When an executive asks ChatGPT, Gemini, Perplexity, or Copilot “which provider should I choose,” the assistant no longer returns a list of ten links: it cites two or three names and explains why. Being included in that answer—or not—has become a business issue. That is the purpose of GEO, or Generative Engine Optimization: ensuring that a company is understood, cited, and recommended by answer engines.
The first step is always the same: find out where you stand. Hence the proliferation of “free GEO audit” offers. Behind the same formula, however, lie two very different exercises: the automatic score produced by an online tool and the audit conducted by an expert. Both are useful, provided you know what each one measures.
What a GEO audit should measure
A comprehensive GEO audit answers five questions. Does the company appear in responses to the questions its customers actually ask? Which sources do AI systems use to answer in its industry: review sites, directories, comparison sites, the press, or competitors’ pages? Which competitors are recommended instead, and why? Is the site technically readable by AI crawlers: authorized access in the robots.txt file, content visible without JavaScript, structured data, and an llms.txt file? Finally, is the company’s identity consistent everywhere AIs look for it: its Google Business Profile, professional directories, Wikidata, and social media?
The online tool: a technical check in one minute
The principle is simple: enter your website address, and the tool returns a score within a few seconds. Most of these tools check technical signals: AI crawler access, structured-data markup, the presence of an llms.txt file, heading structure, and loading speed. Some also query an AI with a few generic questions.
This is useful for spotting an obvious blockage, such as a site unknowingly prohibiting access to AI crawlers or pages whose content appears only after JavaScript runs. It is free and immediate, and it provides an initial basis for discussion.
The limitations are nevertheless clear. A tool does not know the questions your customers ask: it tests generic formulations, whereas AI responses vary fortement depending on how the question is phrased, the model used, and the location. The scores are not comparable from one tool to another. Above all, a score does not tell you which sources AIs consult in your industry, which competitors they recommend, or where to start.
How to read a score without making a mistake
A score of 40 out of 100 does not mean that the company is invisible to AIs, just as a score of 80 does not guarantee that it is recommended. It only indicates whether certain technical conditions are met. The proper approach is to note the blocking issues identified by the tool, corriger them, and then rerun the analysis. A site whose AI crawlers are autorized, whose pages render without JavaScript, and whose structured data correctly describes the company is starting from a solid foundation. To find out what AIs actually answer your customers, you then need to ask them the real questions.
The expert audit: starting with your customers’ questions
An expert audit starts with a list of real questions, generally fifteen to twenty-five, developed with the company: discovery questions (“how to choose…”), comparison questions (“what is the best…”), and decision questions (“how much does… cost,” “who should I contact in…”). Each question is asked across several AIs, sometimes repeatedly to smooth out response variability. The expert records whether the company is cited, its position, the description used, and, above all, which sources are highlighted.
This matrix is supplemented by a technical audit, verification of the brand’s identity across the web, and a priorized action plan: pages to create or rewrite, listings to complete, third-party sources to target, and reviews to collect. This plan is what gives the exercise its value, because it transforms an assessment into a list of tasks.
The drawback is the timeframe—several days of work—and the cost: according to the pricing study published by La Fabrique du Net, a GEO audit commonly costs between €1,500 and €3,000.
The two approaches, point by point
Timeframe: a few seconds for the tool, a few days for the expert. Questions tested: generic for the tool, drawn from your customers for the expert. Sources cited by AIs: rarely recorded by the tool, central to the expert audit. Recommended competitors: absent from a score, detailed in an audit. Action plan: generic on one side, priorized and costed on the other.
Why some agencies offer an expert audit
Some agencies offer an expert audit without charging for it: this is their way of demonstrating their methodology before proposing ongoing support. There is nothing worng with that, provided you verify what is actually delivered. Five questions help you separate the serious offers from the rest. Which questions will be tested, and who chooses them? In which AIs, and how many times? Will you receive the sources cited by the AIs? Is the technical audit included? Are you committed to anything afterward?
One warning sign should make you walk away: a promise of a guaranteed citation in ChatGPT. No AI guarantees its answers, and a serious provider commits to actions, deliverables, and measurement—not to an outcome it cannot control.
An example of a free expert audit
The French agency One Online, founded in 2018 and based in Neuilly-sur-Seine, illustrates this model. One Online’s free GEO audit tests twenty customer questions across five AIs, records the sources cited and the competitors recommended, includes a technical audit of the site and a 90-day action plan, and then provides a video-call debrief, with no obligation to continue. The agency applies the method to its own site as well: according to the Microsoft Clarity data it shares, one-online.fr was cited 474 times in AI responses during the 30 days preceding October 5, 2026.
What to remember
Online tools and expert audits are not mutually exclusive. The former serves as a free, immediate initial technical check. The latter tells you what AIs actually answer your customers, which sources they rely on, and where to start. The right sequence is to run the tool to remove technical blockages, commission an audit based on your real questions, and then remeasure the same questions in the same AIs every month: this is the only way to know whether the actions taken are changing the responses.