Research Data & Evidence Standards
Establishing transparent principles for research data collection, source selection, evidence verification, provenance, data integrity and reproducibility across SEO, generative search and artificial intelligence research.
Executive Summary
The credibility of research depends upon the quality, relevance and traceability of the evidence supporting its conclusions. Clearly defined research questions are necessary, but they cannot compensate for unreliable sources, inconsistent measurements or undocumented analytical decisions.
Research into artificial intelligence, search engine optimisation, generative engine optimisation and digital discovery presents particular evidence challenges. Search systems are continuously modified, generative outputs may vary between observations, and researchers often lack direct access to proprietary retrieval processes, training information and ranking mechanisms.
CGO Media's approach to research evidence emphasises source suitability, transparent attribution, appropriate verification, documented limitations and careful separation between observed findings and proposed explanations. These principles apply to original observations, external statistics, technical documentation, literature synthesis and conceptual research.
This document establishes a methodological reference for assessing evidence before it is incorporated into research publications. It explains how different source types should be evaluated, how data quality affects interpretation and why reproducibility and provenance matter to research credibility.
It forms the second component of the CGO Media Research Methodology, following Research Scope & Questions, and operates alongside the Ethics & Integrity principles.
Core Principles of Research Evidence
Evidence standards provide a basis for deciding whether information is sufficiently relevant, credible and appropriately documented to support a particular research claim. The standards should be proportionate to the significance of the claim and the consequences of an incorrect interpretation.
Source Relevance
Evidence should address the actual research question rather than merely discuss a related topic.
Traceability
Material claims should be attributable to identifiable sources, datasets or documented observations wherever reasonably possible.
Verification
Statistics, quotations, technical claims and source descriptions should be checked against appropriate evidence.
Data Integrity
Data should not be fabricated, selectively altered or presented in a way that materially misrepresents the underlying observations.
Methodological Context
Evidence should be interpreted with reference to its collection method, population, time period and known limitations.
Proportionate Interpretation
Conclusions should reflect the strength and limitations of the evidence rather than the apparent authority of its presentation.
1. Establishing Research Evidence Requirements
1.1 Evidence Begins with the Research Question
Research evidence should be selected in response to a defined question. A dataset may be extensive, technically sophisticated or widely cited without being suitable for the particular proposition under investigation.
The first evidence decision therefore concerns relevance. Researchers should establish what observations would meaningfully answer the question, which information is necessary to interpret those observations and which sources may introduce uncertainty.
For example, an investigation into AI citation frequency requires evidence concerning actual citation occurrences under specified conditions. General website traffic statistics or published commentary about AI adoption may provide context, but they cannot substitute for observations of citation behaviour.
The relationship between research questions and evidence requirements is examined in Research Scope & Questions. That relationship should remain explicit throughout evidence collection and analysis.
1.2 Defining the Evidence Population
Evidence populations describe the wider set of observations to which a study intends to refer. They may consist of search queries, generated answers, documents, websites, organisations, technical events or published studies.
The accessible dataset may represent only part of the intended population. This distinction matters because research based on a limited selection of observations may not support conclusions about a broader market, platform or user population.
Where evidence is obtained from a specific geographic region, language, industry or collection period, those conditions should be considered when interpreting the findings. The size of a dataset does not eliminate bias arising from an inappropriate selection process.
1.3 Evidence Sufficiency
Evidence sufficiency concerns whether the available material can reasonably support the proposed conclusion. It is not determined solely by the number of references or observations included in a publication.
Several sources repeating the same unsupported claim do not necessarily provide independent confirmation. Likewise, a large number of observations collected through a systematically biased procedure may provide less reliable evidence than a smaller, carefully defined dataset.
Researchers should consider the relevance, independence, completeness and quality of the available evidence. Where those characteristics are inadequate, the conclusion should be narrowed or presented as provisional.
1.4 Evidence and Research Output Types
Different publications require different forms of supporting evidence. Empirical studies depend on documented observations or measurements. Literature reviews require traceable publications and appropriate selection criteria. Conceptual frameworks may draw upon established theory, technical documentation and structured analytical reasoning.
The evidential requirements of a statistical estimate are not identical to those of a conceptual model. A framework may offer a useful interpretation of a system without demonstrating that every proposed relationship has been independently validated.
CGO Media's Research Architecture distinguishes complementary research outputs, while the Research Library provides access to its wider research programme.
Research principle: The strength of a research conclusion depends on the suitability and quality of the supporting evidence, not simply on the quantity of material cited.
2. Research Source Selection and Classification
2.1 Primary Research Sources
Primary sources provide direct evidence relevant to the question being investigated. Depending on the research context, these may include original datasets, documented platform behaviour, official statistics, technical specifications, experimental observations and first-hand records.
Primary sources can offer important advantages because they reduce dependence on another author's interpretation. However, primary status does not guarantee reliability. A first-hand observation may still be incomplete, biased, incorrectly measured or unrepresentative.
Source evaluation should therefore examine both proximity to the underlying phenomenon and the quality of the process through which evidence was produced.
2.2 Secondary Research Sources
Secondary sources interpret, summarise or analyse evidence originating elsewhere. Examples include research reviews, industry reports, academic commentary and technical analyses.
Such sources can provide valuable context, especially when they synthesise a complex body of research. Their usefulness depends upon the transparency of their methods, the quality of the underlying evidence and the accuracy of their interpretation.
Where a secondary source reports a significant statistic, researchers should attempt to identify the original source and examine its methodological context when reasonably possible.
2.3 Academic and Scholarly Publications
Academic literature can contribute theoretical foundations, established methods and empirical findings. Relevant considerations include the research design, publication venue, peer-review status, data availability, limitations and relationship between findings and conclusions.
Peer review is a useful quality-control process but does not establish that every published claim is correct. Similarly, preprints and repository publications may contain valuable research without having undergone independent peer review.
Researchers should distinguish between publication, repository deposit, editorial assessment and external scholarly review rather than treating these processes as equivalent.
2.4 Government and Official Statistical Sources
Government departments, national statistical agencies and public institutions may publish data relevant to technology adoption, business activity, economic behaviour and digital infrastructure.
Official status can provide useful assurance about institutional responsibility, but the suitability of a dataset still depends on its definitions, collection method, population, reference period and revisions.
Statistics collected for administrative purposes may not directly measure the behaviour a search researcher intends to investigate. The original definitions should therefore be examined before the data is repurposed.
2.5 Platform and Technical Documentation
Official platform documentation may be particularly relevant to technical SEO, crawling, indexing, structured data, APIs and publicly documented search functionality.
However, platform documentation may describe intended or supported functionality rather than the full behaviour of a deployed system. It may also change without preserving all historical details.
Researchers should distinguish documented technical capabilities from independently observed outcomes. For example, a statement that a platform supports a particular markup type does not establish that using the markup guarantees visibility or citations.
2.6 Commercial and Industry Research
Industry reports can provide access to large datasets, specialist surveys and operational knowledge that may not be available through academic publications.
These reports should be assessed for sample construction, disclosed methodology, funding relationships, definitions, missing data and the distinction between analysis and marketing claims.
A commercial relationship does not automatically invalidate evidence. Nevertheless, material incentives and conflicts should be considered when evaluating interpretation, consistent with CGO Media's Ethics & Integrity standards.
3. Source Quality, Credibility and Verification
3.1 Assessing Source Credibility
Source credibility should be evaluated through multiple considerations rather than a single institutional label or reputation indicator.
Relevant factors may include identifiable authorship, methodological transparency, technical expertise, evidence traceability, publication history, conflicts of interest and the opportunity for independent scrutiny.
A source may be credible in one field but poorly suited to another. For example, a provider's technical documentation may be authoritative concerning its published API requirements while offering limited independent evidence about market-wide business outcomes.
3.2 Verification of Factual Claims
Material factual claims should be checked against evidence appropriate to their significance. Verification may involve consulting the original publication, examining a technical specification, reviewing a dataset or comparing independently generated observations.
Quotations require particular care because inaccurate wording or missing context can materially change meaning. Numerical claims require confirmation of the figure, unit, population and reference period.
Where a claim cannot be verified, it should not be presented as established fact merely because it appears in a widely circulated article or AI-generated summary.
3.3 Independent Corroboration
Independent corroboration involves comparing a proposition with evidence that does not simply repeat the same underlying source.
Three publications citing one original estimate do not represent three independent measurements. Researchers should attempt to trace apparent agreement back to the underlying data or observations.
Where different evidence sources produce conflicting results, the disagreement should be investigated. Differences in populations, definitions, timing and collection methods may explain why apparently similar studies reach different conclusions.
3.4 Evaluating Source Currency
Research concerning AI search and digital platforms is particularly sensitive to publication dates and collection periods. A technically accurate statement about an earlier interface may no longer describe current behaviour.
Researchers should distinguish the date a source was published from the period during which its evidence was collected. A recently published report may analyse historical observations, while an older theoretical paper may remain relevant to contemporary research design.
Currency should therefore be assessed according to the research question rather than through an arbitrary preference for the newest publication.
3.5 Avoiding Circular Citation
Circular citation occurs when claims gain apparent credibility through repeated references that ultimately depend on one another or on an unsupported original assertion.
This problem is especially relevant in fast-moving technology markets, where reports, articles and AI-generated summaries may repeat statistics without identifying the primary evidence.
Where feasible, source checking should follow the citation chain to the originating study. If the original evidence cannot be located, the limitation should be acknowledged rather than concealed.
3.6 Distinguishing Evidence from Interpretation
Research sources frequently combine direct observations with analysis, opinion and predictions. These elements should not be treated as having equal evidential status.
An observed change in citation frequency is evidence of a measured difference under specified conditions. A proposed explanation involving entity authority or source trust is an interpretation requiring separate evaluation.
The companion Research Analysis Methods document addresses the relationship between observations, analytical inference and research conclusions.
4. Research Data Collection and Documentation
4.1 Defining Data Collection Procedures
Data collection procedures should correspond to the research question, population and intended analytical methods. Where appropriate, a research plan should identify what is collected, when collection occurs, how observations are recorded and which conditions may affect the results.
For AI search research, relevant conditions may include platform, model or interface version where known, query wording, language, geography, session settings and collection timestamp.
Documenting these conditions does not guarantee that another researcher can reproduce an identical output, but it improves the interpretability of the evidence.
4.2 Structured Data Collection
Structured collection uses predefined fields and rules to record observations consistently. These may include query identifiers, response categories, citation presence, source domains, timestamps and coding notes.
Clear field definitions reduce ambiguity and help prevent researchers from applying different interpretations to similar observations.
When a field is added or its definition changes during collection, the potential effect on comparability should be considered and documented.
4.3 Qualitative Evidence Collection
Qualitative research may involve detailed examination of documents, generated answers, platform behaviour, professional interviews or organisational processes.
Such evidence can support interpretation of complex mechanisms and emerging patterns, but it also requires careful documentation of selection criteria, context and analytical judgement.
Illustrative examples should be identified as examples rather than presented as representative of an entire population without supporting evidence.
4.4 Data Cleaning and Transformation
Research data may require cleaning to correct formatting problems, identify duplicates, standardise categories or address missing values.
These procedures should preserve the meaning of the underlying observations. Data should not be excluded merely because it contradicts a preferred hypothesis or weakens an anticipated result.
Where exclusions or transformations materially affect findings, researchers should explain their purpose and consider whether alternative reasonable treatments produce different conclusions.
4.5 Missing Data
Missing observations may arise from technical failures, unavailable records, platform restrictions, incomplete responses or inconsistent source reporting.
The effect of missing data depends on why information is absent. If missingness is related to the phenomenon being studied, simply removing incomplete records may distort results.
Researchers should distinguish genuinely absent observations from values that are zero, unknown, not applicable or unavailable. These categories should not be merged without justification.
4.6 Data Collection Ethics
Research involving identifiable people, confidential organisational information or personal data should account for applicable legal and ethical requirements.
Public availability does not automatically make every form of collection, combination or republication appropriate. Data minimisation, legitimate purpose, confidentiality and proportionate disclosure may all be relevant.
These responsibilities connect evidence standards with CGO Media's wider research integrity principles.
5. Evidence Provenance, Attribution and Traceability
5.1 Understanding Evidence Provenance
Evidence provenance describes the origin and history of information used within a research publication. It includes where the evidence came from, how it was obtained and which transformations or interpretations occurred before publication.
Provenance matters because research findings may become difficult to evaluate when the connection between a reported claim and its original evidence is lost.
A clear provenance record allows readers to distinguish direct measurements from secondary estimates, derived calculations and interpretive statements.
5.2 Citation and Source Attribution
Research should acknowledge external ideas, findings, data and distinctive analytical contributions where appropriate.
Citations should identify sources sufficiently clearly to support verification. A link to a general homepage may be inadequate when a specific report, dataset or technical document supports the claim.
Attribution also requires accurate representation of the cited source. A citation should not be used to imply that an external author endorses a conclusion that the source does not actually support.
5.3 Version Identification
Research sources and datasets may be updated, corrected or replaced. Where version differences materially affect interpretation, the version or access context should be identified.
Persistent identifiers such as DOIs can support reliable source identification, but their presence does not establish methodological quality or independent validation.
Similarly, a repository deposit may provide useful preservation and discoverability without implying that the deposited work has undergone external peer review.
5.4 Research Data Lineage
Data lineage concerns the transformations through which source information becomes an analytical result.
A reported percentage, for example, may depend on an original dataset, a filtered population, a coding decision, a denominator definition and a calculation procedure.
Where these steps materially affect the outcome, documenting them can help readers understand how the result was produced and assess whether an alternative treatment would change the conclusion.
5.5 Responsible Reuse of External Evidence
External evidence should be reused consistently with relevant copyright, licensing, confidentiality and attribution requirements.
Permission to access information does not necessarily provide permission to reproduce an entire dataset, figure or publication.
Researchers should distinguish between citing findings, summarising evidence, reproducing copyrighted material and distributing underlying data. Appropriate permissions or licences may be required for some forms of reuse.
5.6 Traceability Across CGO Media Publications
CGO Media's research programme includes research papers, statistical publications, observations and analytical frameworks. Related publications may use overlapping concepts or draw upon common source material.
Where findings are reused, the relationship between the original evidence and the later interpretation should remain clear.
The Research Architecture provides an organisational structure for these outputs, while the Research Library serves as a starting point for examining individual publications.
6. Data Integrity, Reliability and Reproducibility
6.1 Data Integrity
Data integrity requires that research evidence be represented accurately and without deliberate fabrication, falsification or misleading selective treatment.
Integrity applies throughout the research process, including collection, coding, transformation, analysis, visualisation and publication.
A result may be mathematically correct yet misleading if the underlying population is incorrectly described or important exclusions are omitted.
6.2 Measurement Reliability
Reliability concerns the consistency of a measurement process under comparable conditions.
In search research, repeated measurements may differ because of personalisation, interface changes, retrieval variability, generative randomness or differences in collection settings.
Researchers should consider whether observed differences represent meaningful changes in the phenomenon or variation introduced by the measurement process itself.
6.3 Reproducibility and Replicability
Reproducibility generally concerns obtaining consistent results using the same data and analytical procedures. Replicability concerns whether a comparable investigation using new evidence produces consistent findings.
These concepts are related but distinct. A documented calculation may be reproducible even when the original AI-generated response cannot be recreated exactly.
Conversely, similar findings across independent collections may strengthen confidence in a pattern without demonstrating that every internal platform mechanism is understood.
6.4 Challenges in Generative AI Research
Generative systems may produce different outputs in response to similar or identical prompts. Retrieval sources, ranking decisions, model updates and presentation formats may also change.
Consequently, perfect output-level repeatability may be unrealistic in some settings. Researchers should document the conditions of observation and distinguish limitations inherent in the system from weaknesses in the research design.
Repeated sampling, stable coding definitions and sensitivity analysis may improve understanding of variability where these approaches are feasible.
6.5 Quality Assurance
Research quality assurance may include checking source references, reviewing data definitions, identifying duplicates, examining calculations and assessing consistency between evidence and conclusions.
Different checks address different risks. Correct arithmetic cannot validate a biased sample, and accurate citations cannot compensate for an unsupported causal interpretation.
Quality assurance should therefore consider the complete chain from research question to publication rather than treating proofreading as equivalent to methodological review.
6.6 Reporting Uncertainty
Uncertainty may arise from sampling variability, measurement error, missing evidence, ambiguous classification, inaccessible platform processes or rapidly changing technical conditions.
These uncertainties should be communicated in language proportionate to the available evidence. Where appropriate, quantitative studies may report intervals or sensitivity results, while qualitative studies may explain interpretive alternatives and contextual limits.
The Research Limitations & Transparency companion document addresses these issues as part of research governance.
Research principle: Reproducibility is strengthened by clear documentation, but a reproducible procedure is not automatically a valid measurement of the intended phenomenon.
7. Evidence Standards for AI Search, SEO and GEO Research
7.1 Search Engine Evidence
SEO research may draw upon search result observations, crawling data, indexing information, technical documentation, website performance measurements and published search guidance.
These evidence types should not be treated as interchangeable. A crawler's ability to access a page does not prove that a search engine has indexed it, while indexing does not guarantee ranking or traffic.
Researchers should distinguish technical eligibility, observed search visibility and subsequent user behaviour.
7.2 AI Citation Evidence
AI citation research examines how sources are attributed or referenced within specified generative interfaces.
Evidence collection should define what qualifies as a citation, whether links and unlinked references are treated differently, how duplicate sources are handled and which response conditions are included.
Citation occurrence should not automatically be interpreted as endorsement, factual reliability or proof that the cited page directly influenced every part of an answer.
Related research themes are represented within CGO Media's Research Library.
7.3 Entity and Brand Evidence
Entity and brand research may investigate organisational identification, naming consistency, source representation, knowledge graph relationships and generated recommendations.
Observable associations between these characteristics and AI visibility can support research questions, but they do not independently establish a platform's internal weighting of authority signals.
Conceptual approaches to these questions are presented within the CGO Entity Authority Framework and CGO Brand Signal Framework.
7.4 GEO and Generative Discovery Evidence
Generative engine optimisation research may examine source accessibility, content structure, citation occurrence, brand representation and the relationship between traditional search visibility and AI-mediated discovery.
These measurements describe different stages or characteristics of discovery. They should not be combined into a single claim of effectiveness without an explicit analytical model and appropriate supporting evidence.
Researchers should also distinguish observed platform behaviour from proposed optimisation mechanisms. A correlation between a content characteristic and citation occurrence does not necessarily show that changing the characteristic will produce more citations.
7.5 Sector-Specific Evidence
Evidence standards must account for differences between industries. Healthcare, financial services, legal information, ecommerce and travel research may involve distinct regulatory, commercial and informational contexts.
A source suitable for examining consumer travel discovery may not provide an appropriate basis for conclusions about medical information trust or financial decision-making.
CGO Media's Sector Research programme provides the broader organisational context for industry-specific analysis.
7.6 Statistics and Research Observations
Statistical research and observational analysis serve related but different purposes. A statistical publication may synthesise numerical evidence across defined populations, while an observation paper may describe an emerging pattern without claiming representative measurement.
The distinction should remain clear in titles, methodological descriptions and conclusions.
CGO Media maintains separate collections for the Statistics Library and Research Observations Library.
8. Evidence Governance and Responsible Publication
8.1 Linking Evidence to Published Claims
Every substantive research conclusion should remain connected to evidence that is relevant to the claim being made.
Where a publication combines direct observations, external literature and conceptual interpretation, the distinctions between these elements should remain understandable to the reader.
Research credibility depends not merely on providing references, but on ensuring that the cited material supports the specific propositions associated with it.
8.2 Avoiding Evidence Overstatement
Evidence should not be described as stronger, more representative or more conclusive than its methodology permits.
Examples of overstatement include presenting correlation as causation, treating a convenience sample as a population estimate, describing a conceptual model as experimentally validated or interpreting a platform statement as proof of a universal ranking mechanism.
Appropriate qualifications should be included where uncertainty materially affects interpretation.
8.3 Artificial Intelligence as a Research Tool
Artificial intelligence tools may assist with organising information, identifying themes, drafting, coding support and preliminary analytical tasks.
However, AI-generated content should not automatically be treated as verified evidence. Generated references, quotations, statistics and factual claims require checking against appropriate sources before they are relied upon in a publication.
Responsibility for published findings remains with the named researcher or research team, consistent with CGO Media's Ethics & Integrity policy.
8.4 Evidence Review and Corrections
Research evidence may require reassessment when a source is corrected, a dataset is revised, a technical platform changes or a methodological error becomes apparent.
Material corrections should be evaluated according to their effect on the original interpretation. A revised statistic may require a limited amendment, while a fundamental error in source selection may affect the validity of the broader conclusion.
Research should remain open to legitimate challenge and evidence-based correction after publication.
8.5 Transparency About Evidence Limitations
Readers should be able to identify material limitations that affect a publication's conclusions. These may include restricted access to data, incomplete sampling frames, uncertain measurement definitions, proprietary systems or reliance on secondary reporting.
Not every limitation invalidates a study. The relevant question is whether the limitation changes what the evidence can reasonably establish.
The companion Research Limitations & Transparency document provides a dedicated framework for discussing uncertainty and disclosure.
8.6 Integrating Evidence Standards into Research Architecture
Evidence standards support the relationship between CGO Media's research papers, statistics, observations and analytical frameworks.
Individual publications may employ different methodologies, but their claims should remain proportionate to the evidence used and their source relationships should be understandable.
The Research Architecture explains how these publication types connect, while Research Analysis Methods addresses the subsequent interpretation of collected evidence.
8.7 Research Accountability
Evidence standards require continuing attention because sources, technologies and research questions change.
The purpose of research governance is not to suggest that every finding is final. It is to support accurate reporting, transparent limitations, responsible attribution and willingness to revise conclusions when justified.
For CGO Media, these principles provide a methodological foundation for research into the evolving relationship between search systems, organisations, information sources and AI-mediated discovery.
Central methodological principle: Research evidence must be relevant, traceable and interpreted within its methodological limits. No source should be treated as conclusive solely because of its reputation, publication format or technical presentation.
Practical Evidence Assessment Workflow
The following sequence provides a recommended approach to assessing research evidence. It describes methodological guidance rather than asserting that every previous CGO Media publication has undergone an identical documented process.
01. Define the Claim
Identify the exact proposition that requires evidence and the level of certainty being considered.
02. Identify Sources
Locate relevant primary evidence and suitable secondary material, recording important source relationships.
03. Assess Quality
Examine methods, definitions, sample selection, publication context, incentives and known limitations.
04. Verify Material Claims
Check statistics, quotations, technical statements and citations against appropriate original evidence.
05. Document Provenance
Record the source, relevant version, collection context and material data transformations.
06. Review Interpretation
Assess whether the proposed conclusion is proportionate to the evidence and its uncertainties.
Explore the Research Methodology Collection
This document forms part of CGO Media's five-part research methodology collection. Each companion document addresses a different stage of research design, analysis and publication.
Conclusion
Reliable research depends upon evidence that is relevant, traceable and appropriate to the questions being investigated. Source credibility, verification, provenance and data integrity all contribute to the quality of research interpretation.
These requirements are particularly important in AI search and generative discovery, where rapidly changing systems, proprietary technologies and incomplete observational access create substantial methodological challenges.
CGO Media's evidence standards emphasise responsible source selection, transparent limitations and conclusions proportionate to the available information. Together with clearly defined research questions and appropriate analytical methods, these principles provide a foundation for credible research publication.
References and Methodological Foundations
The following publications and standards provide established background for data management, evidence assessment, reproducibility and responsible research reporting. Their inclusion does not imply that every procedure they describe has been implemented in every CGO Media publication.
- National Academies of Sciences, Engineering, and Medicine (2019). Reproducibility and Replicability in Science. National Academies Press. https://doi.org/10.17226/25303
- Wilkinson, M. D. et al. (2016). The FAIR Guiding Principles for scientific data management and stewardship. Scientific Data, 3, 160018. https://doi.org/10.1038/sdata.2016.18
- Munafò, M. R. et al. (2017). A manifesto for reproducible science. Nature Human Behaviour, 1, 0021. https://doi.org/10.1038/s41562-016-0021
- Page, M. J. et al. (2021). The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ, 372, n71. https://doi.org/10.1136/bmj.n71
- von Elm, E. et al. (2007). The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement. PLoS Medicine, 4(10), e296. https://doi.org/10.1371/journal.pmed.0040296
- Committee on Publication Ethics. COPE Core Practices. https://publicationethics.org/core-practices
Research Data & Evidence Standards: 24 Frequently Asked Questions
These questions explain how research evidence should be selected, verified, documented and interpreted within the CGO Media methodology.
1. What are research evidence standards?
Research evidence standards are principles used to assess whether information is relevant, credible, traceable and suitable for supporting a research conclusion. They concern source selection, verification, data integrity and the relationship between evidence and interpretation. They complement the wider Research Methodology.
2. How does CGO Media evaluate research sources?
Source evaluation considers relevance to the research question, identifiable origin, methodological transparency, evidence quality, collection period and material limitations. Institutional reputation alone is not sufficient to establish that a source supports a particular claim.
3. What is the difference between primary and secondary evidence?
Primary evidence provides direct information about the phenomenon being studied, while secondary evidence interprets or summarises material originating elsewhere. Both can be useful, but their limitations differ and should be considered when constructing research conclusions.
4. Why is source verification important in AI search research?
AI search research frequently involves rapidly changing platforms and widely repeated technical claims. Verification helps distinguish documented platform behaviour, original observations and unsupported interpretations. Material statements should be checked against suitable sources wherever feasible.
5. Does a peer-reviewed publication automatically provide reliable evidence?
No. Peer review can improve quality control, but individual studies may still contain methodological limitations, errors or findings that do not generalise to a new context. Researchers should examine the study's design, evidence and conclusions rather than relying on publication status alone.
6. How should third-party statistics be used?
Third-party statistics should be interpreted in relation to their original population, definitions, methodology and collection period. Where possible, the originating publication should be consulted. The CGO Media Statistics Library provides a dedicated context for statistical research.
7. What is evidence provenance?
Evidence provenance describes where information originated, how it was collected and which transformations or interpretations occurred before publication. It helps readers trace a claim back to its underlying evidence and evaluate the reliability of the analytical process.
8. What is the difference between reproducibility and replicability?
Reproducibility generally concerns obtaining consistent results from the same data and analytical procedures. Replicability concerns obtaining consistent findings through a comparable investigation using new evidence. Both support research credibility, although closed AI systems may limit exact repetition.
9. Can AI-generated content be used as research evidence?
AI-generated outputs can be the subject of direct observation, but factual statements produced by AI should not automatically be accepted as verified evidence. References, statistics and material claims require appropriate checking. CGO Media's Ethics & Integrity policy addresses responsible AI use.
10. How should conflicting research evidence be handled?
Conflicting evidence should be examined for differences in research populations, definitions, timing, methods and measurement quality. Researchers should not automatically select the result that supports their preferred explanation. Unresolved disagreements should be reflected in the conclusion.
11. What makes a research dataset reliable?
A reliable dataset depends on consistent collection and measurement procedures, appropriate definitions and suitable quality controls. Reliability alone does not establish validity; a dataset may be consistently measured yet fail to represent the concept under investigation.
12. Why are original sources preferable when available?
Original sources may provide direct access to definitions, methodology and evidence that secondary summaries omit. However, primary sources must still be evaluated for quality and relevance. The most appropriate source depends on the specific research question.
13. How should missing data be treated?
Missing data should be identified and assessed according to its cause and potential effect on findings. Unknown, unavailable, not applicable and zero values should not be treated as equivalent without justification. Material exclusions should be documented.
14. What is data cleaning?
Data cleaning involves identifying and addressing problems such as duplicates, inconsistent formatting, invalid entries or incomplete records. Cleaning should preserve the meaning of observations and should not be used to remove inconvenient evidence.
15. Why is the collection period important?
Evidence may become less representative as platforms, markets and technologies change. Recording the collection period helps readers determine which conditions the observations describe and whether subsequent changes may affect their interpretation.
16. What is independent corroboration?
Independent corroboration involves comparing a claim with evidence that does not merely repeat the same original source. Multiple publications citing one estimate should not be counted as independent measurements of the phenomenon.
17. Can commercial industry reports be used in research?
Yes, when they provide relevant evidence and their methodology, limitations and commercial context are appropriately assessed. Sponsorship or commercial authorship does not automatically invalidate evidence, but material conflicts and incentives should be considered.
18. How do evidence standards apply to research observations?
Observational research should identify what was observed, under which conditions and with what limitations. Emerging patterns should not be presented as representative population statistics without an appropriate design. See the Research Observations Library.
19. How do evidence standards support CGO Media frameworks?
Evidence standards help distinguish established findings, conceptual assumptions and relationships requiring further investigation. Frameworks should not be described as empirically validated solely because they organise concepts coherently. The Research Frameworks & Models methodology page examines this distinction.
20. What is the difference between a citation and evidence of causation?
A citation identifies a source associated with a statement or generated answer. It does not itself establish a causal mechanism. Demonstrating causation requires a suitable research design and consideration of competing explanations.
21. Why should evidence limitations be disclosed?
Disclosure allows readers to understand the conditions under which findings may apply. Restricted datasets, sampling bias, ambiguous measures and platform changes can all affect interpretation. The Research Limitations & Transparency document addresses these responsibilities.
22. How are research data and analytical methods connected?
Data quality determines which analyses are defensible, while analytical methods determine how evidence is compared and interpreted. Researchers should select methods suited to the evidence rather than forcing available data into an inappropriate model. See Research Analysis Methods.
23. Where can readers find CGO Media's published research?
Readers can explore the Research Library, Statistics Library, Research Observations Library and Sector Research collections.
24. Who is responsible for the Research Data & Evidence Standards document?
This document is attributed to the CGO Media Research Team. The team is responsible for the published document and for addressing material inaccuracies where identified. Authorship does not imply independent external peer review unless such review is separately documented.
Research Attribution and Publication Information
Document: Research Data & Evidence Standards
Research author: CGO Media Research Team
Publisher: CGO Media
Research programme: CGO Media Research Methodology
Subject areas: Research data, evidence standards, source verification, data provenance, reproducibility, research integrity, SEO, GEO and AI search.
Canonical publication: https://cgomedia.com/research-data-and-evidence/
Research governance: CGO Media Ethics & Integrity
Parent methodology: CGO Media Research Methodology
Publication date: 22 October 2026