Tutte le risorse
Brand e reputazione6 min di lettura

The day a chatbot mistook your brand for another

Two companies, one name: AI chatbots merge their identities into a single confident answer. Why it happens, what it costs, and how to make your brand identifiable.

The day a chatbot mistook your brand for another

A company is called Lumio. A French specialist in connected lighting, twelve employees, based in Nantes. One evening, its founder asks ChatGPT to tell her about her own company. The answer blends two worlds: her actual business, and the funding of a US company of the same name that sells video-conferencing software in Austin. Neither company has any link to the other. The chatbot has made up a third one, which exists only in its answer.

This case is invented, but the mechanism is not. With hundreds of thousands of trademarks filed worldwide every year, short, evocative, easy-to-pronounce names get recycled constantly. Two unrelated companies share the same name, in different countries, different sectors, sometimes decades apart. A traditional search engine copes reasonably well: it shows ten links, and the user sorts things out by reading titles and snippets. An AI chatbot writes a single answer, in the first person, with the assurance of an expert who has already checked their sources. Except it checks nothing of the kind. It assembles text fragments that are statistically close to the name typed, and presents them as a coherent whole.

The finding: a wrong answer, and no one to correct it

A negative Google review is visible. You can reply to it, report it, sometimes get it removed. A wrong ChatGPT answer about your company's identity, on the other hand, leaves no visible trace. It appears in no customer-review dashboard. It triggers no alert. A prospect asks the question once, gets an answer that mixes your business with another company's, and closes the conversation believing they learned something true.

The problem shows up mostly in three specific situations. First, direct homonymy: two brands, one name, different sectors. Second, phonetic or spelling proximity: two names that are written almost the same, which a chatbot treats as a spelling variant of one word rather than as two distinct entities. Finally, a rename or an acquisition: a company that has changed identity keeps, in everything chatbots learned before the switch, its old name attached to old information, sometimes inaccurate for a long time.

In all three cases, the result looks the same to the end user: a false identity record, delivered in the neutral, confident tone typical of these tools. Nobody doubts an answer that shows no doubt.

Why chatbots confuse brands

An AI chatbot does not consult an official company register before answering. During training it learned from a vast volume of text in which your brand appears perhaps a few dozen times, and in which another entity with a similar name appears thousands of times. When it has to answer a question about your name, it draws on the statistically densest space around that name. If that space is dominated by another company, older, more covered by media or simply better indexed, that company colours the answer.

Chatbots that fetch information from the web in real time, such as Perplexity or recent versions of ChatGPT and Gemini, add a second source of confusion. To answer quickly, they run a handful of searches around the name typed and aggregate the pages that come up. If your site never states in its content your exact sector, your city, your legal form or the year of creation, nothing allows the algorithm to decide between you and the homonym. It takes whatever looks most like a complete answer, whether it belongs to you or not.

One detail makes things worse for young brands. The longer an entity has existed on the web, the more pages, articles and mentions it has generated, and the more of the statistical space around its name it occupies. A two-year-old start-up with a name already used by a century-old company starts with a structural disadvantage. It is not a question of content quality. It is a question of mass.

The invisible cost of a wrong identification

The damage is hard to measure, and that is precisely what makes it dangerous. A prospect who asks a chatbot to compare your offer with a competitor's, and gets back information mixed with another company's, forms an opinion on a false basis. They will not come back to tell you your AI-generated identity record was inaccurate. They will simply move on, or worse, they will quote that false information in an internal meeting where nobody thinks to check it.

For a company that sells to other companies, the risk goes up a notch. A professional buyer preparing a supplier list increasingly asks a chatbot before picking up the phone. If they land on an answer that attributes to your company a funding round, a team size or a controversy that actually belong to a homonym, your file can be dismissed before the first conversation. No alert, no signal, just a line removed from a shortlist.

There is also the opposite scenario, just as awkward: your brand mistakenly inherits the merits of another company. That may seem flattering at first, but a prospect who discovers the gap between the promise and reality at the moment of sale immediately loses trust, and this time in both directions: towards the chatbot, and towards you.

Making yourself identifiable to a machine that verifies nothing

The good news is that this problem is fixed with the same levers as any clarity work on a site. The difference is that you have to use them with disambiguation explicitly in mind, and not only conversion or style.

First reflex: repeat, on the most visited pages and in the site's metadata, the elements that make your identity unique. Full legal name, precise sector of activity, city or country of operation, year of creation, name of the founder or founders. A chatbot that cross-checks these signals several times on your own domain has far more to work with than one that must guess from a name alone.

Second reflex, more direct: if a homonym exists and already causes documented confusion, a page that explains it clearly, citing both names and both sectors, paradoxically helps your own identification. It seems counter-intuitive to write about a competitor that is not one, but a well-built disambiguation page often becomes the clearest source on the subject, and therefore the one chatbots pick up.

Third reflex: monitor, not once but regularly, what the main chatbots answer when asked who you are. A brand confusion is not detected by intuition. It is detected by systematically asking ChatGPT, Gemini, Claude and Perplexity the question, comparing the answers, and spotting the details that do not fit: a wrong location, an activity that is not yours, a figure that visibly belongs to another company. It is cold, repetitive checking work, and it is exactly the kind of task better automated than left to an occasional manual check, which always comes too late, once the prospect has already closed the conversation.

Vurto has built in a safeguard that anchors each brand analysis in the real identity of the audited site, so that its own analyses do not fall into the same trap as the chatbots it monitors, and to spot more quickly the cases where a brand is overtaken by a homonym in AI answers.