How to Audit AI Search Visibility for AEO and GEO

What Is an AI Search Visibility Audit?

An AI search visibility audit examines how a brand, organisation, website, person, product or other entity appears across relevant AI-powered search and answer environments.

Depending on the project, the audit can investigate:

• Brand mentions.
• Website citations.
• Cited URLs.
• Source selection.
• Topic associations.
• Entity recognition.
• Entity accuracy.
• Competitor or peer mentions.
• Recommendation contexts.
• Answer consistency.
• Citation consistency.
• Representation across different query types.

The audit can then compare those observations with the website’s:

• Technical accessibility.
• Content.
• Information architecture.
• Entity clarity.
• Structured data.
• First-party evidence.
• External corroboration.
• Search visibility.

This helps identify plausible areas for improvement without pretending that every generative response can be reverse-engineered.

AEO and GEO Are Related but Different

I use Answer Engine Optimisation and Generative Engine Optimisation as related disciplines with different analytical emphasis.

Answer Engine Optimisation

AEO focuses on making useful information easier for search and answer systems to retrieve and present in response to a question or information need.

A simplified model is:

Question → Intent → Information → Answer → Retrieval

Generative Engine Optimisation

GEO focuses more broadly on how entities, information, evidence and sources may be understood, selected and synthesised within generative search experiences.

A simplified model is:

Entity → Information → Evidence → Sources → Synthesis → Representation

These areas overlap substantially.

An AI search visibility audit can examine both.

Start With the Entity

Before testing prompts, define what is being audited.

For a company, document:

• Official name.
• Website.
• Products or services.
• Main topics.
• Important people where relevant.
• Locations where relevant.
• Distinguishing attributes.
• Official profiles.
• Important first-party sources.

For a person, this might instead include:

• Name.
• Professional role.
• Website.
• Areas of expertise.
• Professional profiles.
• Published work.

This creates a reference against which AI-generated representations can be compared.

Define the Topics That Matter

Do not begin with hundreds of random prompts.

Identify the topics for which visibility would actually matter.

These may include:

• Core services.
• Products.
• Problems the organisation solves.
• Important informational subjects.
• Commercial categories.
• Brand-specific questions.
• Comparison contexts.
• Industry topics.

Prioritise according to relevance.

AI visibility for an irrelevant topic is not useful merely because the brand was mentioned.

Build a Prompt Set

Create a controlled set of prompts representing important information needs.

Useful categories can include:

The exact categories should reflect the project.

Use Natural Prompt Variations

Users do not all phrase questions identically.

For an important information need, test reasonable variations.

For example:

What is technical SEO?

How does technical SEO work?

What does technical SEO include?

These are related but not identical prompts.

The purpose is not to generate dozens of artificial variants.

It is to determine whether representation remains reasonably consistent across realistic ways of asking about the subject.

Document the Testing Environment

AI systems change frequently.

Record enough information to understand the conditions of the observation.

Depending on the platform and what is visible, document:

• Platform.
• Date.
• Prompt.
• Response.
• Mention status.
• Citation status.
• Cited URLs.
• Other cited sources.
• Relevant observations.

If account state, location, conversation context or other known variables may influence the response, record those where practical.

The purpose is not to pretend the environment is perfectly controlled.

It is to make the test more reproducible.

Establish a Baseline

Before changing the website, establish current visibility.

For each prompt, record whether the target entity is:

• Mentioned.
• Not mentioned.
• Cited.
• Mentioned without citation.
• Represented accurately.
• Represented inaccurately.
• Associated with the intended topic.
• Associated with unexpected topics.

Also record which other entities and sources appear.

This baseline gives future testing something to compare against.

Track Brand Mentions

A brand mention indicates that the system included the entity in its response.

Record:

• Prompt.
• Whether the brand appeared.
• Context of the mention.
• Whether the mention was relevant.
• Whether it was positive, neutral or negative only when this classification can be made objectively enough for the project.
• Whether a citation accompanied the mention.

A mention alone does not prove that the website was used as a source.

Keep mentions and citations separate.

Track Website Citations

When the system provides sources, record whether the target website is cited.

Capture:

• Domain.
• Exact URL.
• Prompt.
• Location or role of the citation where observable.
• Information apparently supported by the citation.
• Other cited domains.

This can reveal which pages are being selected and for what kinds of information.

Do not assume a citation proves that every part of the generated answer came from that page.

Track Cited URLs

Domain-level visibility is useful, but URL-level visibility is more actionable.

A website may be cited primarily through:

• Homepage.
• Service pages.
• Guides.
• Research.
• Product pages.
• Documentation.
• About pages.

Understanding which URLs appear can reveal what types of content the system currently finds useful for particular information needs.

Analyse Source Selection

When citations are available, examine the broader source set.

Ask:

• Which domains are repeatedly selected?
• What type of source are they?
• Are they first-party or third-party?
• Are primary sources being used?
• Are publishers, institutions, documentation or community sources appearing?
• What information do those pages provide?
• What does the target website lack by comparison?

The objective is not to copy competing pages.

It is to understand the evidence environment surrounding the topic.

Evaluate Entity Recognition

Test whether the system understands the target entity correctly.

For example:

• What does the organisation do?
• What services does it provide?
• Where does it operate?
• Who is associated with it?
• What topics is it known for?

Compare the generated information with authoritative first-party information.

Record:

• Correct information.
• Missing information.
• Ambiguous information.
• Incorrect information.
• Confusion with similarly named entities.

Entity clarity is particularly important when names are ambiguous.

Evaluate Entity Accuracy

Visibility without accuracy can be harmful.

Check whether generated answers correctly represent important facts.

For example:

• Organisation type.
• Products or services.
• Locations.
• Professional role.
• Relationships between entities.
• Claims about capabilities.

Where the system is wrong, investigate whether authoritative information is:

• Missing.
• Inconsistent.
• Difficult to access.
• Ambiguous.
• Contradicted by other sources.

Do not assume the website alone controls the model’s representation.

Generative systems may rely on multiple sources and internal model knowledge.

Analyse Topic Association

Determine whether the entity appears in contexts relevant to its actual expertise or business.

For example, an SEO specialist might want to understand whether they are associated with:

• Technical SEO.
• Information architecture.
• Content SEO.
• AEO.
• GEO.

Topic association should be evaluated against actual expertise and published evidence.

The objective is not to manufacture associations unsupported by reality.

Compare Peer and Competitor Representation

Where appropriate, record other entities that appear for the same prompts.

This can help answer:

• Who is being mentioned?
• Who is being cited?
• Which sources recur?
• Which entities are strongly associated with the topic?
• What evidence supports those associations?

This is not an automatic ranking.

Generative answers can include entities for different reasons.

Use comparison to identify patterns requiring further investigation.

Repeat Important Tests

A single AI response is weak evidence.

Important prompts should be repeated over time and, where practical, across controlled variations.

Record whether:

• The entity repeatedly appears.
• Citations persist.
• Different URLs are selected.
• Different sources replace each other.
• Entity descriptions remain consistent.
• Topic associations change.

This helps distinguish a recurring pattern from a one-off response.

Do Not Treat AI Outputs as Traditional Rankings

A conventional search result has an observable ordered result set.

Generative answers behave differently.

An entity appearing first in an answer does not necessarily mean it holds a stable “number one AI ranking”.

Likewise, being omitted once does not establish permanent invisibility.

Avoid converting generative outputs into traditional ranking terminology unless the platform actually exposes such a ranking system.

Create an Observation Matrix

A practical audit spreadsheet can include:

• Prompt ID.
• Topic.
• Intent.
• Prompt.
• Platform.
• Test date.
• Entity mentioned.
• Website cited.
• Cited URL.
• Other sources cited.
• Entity accuracy.
• Topic association.
• Notes.
• Screenshot or evidence reference.
• Retest date.

This creates an auditable record rather than relying on memory.

Separate Observation From Interpretation

Suppose a website is not cited for ten important prompts.

The observation is:

The website was not cited in these recorded tests.

The conclusion is not automatically:

The website lacks authority.

Possible explanations may involve:

• Source preferences.
• Content relevance.
• Entity understanding.
• Evidence quality.
• Query interpretation.
• Platform behaviour.
• Technical accessibility.
• Limited test coverage.
• Normal output variability.

Further investigation is required.

This distinction is essential in an emerging field.

Audit Technical Accessibility

AEO and GEO still depend on many traditional technical foundations.

Review whether important information is:

• Crawlable.
• Indexable where appropriate.
• Available in rendered content.
• Internally discoverable.
• Canonicalised correctly.
• Served successfully.
• Included in a coherent architecture.

If search systems cannot reliably access important information, AI optimisation tactics built on top of it have a weak foundation.

Audit Information Architecture

Examine whether the website clearly communicates:

• Main topics.
• Subtopics.
• Services or products.
• Entity relationships.
• Supporting information.
• Page hierarchy.

Poor architecture can make information fragmented or difficult to contextualise.

Clear architecture does not guarantee AI visibility.

It improves the underlying information environment.

Audit Content Structure

Review whether important pages:

• Address a clear information need.
• Provide direct explanations where appropriate.
• Use logical headings.
• Define important concepts.
• Provide sufficient context.
• Connect related information.
• Distinguish facts from claims.
• Support important claims with evidence.

The objective is not to write in an artificial “AI-friendly” style.

The objective is to make useful information clear and retrievable.

Audit Entity Clarity

Review how clearly the website communicates important entities and relationships.

This may include:

• Organisation.
• People.
• Services.
• Products.
• Locations.
• Topics.
• Relevant relationships between them.

Check consistency across:

• Website content.
• About information.
• Professional profiles.
• Structured data.
• External authoritative references where relevant.

Do not create artificial entities simply for SEO.

Clarify real ones.

Audit First-Party Evidence

Generative systems frequently need information that can be supported.

Evaluate whether the website contains useful first-party evidence such as:

• Original research.
• Methodologies.
• Case Studies.
• Projects.
• Data.
• Documentation.
• Expert analysis.
• Definitions based on genuine expertise.
• Transparent company information.

A website composed entirely of generic summaries may provide little unique evidence.

First-party information can create material worth referencing.

Audit Claims and Sources

For factual claims, ask:

• Is the claim necessary?
• Is it accurate?
• Is it current?
• Is a source needed?
• Is the source primary or authoritative where possible?
• Does the source actually support the claim?

Citation quality matters for users even when it does not produce an observable AI-search effect.

Evidence should exist because the content requires it, not merely because citations are assumed to be a GEO tactic.

Audit Structured Data

Structured data can help describe page content and entities in machine-readable form.

Review:

• Appropriate schema types.
• Accuracy.
• Consistency with visible content.
• Entity relationships.
• Required properties where relevant.
• Duplicate or conflicting markup.

Do not assume schema guarantees inclusion in an AI-generated answer.

Structured data is one part of machine-readable clarity.

Audit External Corroboration

An organisation’s own website is not the only source of information about it.

Where relevant, investigate whether important facts are consistently represented across credible external sources.

This might include:

• Professional profiles.
• Industry publications.
• Primary databases.
• Institutions.
• Reputable media.
• Relevant directories.
• Other authoritative sources.

The appropriate sources depend entirely on the entity and industry.

Do not manufacture mentions or pursue low-quality citations simply to create an artificial footprint.

Analyse Citation Gaps

When other sources are repeatedly cited but the target website is not, compare the information available.

Ask:

• Does the cited source answer the question more directly?
• Does it contain evidence absent from the target site?
• Is it more specific?
• Is the information easier to retrieve?
• Does it provide original data?
• Is it a primary source?
• Does it have stronger corroboration?

A citation gap is an investigation opportunity, not proof of a single ranking factor.

Analyse Representation Gaps

The system may recognise the entity but describe it incompletely.

For example, it may understand one service but omit another important area.

Investigate whether the missing association is clearly supported by:

• Website architecture.
• Dedicated content.
• Projects.
• Case Studies.
• Professional profiles.
• External evidence.

The appropriate response may be to strengthen genuine evidence around the missing area.

Prioritise Findings

An AI search visibility audit can generate many observations.

Prioritise using factors such as:

Business relevance: Does the prompt matter?

Entity importance: Is the missing or incorrect information important?

Evidence: Is the pattern repeatable?

Accuracy risk: Could incorrect representation cause problems?

Visibility opportunity: Is the topic strategically relevant?

Website gap: Is there a clear weakness that can reasonably be improved?

Effort: What would improvement require?

Uncertainty: How confident are we in the diagnosis?

Uncertainty should affect priority.

Turn Findings Into Actions

A finding should lead to a defensible action.

For example:

This is more defensible than:

Add more keywords so ChatGPT cites the website.

Retest After Changes

After meaningful changes, repeat the relevant prompt set.

Compare against the baseline.

Look for changes in:

• Mentions.
• Citations.
• Cited URLs.
• Entity descriptions.
• Topic associations.
• Source selection.

Do not expect every test to change immediately or consistently.

The purpose is to accumulate evidence over time.

Measure Trends Rather Than Isolated Outputs

Useful AI visibility measurement may include:

• Mention frequency within the controlled test set.
• Citation frequency within the controlled test set.
• Number of distinct cited URLs.
• Topics associated with the entity.
• Entity accuracy.
• Source recurrence.
• Representation changes over time.

These are observations within your testing methodology.

They should not be presented as universal platform-wide visibility metrics.

That distinction is important.

AI Search Visibility Audit Workflow

The complete process can be summarised as:

1. Define the entity.
2. Define strategically relevant topics.
3. Build a controlled prompt set.
4. Establish testing conditions.
5. Run baseline tests.
6. Record mentions.
7. Record citations and cited URLs.
8. Analyse other selected sources.
9. Evaluate entity recognition and accuracy.
10. Evaluate topic associations.
11. Repeat important tests.
12. Audit technical accessibility.
13. Audit information architecture.
14. Audit content structure and retrievability.
15. Audit entity clarity.
16. Audit first-party evidence.
17. Audit claims, sources and structured data.
18. Investigate external corroboration.
19. Identify citation and representation gaps.
20. Prioritise findings.
21. Implement justified improvements.
22. Retest using the same methodology.
23. Compare results over time.

AI Search Visibility Audit Checklist

Entity

• Entity clearly defined.
• Important attributes documented.
• Official sources identified.

Topics

• Relevant topics selected.
• Commercial relevance considered.
• Irrelevant visibility excluded from priority analysis.

Prompts

• Controlled prompt set created.
• Multiple intent types represented.
• Important variations included.
• Prompt wording documented.

Testing

• Platform recorded.
• Date recorded.
• Responses preserved where practical.
• Repeat tests planned.

Visibility

• Brand mentions recorded.
• Website citations recorded.
• Cited URLs recorded.
• Other cited sources recorded.

Representation

• Entity recognition checked.
• Entity accuracy checked.
• Topic association checked.
• Confusion with other entities investigated.

Website

• Technical accessibility reviewed.
• Information architecture reviewed.
• Content structure reviewed.
• Entity clarity reviewed.
• Structured data reviewed where relevant.

Evidence

• First-party evidence assessed.
• Claims reviewed.
• Sources assessed.
• External corroboration investigated where relevant.

Analysis

• Citation gaps investigated.
• Representation gaps investigated.
• Competitor or peer appearances recorded where useful.
• Observations separated from interpretations.

Measurement

• Baseline established.
• Repeat testing performed.
• Changes compared over time.
• Uncertainty documented.

What an AI Search Visibility Audit Should Produce

Depending on scope, useful deliverables may include:

• Entity definition.
• Topic map.
• Prompt library.
• Baseline visibility matrix.
• Mention analysis.
• Citation analysis.
• Cited URL inventory.
• Source analysis.
• Entity accuracy assessment.
• Topic association analysis.
• Technical and content findings.
• Evidence gaps.
• Prioritised recommendations.
• Retesting methodology.
• Ongoing observation framework.

The output should make it possible to repeat the analysis rather than relying on screenshots of a few favourable AI responses.

What an AI Search Visibility Audit Cannot Prove

An audit should not claim more than the evidence supports.

It cannot guarantee:

• Future AI citations.
• Stable placement in generated answers.
• Consistent mentions across all users.
• A permanent recommendation.
• A universal AI ranking.
• That one optimisation caused a particular AI response.

Generative systems are dynamic.

Platforms, models, source-selection methods, search indexes and response generation can change.

Measurement therefore needs to acknowledge variability.

AEO and GEO Still Depend on SEO Foundations

AI search optimisation should not be treated as a replacement for SEO.

The same website still needs:

• Technical accessibility.
• Clear information architecture.
• Useful content.
• Search intent alignment.
• Entity clarity.
• Evidence.
• Authority.
• Internal relationships.
• Measurement.

AEO and GEO extend the questions we ask about those foundations.

Traditional SEO asks whether information can be discovered and perform in search.

AEO asks whether useful information can be retrieved effectively for answers.

GEO asks how entities, information, evidence and sources may be represented within generative systems.

The underlying information system remains connected.

The Goal Is Better Evidence, Not AI Manipulation

There will be pressure to turn GEO into a collection of tactics promising guaranteed AI citations.

That is not a useful methodology.

A stronger approach is to improve the things that can be examined and defended:

• Accessibility.
• Information quality.
• Entity clarity.
• Topic coverage.
• Evidence.
• Source quality.
• First-party information.
• Structured relationships.
• External corroboration.
• Measurement.

Then observe whether representation changes.

That is slower than promising a secret AI ranking formula.

It is also much more credible.

Related SEO Resources

This methodology connects with:

• Technical SEO Audit
• Website Information Architecture for SEO
• Internal Linking Audit
• Google Search Console Analysis
• Search Intent Mapping

Together, these resources provide a framework for analysing visibility across both traditional and emerging search environments.