AI visibility audit: measuring your presence in ChatGPT, Perplexity and Google
Before producing anything, know who is cited in your place, on which questions, and why.
An AI visibility audit answers a simple question: when a prospect asks an AI for a recommendation in your field, are you cited, and if not, who is? Without this measurement, GEO comes down to intuitions. With it, every action has a target and an indicator.
Here is the method we apply, in six steps. It takes a spreadsheet and a few hours for a first assessment, or a tracking tool for continuous measurement.
Contents
Step 1: the question set
Everything starts from the real questions your prospects ask, phrased the way they ask an AI: in natural language, with context. We build between twenty and fifty of them, spread across four families.
Category: “which science communication agency in France?”, “who does marketing for biotechs?”.
Comparison: “difference between a generalist B2B agency and a science-specialized agency”, “best agencies for a medtech”.
Problem: “how do I explain a complex technology to investors?”, “how do I get visible on ChatGPT?”.
Brand: “what does DigiObs do?”, “reviews of such and such agency”, to check what the AI knows about you.
Step 2: the engines and the conditions
Each question is asked to ChatGPT with search enabled, to Perplexity, to Gemini, to Google in AI mode or with AI Overview, and to Claude with web search. From France, in French, with no conversation history. Answers vary from one session to the next: we ask each question three times and keep the frequency, not an isolated answer.
Step 3: the reading grid
Mention: is your company named in the answer?
Rank: is it cited first, in a list, or at the end of the answer?
Citation: is there a link to your site, and to which page?
Sentiment: is the description accurate, positive, neutral or outdated?
Competitors: who is cited in your place, and how many times?
Sources: which third-party pages does the engine cite to build its answer?
Step 4: the indicators
Three indicators are enough to steer. Share of voice: the proportion of answers that name you, by engine and by question family. Citation rate: the proportion of mentions that come with a link. Dominant sources: the list of sites the engines use most to answer in your field, with their frequency.
Step 5: reading the sources is finding the action plan
This is the most useful step. If the engines build their answers from agency directories, you need to be listed there with an up-to-date profile. If they cite the trade press, you need to obtain op-eds or mentions there. If they rely on comparisons or forums, you need to be present with verifiable facts. The GEO action plan follows directly from the list of dominant sources, not from a general recipe.
On our own case, this reading showed that Perplexity cited us thanks to specific pages, while ChatGPT relied on directories and third-party articles where we were absent. The resulting plan is different for each engine.
Step 6: tracking over time
A one-off audit gives a starting point; only a monthly measurement shows the effect of the actions. For our clients, we use a dedicated tracking tool that replays the question set on each engine and computes share of voice, sentiment and cited sources. A spreadsheet kept with discipline does the job for a first quarter.
What the audit delivers
- The question set, reusable for tracking.
- Share of voice by engine and by family, with the competitors cited.
- The list of dominant sources and, for each, the action to take.
- The corrections to make on the site: robots.txt, structured data, missing pages, direct answers.
- A measurement calendar.
Going further
Frequently asked questions
How many questions are needed for a reliable AI visibility audit?
Between twenty and fifty, spread across category, comparison, problem and brand questions, each asked three times per engine. Below that, the results depend too much on the variability of the answers.
Can the audit be done without a tool?
Yes, for a first assessment: a spreadsheet, the consumer engines and a few hours. Monthly tracking quickly becomes heavy by hand, which is where a tracking tool earns its place.
What if no AI cites my company?
That is the starting situation of most specialized companies. The list of sources the engines use to answer shows where to appear first; in parallel, the site must offer pages that directly answer the questions in the set.
Let's talk about your project
Leading an innovative research project? Looking for a partner who can bring your idea to life beyond the scientific community?
Which type of engagement suits you best?