week

In its Q2 2026 remarks, Google stated that its AI-powered Search features were sending billions of clicks to websites every week. The company did not disclose an exact figure or provide a breakdown by AI feature, geography or publisher category.

The 100-Statistic GEO Finding

The evidence does not show that traditional search is disappearing. It shows that the search ecosystem is expanding: conventional rankings, AI-generated answers, citations, conversational search, source retrieval, brand mentions and recommendations are increasingly operating together.

The 2026 Search Visibility Model

Traditional Rankings
+
AI Overviews
+
AI Mode
+
AI Assistants
+
Citations
+
Brand Mentions
+
Recommendations
=
Search Ecosystem Visibility

CGO Media GEO Visibility Framework

Based on the evidence reviewed across these 100 statistics, GEO visibility can be considered through seven connected stages:

Entity Understanding
→
Retrieval Eligibility
→
Source Selection
→
Citation
→
Brand Recognition
→
Comparison
→
Recommendation

Conclusion

GEO Is Becoming Part of Search Visibility

The 100 statistics examined in this report show why Generative Engine Optimisation should not be treated as a replacement for SEO or as a short-lived marketing label.

Search systems are increasingly capable of retrieving information, comparing sources, synthesising evidence and constructing answers before the user visits a website. This changes where visibility occurs and how that visibility needs to be measured.

The organisations most prepared for this environment will not simply attempt to “rank in AI”. They will build clear, authoritative and evidence-rich information ecosystems that search engines and generative systems can understand and retrieve.

The objective is no longer visibility in one list of search results. It is visibility across the entire search and answer ecosystem.

Primary Evidence Sources — Statistics 91–100

Google — A New Era for AI Search, May 2026
Google reported that AI Mode had surpassed one billion monthly users, its query volume had more than doubled every quarter since launch and overall Search queries had reached an all-time high.

Google — New Opportunities, Control and Insights for Website Owners, June 2026
Google reported more than 2.5 billion monthly active users for AI Overviews and more than one billion monthly users for AI Mode.

Google I/O 2025
Google reported more than 1.5 billion AI Overview users, more than 10% growth in usage for relevant AI Overview query types in major markets, and AI Mode queries from early testers averaging two to three times the length of traditional searches.

Alphabet Q2 2026 Earnings Remarks
Google stated that AI-powered Search features were sending billions of clicks to websites each week. No exact weekly total or country-level distribution was disclosed.

CGO Media Evidence Standard

Statistics are included only where a numerical claim can be connected to an identifiable source, study or transparent CGO Media calculation. Qualitative observations are not counted as statistics. Where evidence comes from outside the UK, its geography and research context are identified.

1. Ofcom — Adults’ Media Use and Attitudes 2026

Organisation: Ofcom
Geography: United Kingdom
Used for: UK adult AI-tool adoption and adoption among younger age groups.

This is one of the principal UK-specific sources used in the report and provides nationally relevant evidence for the scale of AI-tool use among UK adults.

2. Ofcom — Online Nation 2025

Organisation: Ofcom
Geography: United Kingdom
Measurement partner: Ipsos iris
Used for: UK chatbot reach, ChatGPT reach and the size of the UK online AI audience.

3. Ofcom — The Era of Answer Engines

Organisation: Ofcom
Geography: United Kingdom
Used for: UK ChatGPT visit growth, changing search behaviour and contextual evidence concerning AI answer engines.

4. Semrush / Kevin Indig — AI Citations & Brand Mention Research

Research type: Multi-country AI visibility study
Sample: 3,981 domain appearances across 115 prompts in 14 countries
Platforms: ChatGPT, Google AI Overviews, Google AI Mode and Gemini
Used for: Ghost citations, citation rates, brand-mention rates and comparative-content visibility.

These findings measure a defined prompt and domain dataset and should not be interpreted as universal citation probabilities across all AI usage.

5. Ahrefs — AI Citations vs Google & Bing Rankings

Research type: AI citation / organic-ranking overlap study
Sample: Approximately 15,000 prompts
Used for: Google top-ten overlap and Bing top-ten overlap among AI-cited URLs.

6. Ahrefs — Google AI Overview Citation Analysis

Research type: Large-scale Google AI Overview analysis
Sample: Approximately 863,000 SERPs and four million citation URLs
Used for: AI Overview citations ranking in positions 1–10, positions 11–100 and outside Google’s top 100.

7. Ahrefs — ChatGPT Citation & Google Ranking Overlap

Research type: ChatGPT / Google overlap study
Sample: 3,311 short-tail terms in the referenced analysis
Used for: Exact-URL overlap and domain-level overlap between ChatGPT citations and Google’s organic top ten.

8. Semrush / Kevin Indig — Reasoning, Query Fan-Out & Source Selection

Research type: AI reasoning experiment
Sample: 100 prompts across 20 buyer journeys
Used for: Citation rates, average citations, query fan-out, web-search volume, domain overlap and high-reasoning-only source discovery.

The experiment compared minimal- and high-reasoning configurations. Its findings demonstrate how reasoning depth can change retrieval behaviour within the tested environment.

9. Ahrefs — AI Citation Freshness Research

Research type: AI citation freshness analysis
Sample: Approximately 17 million citations
Platforms: Seven AI platforms
Used for: Relative freshness of AI-cited content, ChatGPT in-text references and citation URL recency.

10. Semrush — ChatGPT Search & Referral Behaviour Research

Research type: Clickstream analysis
Geography: United States
Dataset: More than one billion lines of clickstream data
Used for: ChatGPT referral growth, outbound referral concentration, referrals to Google and ChatGPT web-search usage.

11. Ahrefs — ChatGPT vs Google Website Traffic Research

Research type: Website traffic analysis
Sample: Approximately 76,000 websites
Used for: ChatGPT traffic share, Google traffic share, relative traffic scale and the proportion of ChatGPT activity classified as search or search-like.

12. Pew Research Center — Google Users Are Less Likely to Click on Links When an AI Summary Appears

Publication: July 2025
Geography: United States
Participants: 900 US adults
Searches analysed: 68,879 unique Google searches
Used for: AI-summary prevalence, clicks on conventional results, clicks on cited sources, session-ending behaviour, number of cited sources and query characteristics associated with AI summaries.

13. Pew Research Center — Americans and AI 2026

Geography: United States
Survey period: February 17–23, 2026
Used for: Additional context on reported consumption of AI summaries within search-engine results.

This survey uses a different methodology from Pew’s 2025 browsing study and the two datasets are not treated as directly comparable measures.

14. Ahrefs — AI Citation Source Distribution Research

Dataset: Approximately 17 million citations across seven AI platforms
Used for: Source-domain patterns and citation shares associated with sources including Wikipedia, Reddit and YouTube.

Citation shares are dataset-specific and should not be interpreted as fixed citation probabilities for every topic, country or AI platform.

15. Aggarwal et al. — GEO: Generative Engine Optimization

Research type: Academic experimental research
Reference: arXiv:2311.09735
Conference: KDD 2024
Benchmark: GEO-bench — 10,000 queries
Used for: Experimental GEO visibility improvements and comparison of optimisation methods including citation addition, quotation addition and statistics addition.

The reported “up to approximately 40%” visibility improvement relates to the researchers’ experimental environment and metrics. It is not presented by CGO Media as a guaranteed commercial visibility increase.

16. Google — Google I/O 2025 Search & AI Announcements

Organisation: Google
Publication period: May 2025
Used for: AI Overview user scale, reported growth in usage for AI Overview query types and early AI Mode query-length behaviour.

Google’s usage figures are first-party company disclosures and are identified as such rather than independent audience measurements.

17. Google — AI Search Announcements, 2026

Organisation: Google
Publication period: 2026
Used for: More than 2.5 billion monthly active AI Overview users, more than one billion AI Mode users, AI Mode query growth and Google’s statement that Search queries had reached an all-time high.

18. Alphabet — Q2 2026 Earnings Remarks

Organisation: Alphabet / Google
Used for: Google’s statement that AI-powered Search features were sending billions of clicks to websites each week.

Google did not provide a precise weekly total or a country-, publisher- or AI-feature-level distribution for this statement.

UK statistics are used where reliable UK measurement exists.

US statistics are explicitly identified as US evidence.

Multi-country studies are labelled as multi-country evidence.

Global platform figures are not presented as UK market-share statistics.

Vendor datasets are treated as research evidence from the defined sample, not universal population estimates.

Important Research Limitation

Generative search is changing rapidly. ChatGPT, Google AI Overviews, Google AI Mode, Gemini and other systems can alter their retrieval, citation and answer-generation behaviour without the underlying research datasets being immediately repeated.

Statistics should therefore be interpreted according to their study date, geography, platform, sample and methodology, rather than treated as permanent rules governing AI visibility.

Suggested citation:
CGO Media Research Team (2026), GEO Statistics UK 2026: 100 Generative Engine Optimisation Statistics, CGO Media.

Research Methodology

For more information about how CGO Media evaluates sources, separates research findings from interpretation and develops its search and AI research programme, visit the CGO Media Research Methodology.

View Research Methodology