Where the wrong answers come from
Two places. Stale evidence is the first: engines assemble answers from your website, directories, reviews and articles, and when those sources disagree or lag behind reality, the answer inherits the oldest surviving version of your facts. A menu you changed last year, a branch you closed, a price list from a forgotten aggregator page can all resurface as the current truth. Guessing is the second. When an engine lacks solid information, it sometimes fills gaps with what sounds plausible for a business like yours, a behavior researchers call hallucination. A clinic can end up listed with services it never offered. A restaurant can get credited with a dish that belongs to someone else.
The errors that cost real money
Some mistakes are cosmetic, and four kinds are not. Wrong prices set expectations you then have to walk back at the counter. A salon quoted at last year's rates ends up renegotiating on every booking call. Wrong opening hours and dead locations send customers to locked doors, for example during Ramadan hours you changed months ago, and they blame you rather than the tool. Wrong service descriptions filter out buyers who would have bought, silently. And identity mixing, where the engine blends you with a similarly named company, means you can inherit someone else's one-star reputation without a single unhappy customer of your own. Each of these reads as confident, neutral fact to the person asking.
Why you will not hear about it from customers
A shopper who reads that you closed does not call to verify, and a buyer told your service costs double simply shortlists someone else. The error lives inside a private chat you cannot see, so there is no complaint, no bad review, and no dip in any dashboard to alert you. Businesses usually discover the problem when a loyal customer mentions it in passing, months after the answers started steering people away. The only reliable early warning is checking what AI says about your brand yourself, on a schedule, in both languages your customers use.
What to do about it, starting this week
Run your customers' most common questions through ChatGPT, Google AI Overviews, Gemini and Perplexity and write down every factual error, however small. Then work the correction path for wrong AI information: fix the fact at its source, align your site and profiles so every listing agrees, and use the feedback tools the platforms provide. Most errors trace back to a stale page you can actually change, which makes this problem more fixable than it first appears. From there, monthly monitoring keeps new errors from settling in. That routine is a core part of AI visibility work: being described accurately, not just being mentioned.