When a response warrants investigation
Bring the exact question and response, with its date if available. The issue may concern a trading name or an establishment whose activity differs from the wider company. We first define what is wrong and what a correct description would need to say.
An absence from one answer can also be examined, but it does not demonstrate that the brand is absent from every search or conversation.
What we check
We agree a question panel and record the available context for each assistant. The requested country and language are kept distinct from any location setting the product actually supports. OpenAI documents how context can affect ChatGPT search.
For French business identity questions, we compare names and establishments with the information supplied and relevant public references. The official explanation of SIREN and SIRET helps distinguish those objects. It does not establish which source an assistant used unless the response makes that source observable.
What you receive
Each finding includes the retained response and the factual comparison, with a proposed action. We distinguish a correction on a channel you control from an issue that belongs to an external publisher.
The review explains which statements are demonstrably wrong and where the evidence remains incomplete. We cannot require an assistant to change a response or a third party to amend its page. A changed answer is not automatically attributed to a website correction.
Scope of the engagement
Delivery is available in English and French for the agreed products and questions. Recurring collection is separate: AI visibility tracking for France defines that service. The audit does not include a universal visibility score.
Send the description to investigate, or read our general GEO audit approach.