1. Introduction: The day your client asks you why ChatGPT cites their competitors
Imagine a pretty mundane scene nowadays. You work for a brand that ranks well in SEO. The pages exist. The blog publishes. The service pages are clean. On Google, nothing is perfect, but the site isn't invisible.
Then someone on the team tests a query in ChatGPT, Perplexity or Copilot. The question looks like a very clear business intent: "what tools should I use to track my visibility in AI?" or "which agency should I choose for a GEO audit?". And there, a slightly cold surprise: three competitors appear. Your brand isn't mentioned. Or worse, it's described incorrectly.
That's exactly when a GEO audit becomes useful.
A GEO audit isn't about randomly asking "does my brand appear in ChatGPT?". That would be too fragile. A serious GEO audit measures how a brand, product, service or expert is understood in generative engines. It looks at queries, responses, sources, competitors, errors, stability and editorial opportunities.
The goal isn't to produce an impressive dashboard. The goal is to know what to fix.
This page gives you a concrete method for conducting an actionable GEO audit. We'll talk about scope, query panels, engines to test, reading citations, scoring, prioritization and deliverables. For continuous measurement, you can then dive deeper with GEO monitoring. Here, we focus on the audit: the starting photo, well framed, that lets you decide.
2. Definition: What is a GEO audit?
A GEO audit is a structured analysis of a brand's presence in responses generated by AI engines. It seeks to answer a simple question: when a user asks for a recommendation, definition, comparison or method related to your market, is your brand present, correctly described and supported by good sources?
This definition seems simple. In practice, it requires looking at several layers.
Visibility
Is the brand cited? Is it absent? Does it appear only in sources, but not in the response? Is it recommended or just mentioned? These nuances completely change the reading.
Quality of description
A brand can be cited and yet misunderstood. If a GEO tool is described as just an SEO tool, if a specialized agency is placed in too broad a category, or if a recent offering doesn't appear, the diagnosis isn't positive.
Sources used
Generative engines sometimes rely on visible sources, sometimes on a less transparent mix. An audit must identify the pages cited, competing sources, external mentions and content that seems to influence the response.
Competitive context
You're not just measuring your presence. You're also measuring who's taking your place. If three competitors keep coming up, you need to understand why: better content, clearer category, external mentions, more exploitable documentation, comparatives, more structured pages.
3. GEO audit, SEO audit and brand audit: what changes
A GEO audit borrows from SEO, content analysis, reputation audit and competitive studies. But it doesn't overlap with any of these subjects.
It's not a classic SEO audit
An SEO audit often checks indexation, technical aspects, content, keywords, internal linking, backlinks and performance. All of that remains useful. But a GEO audit looks at another surface: the generated response. You can have a page well-ranked in Google and never be picked up by a generative engine on a recommendation query.
It's annoying, but not illogical. Generative engines don't just copy a ranking. They synthesize, compare, select and rephrase.
It's not just a notoriety audit
A well-known brand can be poorly described. A less-known brand can be well understood on a specific topic. Notoriety helps, but it doesn't replace clarity of entities and quality of sources. On this point, the page on entities and sources complements the audit very well.
It's not a collection of screenshots
Many first GEO audits look like a series of screenshots. That's better than nothing, but insufficient. A screenshot shows one case. An audit must reveal a trend: types of queries where the brand appears, engines where it's absent, dominant competitors, recurring sources, description errors.
4. Defining the audit scope
Before testing queries, define the scope. Otherwise, the audit goes in all directions. It's the most frequent mistake: wanting to measure everything, then not knowing what to decide.
The brand or entity being audited
Are you auditing a brand, a product, a person, an agency, a tool, a local offering, an entire category? This precision changes the queries, competitors and reading criteria.
Markets and audiences
A GEO audit for an SEO/GEO agency doesn't look like an audit for an e-commerce, SaaS or local firm. Queries must reflect real use cases. For an agency, you'll test searches for service providers, methods and comparisons. For a SaaS, you'll test searches for tools, alternatives, features and use cases.
Engines to test
You can test ChatGPT, Perplexity, Google AI Overviews, Claude, Copilot or other interfaces. The most important thing is not to mix results as if they came from the same system. Each engine has its logic, visible sources, limits and variations.
To understand the differences, you can read the GEO engines comparison.
Observation period
A one-off audit gives you a photo. If the topic is strategic, you can re-run some tests over several days to avoid over-interpreting an isolated response. But be careful not to turn the audit into permanent monitoring. These are two different approaches.
5. Building a GEO query panel
The query panel is the heart of the audit. If it's poor, the entire diagnosis will be shaky. A good GEO query must represent a real user intent, not just an SEO keyword turned into a question.
Definition queries
They show who structures the topic. Example: "what is GEO in marketing?", "how does Generative Engine Optimization work?", "what's the difference between GEO and SEO?". These queries often show dominant educational sources.
Method queries
They evaluate a site's ability to be picked up as a practical reference. Example: "how do I run a GEO audit?", "how do I optimize a page for ChatGPT?", "how do I measure AI citations?". These queries are valuable for agencies and experts.
Recommendation queries
They're the most business-sensitive. Example: "best tools to track visibility in ChatGPT", "GEO agencies in France", "solutions to monitor AI responses". Here, the brand's presence or absence becomes very concrete.
Comparison queries
They let you see if the brand is compared to the right competitors. Example: "Geoptim vs [competitor]", "alternatives to [tool] for GEO", "SEO tool or GEO tool to track AI".
Problem queries
They start from a user pain point. Example: "my brand doesn't appear in ChatGPT", "why does Perplexity cite my competitors", "how to fix a wrong description in AI". These queries often reveal very strong content opportunities.
A useful panel can start with 30 to 80 queries. Less than that, you risk seeing too little. Much more, you risk producing a heavy audit before having a stable method.
6. Reading an AI response without getting fooled
Reading a generated response requires more method than you might think. The natural reflex is to look for the brand name, then conclude too quickly. That's insufficient.
Explicit presence
The brand is named in the response. It's the most visible signal, but you need to look at context. Is it recommended? Cited as an example? Mentioned at the bottom of the list? Described as a secondary option?
Implicit presence
The brand may be absent from the main text but present in sources, or picked up through a site page without the brand being highlighted. It's an interesting situation: the content may be useful, but the brand entity isn't connected enough.
Correct description
A poorly worded citation can harm. If the engine says an agency does local SEO when it mainly sells GEO, or that a tool does classic rank tracking when it monitors AI citations, you need to treat it as a comprehension error.
Visible sources
When sources are visible, note them. Are they internal, external, recent, competitive, editorial, commercial? A response citing a competitor comparison while ignoring your method page gives a very clear lead.
Competitors cited
Competitors are sometimes more important than your own presence. If the same names keep appearing across multiple engines, they probably structure the topic. The audit must explain why.
7. Indicators to track during the audit
A GEO audit can quickly become subjective. To avoid that, track simple indicators. You don't need to create an elaborate system. You need signals stable enough to guide decisions.
Presence rate
On how many queries does the brand appear? You can calculate it globally, then by query type, by engine and by intent. An average presence rate doesn't mean much without segmentation.
Mention quality
A mention can be positive, neutral, weak, incorrect or ambiguous. I prefer this reading to a simple yes/no. A brand cited but poorly described hasn't gained much.
Generative share of voice
How many times do competitors appear compared to you? On which intents? In what order? This measure helps you escape self-centeredness. GEO is competitive, even when the response looks like a neutral synthesis.
Sources cited
Which pages appear as sources? Yours? Those of competitors? Media? Forums? Comparatives? This list is often more actionable than the response itself.
Description errors
Note errors with precision: wrong category, old offering, wrong country, wrong audience, confusion with a competitor, invented feature, missing key feature.
These signals can then join a more complete grid in GEO KPIs.
8. Scoring a query in a GEO audit
Scoring must stay readable. If nobody understands the score, it won't help. Here's a simple 5-point scale.
- 0: the brand is absent and no useful source appears.
- 1: a source linked to the brand appears, but the brand isn't mentioned.
- 2: the brand is mentioned, but weakly or ambiguously.
- 3: the brand is cited correctly, but with no clear advantage.
- 4: the brand is cited correctly and associated with a good use case.
- 5: the brand is recommended, well-described and supported by a relevant source.
This score isn't strictly scientific. It's meant to prioritize. A high-intent business query with a score of 0 deserves more attention than a secondary informational query with a score of 3.
Add an importance level
Not all queries are equal. A high commercial intent query should weigh more than an informational query far removed. You can add a priority: low, medium, high.
Add an action status
Each line should lead to a decision: monitor, fix a page, create content, strengthen an external source, clarify an entity, work on a comparative, update documentation.
9. Identifying causes of absence
When a brand doesn't appear, the wrong conclusion would be: "we need to publish more". Sometimes yes. Often, it's more subtle.
The brand isn't clearly defined enough
If your site doesn't clearly explain what you do, for whom, with what deliverables, engines will struggle to recommend you. Entity clarity is a prerequisite.
Content doesn't answer the right intents
You can have lots of articles and miss the query that matters. For example, publishing on "the future of SEO" doesn't replace a real page "how to run a GEO audit".
Competitors have better sources
A competitor may be cited not because it's better, but because it has pages easier to pick up: comparatives, documentation, definitions, studies, external profiles.
External sources tell another story
If directories, profiles or old mentions describe the brand poorly, engines may pick up an outdated version. It's a common problem when a company changes its positioning.
The page exists, but it's too vague
A long page can be weak. If it repeats promises without method, examples and criteria, it gives little material to generative engines. The E-E-A-T GEO page helps strengthen this dimension.
10. Concrete example: GEO audit for an agency
An SEO agency wants to know if it's credible on GEO. It tests 50 queries split into four groups: definition, method, service, comparison.
Result: it rarely appears on recommendation queries, but its articles are sometimes cited on definition queries. Engines understand it as an educational source, not an agency to contact.
So the diagnosis isn't "publish more articles". The real problem is the gap between informative content and visible offering. The agency needs to create or strengthen a GEO service page, add deliverable examples, link its guides to its offering, clarify its client cases and fix a few external profiles still focused on classic SEO.
That's a good example of GEO reading: visibility exists, but it doesn't yet serve the right objective.
11. Concrete example: GEO audit for a SaaS
A SaaS already tracks its SEO positions. It wants to understand why it doesn't appear in "best tools to monitor visibility in AI" type responses.
The audit shows three things. First, cited competitors have more explicit comparative pages. Next, the SaaS's documentation explains features well, but public pages still talk mostly about SEO. Finally, several external sources classify it in an old category.
The action plan becomes clear: create a GEO category page, update feature pages, produce an honest comparative, request corrections to priority external profiles, then track the query across multiple engines.
Without the audit, the team probably would have added three blog articles. With the audit, they know the problem is mainly a category problem.
12. Deliverables of a useful GEO audit
A good audit must be readable by an expert, but also by a decision-maker. If the deliverable is too technical, it will stay in a file. If it's too simplified, it won't help the teams that need to make corrections.
An executive summary
It must say where the brand is visible, where it's absent, which competitors dominate and what problems are priority. No need for ten pages. You need a clear reading.
A query table
Each row should contain the query, the engine, the intent type, the response observed, the brands cited, sources, score, importance and recommended action.
A competitive map
Which competitors come up most? On which engines? On which intents? Are they cited because of their site, media, comparatives, reviews?
A list of errors
Description errors should be isolated. They're often quick to fix if you know where they come from.
An action plan
The action plan should distinguish quick fixes, content to create, external sources to update and signals to monitor.
13. Prioritizing after the audit
An audit without prioritization creates frustration. You get lots of observations, then nobody knows where to start.
Start with absent business queries
If the brand is absent from queries where a user is looking for a solution, service provider or tool, that's priority. These queries carry strong intent.
Fix description errors
A bad description can be worse than an absence. It installs a wrong version of the brand. Fix internal pages first, then manageable external sources.
Strengthen pages already close to the result
If a page is cited as a source but the brand doesn't appear in the response, there may be a quick opportunity. You need to better link the page to the brand entity, offerings and proof.
Create missing content
When no page really answers an important intent, you need to create. But creating after diagnosis is very different from creating at random.
14. Common mistakes in a GEO audit
Testing too few queries
Three prompts don't make an audit. They give an impression. An impression can be useful to start, but not to make serious decisions.
Changing prompts all the time
If each test uses an improvised wording, you can't compare results. Keep a stable base, then optionally add variations.
Only looking at ChatGPT
ChatGPT is important, but it doesn't represent the whole ecosystem. Perplexity, Google AI Overviews, Claude or Copilot can give very different signals.
Confusing citation with recommendation
Being cited isn't always being recommended. A brand can appear as a secondary example or peripheral source. The audit must qualify the mention.
Forgetting sources
Sources often explain the result. If you don't track them, you miss the most actionable part of the audit.
15. How to present a GEO audit to a client
The form of the deliverable matters almost as much as the analysis. If you deliver a too-raw document, the client will see a few screenshots and lots of numbers, but won't know what to do. If you simplify too much, you lose the nuance of the diagnosis.
Start with decisions, not data
I advise opening with three elements: where the brand is strong, where it's absent, and what fixes are priority. Detailed tables come next. A decision-maker must understand in five minutes whether the problem is missing content, bad description, weak sources or competitive dominance.
Show the most telling examples
An audit may contain 60 queries, but not all deserve a screenshot in the summary. Choose examples that best tell the diagnosis: a business query where the brand is absent, a response where a competitor is very well described, an outdated external source, a correct citation but poorly leveraged.
Separate findings from recommendations
The finding says what's happening. The recommendation says what to do. This separation avoids shortcuts. For example: "the brand is absent from recommendation queries" is a finding. "Create a comparative page and strengthen the GEO service page" is a recommendation. Both must stay connected, but not mixed.
Assume the level of uncertainty
Generative responses aren't as stable as server log reports. You need to say so. A good audit doesn't claim to freeze everything. It indicates what was tested, when, with which engines, and what seems recurring. This caution strengthens the credibility of the deliverable.
16. Model table for a GEO audit
A GEO audit table can stay simple. What matters is that it lets you filter, compare and decide.
- Query tested.
- Intent type: definition, method, recommendation, comparison, problem.
- Engine tested.
- Brand present: yes, no, source only.
- Type of mention: mention, example, alternative, recommendation.
- Description quality: correct, weak, wrong, outdated.
- Competitors cited.
- Visible sources.
- GEO score.
- Business priority.
- Recommended action.
This structure avoids turning the audit into free commentary impossible to compare. It also lets you easily move from a one-off audit to regular tracking.
Add a "probable cause" column
This column changes a lot. It forces analysis. Missing dedicated page, unclear entity, outdated external source, better-structured competitor, overly promotional content, insufficient linking. Even if the hypothesis needs verification, it guides action.
Add an "effort" column
Fixing wording on a strategic page doesn't have the same cost as a complete study or an external source update campaign. By crossing priority and effort, you get a realistic roadmap.
17. Moving from audit to roadmap
The real deliverable of a GEO audit isn't the score. It's the roadmap. The score helps you understand. The roadmap helps you move forward.
Quick fixes
These are actions doable in a few hours: clarify an introduction, add a definition, link a page to an offering, fix an FAQ, update a positioning sentence, add a section on covered engines.
Editorial projects
These are content to create or restructure: GEO audit page, KPI page, comparative, case study, documentation, engine-by-engine guide. They take more time, but build a durable foundation.
Source projects
These are off-site fixes: partner profiles, directories, marketplaces, media pages, old mentions, tool descriptions. It's less comfortable than editing your own site, but sometimes necessary.
Measurement projects
Once initial fixes are done, you need to verify if signals move. That's where the audit makes way for tracking. Without this step, you don't know if the work improved perception or just internal content.
18. 30-day action plan
Week 1: define the scope, engines, competitors and query panel. Keep a reasonable number so you can analyze properly.
Week 2: run tests, track responses, citations, sources, competitors and errors. Don't conclude too early.
Week 3: score results, group problems and identify causes: missing content, unclear entity, outdated external source, vague page, better-structured competitor.
Week 4: deliver the action plan. Prioritize business queries, description errors and pages already close to a useful citation.
19. Frequently asked questions about GEO audits
How many engines should you audit?
Three engines are often enough for a first serious audit: ChatGPT, Perplexity and Google AI Overviews where the market supports it. Claude and Copilot can be added if audience or B2B context justifies it.
Should you audit all pages on the site?
No. Start with pages that carry important intents: definition, method, offering, comparison, use case. A GEO audit isn't an exhaustive crawl. It's an analysis of visibility and comprehension.
How often should you redo a GEO audit?
For a strategic topic, a full audit can be redone each quarter. Between audits, lighter tracking shows if fixes produce effects. That's the role of monitoring.
Can you guarantee a citation after an audit?
No, and you should be very clear about that. An audit identifies blockers and opportunities. It doesn't control generative engines. Promising a guaranteed citation is a bad idea.
20. Conclusion: a GEO audit must end in decisions
A GEO audit only has value if it turns generated responses into concrete decisions. Otherwise, it's just a collection of amusing tests.
The right question isn't only "are we cited?". The right question is: "are we understood, recommended in the right contexts, and supported by the right sources?".
In my view, the best GEO audits will be those that stay sober. Few useless metrics, lots of business reading, and a clear action plan. That's where the audit becomes truly useful.