GEO audit: how do you diagnose your visibility in AI engines?

What is a GEO audit?

GEO aims for citation in these answers; the audit tells you where you start from and what is blocking. It measures two things, separately:

• cited as a brand: the engine names the company, which mostly depends on its mentions outside the site;
• cited as a source: a URL of the site appears as a reference, which depends on crawler access, extractability and how well the page fits the query.

It is not a guarantee of citation, a ranking or a one-off reading. Its technical foundation is that of the SEO audit, which checks visibility on Google; carried out together, they make an SEO and AI audit.

When should you run a GEO audit?

A GEO audit is justified as soon as one of these signals appears:

• the brand is missing from the answers on its own topics, or competitors are cited in its place;
• the engines give false or outdated information about the company;
• clicks are dropping on queries that show an AI Overview. Seer Interactive measured this on informational queries, as an average over 2025 (published on April 24, 2026). Organic CTR is 2.07% when the brand is cited in the AI Overview, 0.94% when it is not (about 1.61% excluding one account that makes up 47% of impressions) and 3.35% without an AI Overview;
• a tracking tool is being considered, or a redesign is being prepared.

How do you run a GEO audit, step by step?

A GEO audit is carried out in six steps, in this order. The first two are blocking: a page that crawlers cannot read, or that Google cannot show, will not be cited.

Vishwakarma et al. (Sprinklr, May 25, 2026, accepted at SIGIR ’26) ran 252,000 trials on 6 LLMs. Topical relevance and position in the list weigh the most on the first citation. A review of 45 studies (Martinez, July 15, 2026) finds these levers to be the most reproducible.

1. AI crawler access: crawlers allowed, blocked or filtered.
2. Eligibility on Google: pages indexed and eligible to be shown with a snippet.
3. Prompt panel: baseline of mentions and citations.
4. Citability of key pages: score per criterion, with its evidence.
5. Sources and representation: cited domains, claims about the brand.
6. Diagnosis: case A, B or C, prioritized action plan.

Step 1: check AI crawler access

A GEO audit starts with robots.txt, read crawler by crawler, separating the uses (vendor documentation accessed on October 19, 2026):

• Search index: OAI-SearchBot, Claude-SearchBot, PerplexityBot. According to OpenAI, a site that blocks OAI-SearchBot no longer appears in ChatGPT search answers, except as a navigational link. Anthropic states that blocking Claude-SearchBot may reduce visibility in its results.
• User-initiated fetching: ChatGPT-User and Perplexity-User visit a page when a user asks for it; according to OpenAI and Perplexity, robots.txt may not apply to them. At Anthropic, blocking Claude-User prevents this fetching.
• Training: the other crawlers, detailed on our AI crawlers page.

robots.txt blocks nothing on its own: “These rules are not a form of access authorization” (RFC 9309, IETF, 2022). The firewall, the CDN or the server are what actually block, sometimes without your knowledge. Since September 15, 2026, Cloudflare blocks Training and Agent crawlers by default on pages that display ads, for new domains.

Also check that the text is in the HTML served without JavaScript. In December 2024, Vercel observed no major AI crawler that executes JavaScript, apart from Gemini and Applebot (see our JavaScript SEO guide). The final evidence remains log file analysis: which crawlers visit, on which pages, with which response codes.

Lab finding, on October 19, 2026: the robots.txt of elev8-lab.fr only blocks the admin and login pages. Our French page on measuring AI visibility returns 200 to requests that present themselves as OAI-SearchBot, GPTBot, ClaudeBot or PerplexityBot. This test does not prove that the real crawlers visit: it does not come from their IP addresses.

Step 2: check eligibility on Google

To appear as a link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet. Google adds no other technical requirement and guarantees neither indexing nor display (updated on December 10, 2025).

So check noindex, nosnippet, data-nosnippet and max-snippet: these directives limit what is shown from the page, AI features included. URL Inspection shows the HTML received by Googlebot.

Google’s guide to generative AI (July 10, 2026) rules out, for Google Search, the llms.txt file, splitting content into small chunks and seeking inauthentic mentions. Structured data is not required there.

Step 3: record presence on a prompt panel

The baseline of a GEO audit is recorded on a fixed panel: 50 to 100 prompts according to the lab benchmark, across several engines. Record brand mentions and URL citations separately, and run each prompt several times.

According to the same study, Google Search and the OpenAI API retrieve only 28.3% and 26.2% of the URLs cited by ChatGPT Web.

Add the free reports. The Search Console generative AI report gives impressions without clicks (all sites since August 31, 2026); AI Performance in Bing Webmaster Tools shows citations in Copilot (since February 10, 2026). The panel and its metrics are described on our page on how to measure your AI visibility.

Step 4: score the citability of key pages

A GEO audit selects 5 to 10 pages that are already visible in SEO, each with a target query: the question a user would ask the engine. Score them with the grid below, each score backed by written evidence (HTML excerpt, HTTP header, recorded date). Fixes then follow our GEO optimization method.

Step 5: analyze cited sources and brand representation

On the priority prompts, record every URL cited by each engine, aggregate them by domain and classify each domain: media, comparison site, forum, competitor, directory. Where the brand is missing, note the possible action.

Then check how the brand is represented: ask each engine questions about the company and classify each claim (accurate, outdated, false, missing). Trace the error back to its likely source, fix it at the source, then measure again.

Across 75,000 brands, Ahrefs finds that web mentions are the factor most correlated with presence in AI Overviews (0.664), far ahead of backlinks (0.218). This is a correlation, not proof of cause (May 2025, article modified on April 27, 2026). On the site side, the brand entity is declared in structured data (Organization, sameAs).

Step 6: make the diagnosis and prioritize

The findings of the GEO audit lead to one of three cases, each with its own workstream.

According to the lab procedure, the most frequent case combines B and C. The plan ranks each action by impact and effort, with its page and query, with no promised timeline: no study establishes one.

The Elev8 Lab citability grid: 21 criteria scored from 0 to 3

In a GEO audit, the Elev8 Lab grid scores a page for a target query: 8 families, 21 criteria scored from 0 to 3 on a closed scale. Each level is described, and a criterion without evidence stays blank. The score is used to prioritize GEO optimizations, not to predict a citation.

Each criterion is worth its family’s weight divided by the number of criteria in the family, in proportion to its score out of 3. A blocking family scored below 34% caps the score at 35 (F1 or F2) or at 60 (F3). Bands: A from 75, B from 55, C from 35, D below.

The supporting studies diverge. For Vishwakarma et al., an explicit price and a recent date help, completeness and trust contribute less, and formatting alone almost nothing.

The GEO-16 study by Kumar and Palkhouski (September 13, 2025, 1,702 citations) ranks metadata and freshness first (r = 0.68), followed by semantic HTML (r = 0.65) and structured data (r = 0.63). Zhang et al. (April 29, 2026) observe that the pages picked up most are rich in definitions, figures, comparisons and steps.

Example: our French page on measuring AI visibility

Scoring with the Elev8 Lab grid, on October 19, 2026: French page elev8-lab.fr/geo/mesure/ (English version: AI visibility: how to measure it), target query “mesurer sa visibilité ia” (measure your AI visibility), read from its public HTML. Result: 78/100, band A. The score is provisional: organic ranking (criterion 2.2) was not recorded and counts as zero, leaving up to 5 points pending. Elev8 Lab is scoring its own page: a demonstration of method, not an independent assessment.

The three resulting priorities:

1. Date: display an update date on the page and serve a Last-Modified header consistent with dateModified (blocking family, up to 4.4 points).
2. Author: name the author on the page, with a short bio and a link to their page, and declare them in the markup (up to 4.4 points).
3. Standalone sections: open the sections on Search Console and on monthly metrics with an answer sentence, before the table (1.7 points).

What should a GEO audit deliverable contain?

The deliverable of a GEO audit must make it possible to repeat each measurement. Check that it contains:

• the scope: engines, country, language, dates;
• the prompts, their parameters and the number of repetitions;
• mentions and citations, counted separately;
• the cited sources, aggregated by domain;
• the scores per criterion, with the evidence and the target query of each page;
• the access findings: robots.txt, CDN, server logs;
• the case identified (A, B or C) and the prioritized action plan;
• the written limits: variability, sample size, engines not covered.

Free GEO audit, tool or service: which should you choose?

A free GEO audit is enough for a first diagnosis. The choice depends on what you need to see.

No public price list exists: compare quotes for the same scope (prompts, engines, repetitions, pages scored, competitors). Our SEO and GEO audit service covers both sides.