Your business name has namesakes. Which one does the AI pick?

There are hundreds of bakeries called « Le Fournil » in France. Dozens of « Bistrot du Coin ». A « Garage Central » in almost every département. If you carry one of those names - or any name others share - a question follows: when a customer asks an AI « where is Le Fournil? », who is it talking about?

We asked this morning. Three deliberately ordinary business names, two questions each - « what can you tell me about… » and « where is it and what are its opening hours? » - put to all three assistants: ChatGPT, Gemini, Perplexity. Eighteen answers, collected on 9 August 2026, never naming a city. The finding: the three engines do entirely different jobs when facing a namesake. Two abstain. One elects.

What each engine did

ChatGPT (queried via API, without web search) named no city and gave no opening hours across its six answers. It explains what a bakery is, offers to narrow things down, and stops there. Gemini flags the ambiguity five times out of six - « several establishments carry this name » - and sometimes cites cities as examples, without ever pinning an address.

Perplexity, the only one of the three that reads the web while answering, does something else. On all three « where is it? » questions, it named one specific establishment, full address included.

« The most likely bakery Le Fournil, based on the results, is the one at 279 av. Paris, 79000 Niort. » (our translation)

Among hundreds of French « Fournil », the one in Niort - a town of 60,000 - was elected, opening hours attached. Same for the bistro, tied to 73 avenue Ledru-Rollin in Paris, and the garage, placed in La Chapelle-en-Serval. To its credit, Perplexity then adds an honest line: « if you meant a different one… ». But the first paragraph is what a voice assistant reads aloud, and what the eye keeps.

Why the engine that helps is also the one that decides

That is the paradox: the most useful engine - the one that fetches a real address rather than serving a generality - is also the one that must choose. And its choice depends neither on how close the customer is, nor on how good your bread is. It depends on sources: whichever namesake has the most complete, most cited, most consistent presence wins the answer.

Specialists have a name for this, as of this year: entity disambiguation. The principle is simple to state - faced with several candidates sharing a name, the model keeps the one whose signals are strongest in what it has read. In other words: your name does not identify you. The bundle of consistent information attached to that name does.

We live it ourselves, and may as well say so. « Waseit » is also an Arabic word (وسيط, « intermediary ») carried by long-established sites. Measured on Google this morning: our site comes first in France and Brazil, and does not appear on page one in the United Kingdom, where those namesakes are better established. The same mechanism, applied to us.

What should you do tomorrow morning?

• Never let your name travel alone. Everywhere you write it - website, Google profile, directories, socials, email signature - attach the trade and the town: « Le Fournil, bakery in Niort », not « Le Fournil ». It is the cheapest and most effective fix there is.
• Write the same thing everywhere. An abbreviated address here, a stale opening time there: every inconsistency weakens your bundle and strengthens your namesake's.
• Link your accounts to each other. The site points to your profiles, the profiles point back to the site: that reciprocity is exactly what lets a machine know which « Fournil » you are.
• Check what is being said under your name. Our free audit queries the assistants in 60 seconds, no account - and shows you whether they are talking about you, or about someone who resembles you.

One honest note to close: we are not claiming these answers are false. The Niort « Fournil » probably exists, and its listing is likely accurate. What the measurement shows is the mechanism - among several legitimate businesses sharing a name, one was designated, on the strength of its sources. Eighteen answers, a dated snapshot, three engines: not a census. But if your business carries a common name, the question deserves asking before a customer asks it in your place.