Key Takeaways
- What the audit measures: the brand's presence in the responses, citations of the website as a source, the brand's position within the response, and competitors or third-party websites cited instead of your own.
- The protocol: 20 to 30 representative queries, 4 to 6 search engines, 3 sessions per query in incognito mode, two data collections one week apart.
- The schedule: an occurrence score ranging from 0 to 3 per response, aggregated into a citation rate by search engine and query family.
- The expected result: A list of won, lost, and missing queries, along with the competing sources to match, before any content work begins.
Before investing in the GEO, you need to know where you’re starting from. AI engines don’t provide any ranking reports: the only way to determine your visibility is to query them using a specific protocol and record their responses. This article describes a reproducible audit method that can be completed in half a day for an SME, as well as the pitfalls that can skew the results.
What an AI Visibility Audit Actually Measures
A generated response is not a ranking. To interpret it, you need to distinguish between four levels of occurrence, from the weakest to the strongest:
| Level | What We Observe | Score |
|---|---|---|
| Absent | Neither the brand nor the website appears in the answer or in the sources | 0 |
| Note | The brand is mentioned in the text, without a link or source | 1 |
| Quote | A page from the website appears in the list of sources but is not included in the main text | 2 |
| Sourced recommendation | The brand is mentioned in the response, and a page from the website is cited as supporting evidence | 3 |
In addition to this score, three qualitative factors are considered: the position of the mention (first or last option listed), the tone (neutral, positive, reserved), and the accuracy of the facts presented about the company. An inaccurate review (incorrect prices, discontinued service) is worse than no review at all, because it misleads the prospect. Details on the metrics to track over time can be found in GEO's KPIs.
Step 1: Build the query set
The quality of the audit depends on the questions asked. A set of 20 to 30 questions is sufficient for an SME, divided into four categories:
| Family | Example | Game Share |
|---|---|---|
| Recommendation | "What marketing tracking software is best for an SME with 20 employees?" | 40 % |
| Comparison | "GreenRed or Looker Studio for a GA4 dashboard?" | 20 % |
| Information | "How do I track traffic from ChatGPT in GA4?" | 25 % |
| Brand | "What is GreenRed worth? What are its rates?" | 15 % |
Formulate questions the way a user would type them into a search assistant: complete sentences, including context (company size, industry, city for a local business), in French. Two-word queries inherited from SEO are useless here. To find the right phrasing, start with questions asked during sales meetings, support tickets, and Search Console; the complete method is in Keyword Research for GEO.
Step 2: Test the engines under neutral conditions
Assistants tailor their responses based on the account’s history and data. An audit conducted from the account of an executive—who talks about their brand every day—overstates visibility. The rules:
- Private browsing or incognito mode, cache disabled, no custom instructions.
- Web search enabled on ChatGPT (answers that do not include web search are based on outdated training data).
- Three runs per query and per search engine, since the results vary from one run to the next; use the average.
- Two readings taken one week apart to smooth out variability.
- Six platforms set to dominate the French market in 2026: ChatGPT, Perplexity, Gemini, Google’s AI Mode and AI Insights, Claude, and Mistral. How each one works is described in Understanding AI Engines.
For each result, record the following in a spreadsheet: search engine, query, date, score (0–3), brands mentioned, sources cited (domain and URL), and tone. Thirty queries multiplied by six search engines and three passes yield 540 results: this is the threshold at which an automated tool becomes more cost-effective than manual data collection; the options are compared in AI-powered visibility monitoring tools.
Step 3: Read the results
Aggregate the scores into three indicators:
- Attendance rate: Percentage of responses with a score of at least 1. In 2026, an SME not using GEO typically scores between 0 and 15% on recommendation queries and between 60 and 100% on brand queries.
- Citation rate: Percentage of responses where a page on the site is the source (score of 2 or 3). This is the metric that changes when the content is revised.
- Market share: The number of mentions of your brand divided by the total number of brands mentioned in the same group of search queries.
Then categorize the search queries into three groups: won (consistent presence on at least three search engines), contested (presence on one or two search engines, or inconsistent between checks), and absent. The “contested” group is where content efforts yield results the fastest. For each absent query, identify the three most frequently cited sources: these are the pages you need to match or surpass in terms of accuracy.
Step 4: Check the technical requirements
An absence may be due to a technical issue rather than an editorial one. Check the following, in this order:
- The robots.txt file: GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, and Google-Extended must not be blocked; a firewall or CDN may also filter them without any entries appearing in the file.
- Server logs from the last 30 days: The presence of these bots confirms that the site is being visited; their absence indicates that access is blocked or that the site is considered uninteresting.
- Indexing in Bing, which ChatGPT relies on: a website that isn't indexed in Bing is invisible for half of the queries tested.
- HTML rendering: Content loaded exclusively via JavaScript is not read by most AI bots. The mechanism is detailed in the indexing of content by AI.
From Audit to Action Plan
The audit report is just one page long: the three rates per search engine, the list of queries by group, the ten most frequently cited competitor sources, and the technical issues found. Add to that the measurement of actual traffic coming from AI assistants, available in GA4 as referrals from chatgpt.com, perplexity.ai, or copilot.microsoft.com; the setup is described in GA4 and Search Console for Measuring GEO. The resulting action plan is then standard: fix the issues, rewrite the pages that address the disputed search queries, create the missing pages for the unaddressed search queries, and then run the analysis again in 8 weeks.
How GreenRed Can Help You
Instead of juggling multiple tools, GreenRed's GEO Tracking module brings these metrics together in a single dashboard, compares them over time, and shows you which actions to prioritize. You can try it for free—no credit card required—on the page Rates.
Frequently Asked Questions
How many queries are needed for a reliable GEO audit?
Between 20 and 30 queries for a single-business-line SME, up to 60 for a company with multiple offerings or locations. Beyond that, manual tracking becomes too time-consuming, and an automated tool is necessary. The key factors are the breakdown between recommendation, comparison, informational, and brand-related queries, as well as the stability of the data over time.
Why do the answers change from one passage to the next?
The models generate a different response with each run, and the associated web search does not always return the same pages. A single run is not meaningful; three runs per query and per search engine, averaged, provide a usable metric. Variability itself is a form of information: an unstable presence indicates a contested query.
Should Google's AI previews also be audited?
Yes, because they appear in an increasing proportion of informational searches in France and reduce the number of clicks on traditional search results. For each query, check whether a snippet appears, which pages it lists, and whether yours is among them. Search Console does not yet isolate them; manual checking is still necessary.
Does a GEO audit replace an SEO audit?
No, it complements it. AI engines rely on Google and Bing indexes: a poorly ranked page will rarely be cited there. The GEO audit includes measuring visibility in search results, analyzing competitor sources, and verifying AI bots—all of which are not part of a traditional SEO audit.