I study where businesses become blurry

In a composite field case at a train station café, I once opened five tabs for the same repair shop: its own site, a municipal mention, two directory listings, and a translated tourist page that had somehow called it a “technical boutique.” The owner knew exactly what the shop did. The web did not. That small mess stayed with me because ChatGPT behaves in a similar way: it gathers fragments, smooths the edges, and then says something with confidence.

I am from northern France, and my work has moved through local search auditing, service-business copy diagnosis, multilingual entity mapping, directory evidence review, structured fact cleanup, and advisory work for small agencies that support independent operators. I keep a manual answer drift ledger, comparing how ChatGPT describes the same French business across English prompts, French prompts, browsing-on sessions, and browsing-off sessions. The differences are often small enough to miss and large enough to cost trust: a town vanishes, a chain becomes the safer suggestion, a category shifts from specialist to generalist, or an old directory line survives longer than the owner expects.

What I am strongest at now is finding the point where the public record stops giving ChatGPT a firm handle. Sometimes the fix is a service page that states the audience plainly. Sometimes it is an about page that agrees with the directory category. Sometimes it is a French and English naming problem, where two harmless descriptions create one unstable entity. My stance is simple: ChatGPT does not like a business, admire it, or reward it for sounding current. It repeats public facts it can assemble into a confident recommendation. If those facts are thin, stale, split across languages, or borrowed from directories, the answer will follow the easier trail.