GEO audit: analyze your visibility in ChatGPT and AI engines

01

What is a GEO audit?

A GEO audit is a diagnosis of a brand’s visibility in generative engines. It answers two separate questions: what do AI systems show when asked about your market, and why do they show it.

GEO stands for Generative Engine Optimization. The term was formalized in the research paper “GEO: Generative Engine Optimization” (Aggarwal et al., presented at KDD 2024). It covers the actions that increase the probability that a piece of content is retrieved and reused in a generated answer.

Level 1: measure what AI systems show

The first level of a GEO audit observes the answers produced on a set of questions representative of the market. It records:

• whether the brand is present in the answer;
• citations of the brand’s domain as a source;
• competitors cited instead of, or next to, the brand;
• the sources the engine relies on;
• the brand’s position within the answer;
• the context in which the brand is mentioned;
• the sentiment attached to the brand.

Level 2: understand why

The second level of a GEO audit looks for the causes of the observed result. It analyzes crawler accessibility, content retrievability, information structure, entities, semantic coverage, authority, external corroboration, freshness, structured data, SEO foundations and third-party sources.

Measuring citations tells you the outcome. Auditing GEO also means identifying the mechanisms behind that outcome. A GEO audit limited to logging ChatGPT answers describes a symptom without naming its cause.

02

Why run a GEO audit?

A GEO audit is worth running when a company sees a gap between its actual market position and the way generative engines present it, or when it wants a measured starting point before investing. The most common situations:

• Your competitors are recommended by ChatGPT, your brand is not, on questions where you are a legitimate answer.
• Your company is well known but your site is never cited: AI systems talk about you through third-party sources.
• AI systems give wrong information about your company: outdated offer, wrong price, wrong area, confusion with a namesake.
• Your brand is cited in the wrong context, for instance as a budget option when you are positioned at the high end.
• Some offers or areas of expertise are not associated with your brand, even though they make up a large share of your business.
• You produce a lot of content without knowing whether generative engines can use it.
• You want to fold GEO into an existing SEO strategy without duplicating work.
• You want a baseline to measure the effect of a GEO strategy later on.

03

What does a GEO audit include?

A complete GEO audit covers fourteen pillars, grouped in three families: what the engines show (pillars 1, 2 and 13), what drives their choices (pillars 3 to 12) and what it produces for the business (pillar 14). They do not all weigh the same for every site: scoping decides which ones are examined in depth.

01 Actual visibility in generative engines

The first pillar measures the brand’s current presence in generative engine answers, on a panel of prompts that reproduces the questions its customers actually ask. Prompts are grouped by intent:

• discovery and information;
• problem to solve;
• comparison and alternatives;
• recommendation and vendor selection;
• named brand, product or service;
• transactional queries.

The engines tested depend on the market: ChatGPT, Gemini, Google AI Overviews and AI Mode, Perplexity, Claude, and any other engine the audience uses.

A generative answer is probabilistic: the same question asked twice can produce two different answers. Each prompt is therefore run several times per engine, and answer stability becomes a measure in its own right. The resulting measures: mention rate, citation rate, share of voice, presence per engine, frequency, competitors present, relative position and stability.

Mention rate measures how often a brand appears in generated answers. Citation rate measures how often its own domain is used as a source. A brand can therefore be mentioned frequently without its site ever being cited.

02 Analysis of the sources AI systems use

The second pillar maps the sources that shape generative engine answers. For each prompt family, the audit records the domains cited, the pages cited, the source type, its owner, its date, the competitor it benefits and the information extracted from it.

Sources fall into recurring families: media, specialist sites, comparison sites, forums and Reddit, directories, documentation, institutional sites, brand sites, professional profiles. Identifying these families shows where a piece of information needs to exist to stand a chance of being retrieved, which often goes well beyond the brand’s own site.

03 Accessibility to engines and crawlers

The third pillar checks that no technical obstacle prevents engines from reading the site. Content a crawler cannot fetch cannot be cited, however good it is. Items checked: HTTP status, indexability, robots.txt, meta robots, canonicals, redirects, orphan pages, architecture and depth, sitemap, HTML rendering, JavaScript dependency, actual access for the relevant crawlers, and any blocking at CDN or web application firewall level.

OpenAI’s crawlers have distinct roles, documented by OpenAI. OAI-SearchBot is used to surface websites in ChatGPT’s search features: a site that blocks it does not appear in ChatGPT search answers. GPTBot crawls content that may be used to train OpenAI’s models. The two are configured separately in robots.txt.

Allowing GPTBot is not required to appear in ChatGPT search. Access for OAI-SearchBot is what determines eligibility; GPTBot is a separate decision about using content for training.

OpenAI states that a robots.txt change can take about 24 hours to take effect, and publishes the IP ranges of its crawlers. A CDN or firewall setting can block these crawlers without robots.txt showing it: the audit therefore checks actual access, not only the declaration.

04 Retrievability

Retrievability is how easily a relevant piece of information can be found, isolated and retrieved from a piece of content to answer a given question.

Retrievability is the pillar that sets a GEO audit apart from a classic content audit. A page can contain the right answer and still be unusable if that answer is scattered, context-dependent or buried in low-value text. The audit analyzes:

• paragraph autonomy and information granularity;
• hierarchy and heading clarity;
• the presence of direct answers and the distance between a question and its answer;
• the context needed to understand a passage;
• referential ambiguity (“it”, “this solution”, “we” with no antecedent);
• factual density and editorial noise;
• duplication and information spread across several pages.

Retrieval cost is the effort a system has to make to extract a piece of information from content. It rises when the system has to read through a lot of text, resolve ambiguous references, combine several sections or infer implicit relationships. Position matters too: the “Lost in the Middle” study (Liu et al., TACL 2023) showed that language models use information placed in the middle of a long context less effectively than information at the beginning or the end.

The goal is to reduce tokens consumed, ambiguity and context dependency, while increasing the useful information, precision and semantic autonomy of each passage.

05 Information structure and citability

The fifth pillar assesses whether a page’s passages can be reused as they stand in an answer. The audit checks for explicit questions, direct answers, definitions, lists, tables, comparisons, sourced figures, causal relationships, examples, nuances and exceptions.

Every important unit should work as a self-contained passage. “As explained above, this method makes it possible” cannot be reused outside the page. “A GEO audit identifies the sources generative engines use when they answer queries related to a brand” can.

06 Semantic coverage

The sixth pillar checks that content covers the topic and spells out the relationships between its concepts instead of listing them side by side. For the topic of this page, the structuring relationships are:

• a GEO audit measures a brand’s AI visibility in generative engines;
• a generative engine (ChatGPT, Gemini, Perplexity, Claude, AI Overviews, AI Mode) produces its answer with an LLM, often after a search step;
• RAG (retrieval-augmented generation) retrieves passages from sources before generating the answer: a piece of content’s retrievability determines whether it enters that process;
• a citation links the answer to a source; a brand mention names the company without necessarily citing its site;
• crawling and indexing determine whether engines can access the sources;
• entities and the knowledge graph structure what an engine associates with a brand; structured data makes them explicit;
• authority, reputation and sentiment shape the trust given to sources and the way the brand is described;
• prompts and analytics tools are used to measure the outcome.

GEO visibility depends in part on a generative engine’s ability to find relevant information in the available sources before including it in its answer.

07 Entities and brand understanding

The seventh pillar compares what the brand says about itself with what generative engines appear to associate with it. Relationships examined: brand and business, products or services, expertise, people, geographic area, competitors, attributes and reputation. For each one, the audit classifies the relationship as correct, missing, ambiguous, outdated or contradictory.

Each observed row is linked to the sources that carry it, so you know where to correct it.

08 Authority and external corroboration

The eighth pillar measures brand authority beyond backlinks: citations, brand mentions, links, articles, profiles, interviews, reviews, databases, press, communities, professional associations, third-party documentation.

The central question of authority in GEO: are there enough independent sources to corroborate what the brand claims about itself?

A site on its own can state “we are GEO experts”. The same relationship stated on several independent sources becomes much easier for an engine to corroborate.

09 Expertise, trust and evidence

The ninth pillar checks that every important claim can be verified. The audit looks for an identifiable author, a biography, demonstrated experience, sources, a methodology, references, real cases, dates, an update history and clear editorial responsibility. The test applied to every claim: how can this be verified?

10 Information freshness

The tenth pillar spots outdated information: dates, statistics, platforms, features, documentation and technologies mentioned. GEO moves fast, and a page that was accurate a year ago can contain claims that have since become false.

Freshness is managed through a real update policy, reflected in the datePublished and dateModified properties. Changing a date without changing the substance is a misleading signal, and the audit flags it as such.

11 SEO foundations

The eleventh pillar covers SEO, because a GEO audit cut off from SEO misses part of the causes. Generative engines rely on search indexes and retrieval steps: poor classic discoverability limits generative discoverability. Google states that a page must be indexed and eligible to be shown in Google Search with a snippet to appear in AI Overviews and AI Mode.

Items analyzed: crawlability, indexability, architecture, content, search intent, internal linking, duplication, cannibalization, authority, backlinks, performance, SERPs and the match between query and page.

12 Structured data and machine-readable signals

The twelfth pillar reviews the Schema.org structured data relevant to the site: Person, Organization, Service, Article where justified, BreadcrumbList, and any other types that genuinely match the content.

Structured data is not a magic citation factor. Google states that no special markup is needed to appear in AI Overviews or AI Mode. Its role is to make information that already exists on the page explicit in a machine-readable format.

Adding properties that match nothing visible on the page goes against Google’s structured data guidelines. The audit flags it as a risk, not as an optimization.

13 Sentiment and brand representation

The thirteenth pillar measures how the brand is described, not only whether it appears. The audit builds a lexicon of attributes (premium, budget, expert, innovative, reliable, local, specialist, generalist…) and compares the positioning the brand intends with the positioning the engines generate.

It also identifies recurring criticism, negative information, ambiguities and the competitors systematically associated with the brand.

14 Business measurement

The fourteenth pillar connects GEO visibility to data the business can act on. Where the data exists, the audit analyzes visits referred by ChatGPT, Perplexity and Gemini, landing pages, conversions, leads, revenue and how they change over time.

The audit keeps three levels apart: visibility, traffic and conversion. More citations do not mean an equivalent rise in visits, because a generated answer often satisfies the question without a click. On Google’s side, Search Console includes traffic from AI features in the Performance report, under the “Web” search type, without separating it from the rest.

04

How does a GEO audit work?

A GEO audit runs in six steps, from scoping to roadmap. It is the method I apply to every audit, published here as the Julien Mouttet GEO audit framework, version 1.0 (October 2026). I version it so that every change stays traceable.

01 Scoping

Scoping identifies offers, audiences and personas, competitors, markets and goals. It sets the scope: number of prompts, engines, competitors, pages and markets.

02 Building the prompt set

Customers’ search journeys are turned into families of questions, from discovery to choosing a vendor. Prompts that name the brand are kept apart from those that do not: the former measure what engines know about it, the latter its discoverability.

03 Visibility baseline

Prompts are run several times on each selected engine. The baseline measures citations, mentions, competition, sources and sentiment. It is dated and serves as the starting point for every later measurement.

04 Causal analysis

The causal analysis audits technical setup, content, retrievability, entities, authority and external sources, to explain every gap observed in the baseline.

05 Gap analysis

The brand is compared with its competitors, with the sources engines already select, and with the information those engines appear to look for when answering.

06 Roadmap

Gaps become actions. Each recommendation states:

• the finding;
• the expected impact;
• the solution;
• the priority;
• the page or scope concerned;
• the team concerned (content, technical, PR, marketing).

Version history. v1.0, October 2026: first publication, fourteen pillars and six steps.

05

GEO audit metrics

A GEO audit uses ten metrics, each one computable from the collected answers or from analytics data. None of them is an opaque proprietary score.

Diagnostic matrix

Crossing three observations (is the brand mentioned, is its domain cited, do third-party sources talk about it) gives a first diagnosis and points to priorities.

06

GEO audit vs SEO audit

An SEO audit measures a site’s position in search results; a GEO audit measures how a brand and its content are picked up in generated answers. The units of analysis change:

SEO and GEO are not two independent disciplines. A large part of the technical, semantic and reputational foundations is shared: a serious GEO audit therefore builds on SEO foundations instead of setting the two against each other.

07

What do you receive at the end of a GEO audit?

A GEO audit produces seven deliverables. Their value lies in the decisions they make possible, not in their page count.

1. AI visibility baseline: brand presence by prompt, by engine and by competitor, dated to serve as a reference point.
2. Source map: the domains and pages that shape the answers, grouped by family and by competitor.
3. GEO technical audit: access, rendering and retrieval issues, with the expected fix.
4. Editorial and retrievability audit: pages analyzed, problem passages and recommended rewrites.
5. Entity graph: expected and observed relationships around the brand, with the sources that carry them.
6. Competitive analysis: why some sources or brands are selected instead of yours.
7. Prioritized roadmap: concrete SEO and GEO actions, with impact, priority, scope and owning team.

08

What a GEO audit cannot guarantee

A GEO audit cannot guarantee that a brand will be cited by ChatGPT or any other generative engine. Any offer that promises guaranteed citations promises something no provider controls. The reasons:

• LLM answers are probabilistic and vary from one run to the next;
• engines and their models change, sometimes without notice;
• the sources used vary with wording, timing and search mode;
• no optimization guarantees a citation, it raises its probability;
• a single measurement (one pass over the prompts) is not robust enough to draw conclusions;
• a correlation between an action and a visibility gain does not prove the action caused it.

What the audit does guarantee: a reproducible measurement, documented causes, and actions tied to each finding.

09

Why trust me with your GEO audit?

I am an SEO and GEO consultant with six years of experience in SEO agencies: sales, then SEO project officer at the SEO.fr agency (2020-2023), SEO project manager at netlinking.fr (2023-2024), and SEO consultant at Arkée by Jaws since November 2024, where SEO and GEO audits and their action plans are at the core of my work. I have carried out more than fifty SEO and GEO audits.

What I bring to your GEO audit:

• Technical and editorial SEO experience: technical, content, architecture, internal linking and link building audits across client portfolios.
• Understanding of retrieval systems: I build RAG systems myself (chunking, embeddings, re-ranking, context management), the very mechanisms retrievability is meant to serve.
• Tooled measurement: data collection through the DataForSEO API, analysis in GA4, Semrush and Ahrefs, and a fleet of AI agents I host myself to automate recurring analysis tasks.

SEO experience, an understanding of retrieval systems and measurement of generative engines: this combination is what makes it possible to analyze GEO beyond content, down to its technical and reputational causes.

10

How much does a GEO audit cost?

The price of a GEO audit depends on the volume to analyze, not on a single flat fee. It is set after a scoping call that fixes the number of markets, competitors, prompts and pages to analyze. The variables that move it:

• site size and number of pages analyzed;
• number of markets and languages;
• number of competitors compared;
• number of engines tested;
• size of the prompt set and number of runs per prompt;
• depth of the causal analysis.

The prompt set weighs the most: each prompt is run several times on each engine, so doubling the intents covered or the engines tested multiplies the collection work. To get a quote, describe your situation: market, competitors, goals.

11

GEO audit FAQ