When AI gets your business wrong (and nobody tells you)

In short: There's now an AI layer talking to your customers about your business without asking you — and it gets things wrong more often than you'd think. Business-profile accuracy on AI platforms is just 68% [1]. One in three facts AI states about local businesses is wrong or out of date. You can't edit what ChatGPT answers, but you can edit the raw material it answers from.

A dog, a restaurant in Puebla, and a one-star review

A customer asks the AI whether she can bring her dog to a restaurant in Puebla. The AI says yes — confidently, in detail, as if it had spoken with the owner that morning. The restaurant doesn't allow pets.

The customer arrives with the dog. The server tells her no. There's an awkward conversation at the entrance, table reserved, dog tugging at the leash. She leaves annoyed, finds another place, and that night leaves a one-star review.

The review doesn't mention the AI. It mentions you.

That's what makes this problem different. You didn't make the mistake, you can't see it, and no alert reaches you. One day you simply start getting annoyed customers over promises you never made. The AI spoke in your name, got it wrong, and the bill landed at your door.

The figure that puts it to scale: 68%

This isn't an isolated case or an abstract fear. Whitespark — one of the most cited authorities in local search — documents these failures in AI-mode answers: stale data, invented attributes, one business's information attributed to another with a similar name [2].

And there's a number that puts it in scale. According to SOCi, a platform that analyzes the digital presence of multi-location businesses, business-profile accuracy on AI platforms is just 68% [1]. Translated: one in three facts AI states about local businesses is wrong or out of date. Not for some odd business. On average.

What changed: a layer that talks about you without asking

Your digital presence used to be what you published. Your site, your Google profile, your social channels. If a customer wanted your hours, they landed on a surface you controlled and saw what you put there.

Now there's a layer on top. When someone asks ChatGPT, Gemini or Perplexity about your business, the model doesn't send them to your page: it synthesizes. It gathers what it finds — wherever it finds it — and delivers it with a tone of certainty, in a single answer, usually with no visible sources. If it finds a 2023 schedule, an old menu in a PDF, or a similarly named business's data, that's what the customer hears.

The gap is brutal on one point: the customer who reaches your page sees the corrected notice, the "closed for remodeling" sign, the new menu. The one who asks the AI never passes through your page. The error travels straight from the answer to the decision. There's no intermediate surface where you can fix it.

And models don't invent out of malice: they fill gaps. When the available information is scarce, old or contradictory — one schedule on Google, another on Facebook, a third in a directory nobody has touched since 2021 — the model picks one or blends them. The root is almost always the same: inconsistent data across platforms and the absence of a structured, fresh source the model can prefer over the noise.

What you can control

You can't edit the AI's answer. But every error is born from raw material that is in your hands. Four concrete levers:

1. Make your information identical everywhere. A dental clinic in Oaxaca that stopped opening Saturdays a year ago but left "Sat 9-2" alive in an old directory is handing the AI two truths. The model will pick one, and not necessarily the right one. Outdated hours are among the most expensive errors: they send people to a closed door. Every cross-platform inconsistency is an invitation for the model to choose wrong.

2. Keep your data in a format machines read without interpreting. A restaurant in Veracruz whose menu and prices live only in a photo or a PDF is asking the AI to guess. Structured data — services, prices and attributes in fields, not in an image — reads with no margin for error. A photo of the menu is pretty for the customer and opaque for the machine.

3. Make your services say what you offer today. A spa in Tulum that stopped offering prenatal massage but never took it down across channels will get clients asking for exactly that. The reverse hurts just as much: if you added a new service and the AI doesn't know, it won't recommend you for what you actually do. Services you no longer offer — or never did — are among the errors that create the most friction at the door.

4. Give your data freshness signals. A real-estate agency in Querétaro with list prices from two years ago competes at a disadvantage: models cite stale content less. A fact with a recent date beats the 2021 PDF in the fight to be the source the model prefers. Freshness isn't cosmetic; it's a signal that the fact can be trusted.

There's a fifth front that sums up the rest: confusion between similarly named businesses. Two "San Rafael" clinics in Guadalajara, one downtown and one in Zapopan, blend easily in the AI's answer if neither has a structured source that clearly sets it apart. A verified listing, with unique and consistent data, is what tells the model which one is you.

The AI will answer anyway

This is the point that closes the loop, and it's worth saying plainly: the AI will answer about your business with or without your permission. It already is. The only open question is which data it answers with: yours — verified and fresh — or whatever it found lying around: the old PDF, the dead directory, the wrong business's data.

You don't need to control the answer. You need to control the raw material, and give the model a source it prefers over the noise. That's where Plexa comes in: a structured, verified listing with freshness signals, legible to AI crawlers by design. It doesn't replace your Google profile or your social channels; it puts a source on top of them that machines read without guessing.

Before fixing what the AI gets wrong, it's worth seeing what it says. Plexa's visibility report shows you in what fraction of answers you appear today and with what data. Claim your listing, verify it, and stop learning about your own errors from a one-star review.