Why does AI get facts about businesses wrong?
Three causes account for nearly everything. Stale sources: you moved, rebranded or changed prices, and old records still say otherwise. Conflicting records: your listings disagree with each other, so the engine picks a version or blends them badly. Entity confusion: a business with a similar name gets merged with yours, and you inherit its address, reviews or reputation. Each cause leaves a different trail, which is why diagnosis beats guessing.
How do you trace where the error comes from?
- Capture the wrong answer with a screenshot, noting the platform, the exact question, the language and the date.
- Ask the platform directly: "what sources say this?" Perplexity lists citations by default, and the other platforms often name their sources when asked.
- Search the wrong fact verbatim, in quotes, on regular search. The pages repeating it are your correction targets.
- Check your own properties first: old service pages, an outdated footer, a forgotten location page. Engines quote you against yourself more often than owners expect.
How do you correct the error at its source?
Start where you have control: your website, your structured data, your llms.txt file, your business profiles. State the correct fact plainly on a page engines can read. Then work outward through the directories and listings you found in step three, using each platform's edit or correction process. For articles and third-party coverage, a short polite correction request to the publisher works more often than owners assume, especially with the evidence attached.
Consistency is the multiplier: an engine that finds the corrected fact agreeing across six sources stops repeating the old one far sooner than if one lonely page changed.
Keep a simple correction log along the way: the wrong fact, every page carrying it, the date you fixed each, and the date the answer finally changed. The log turns a frustrating chase into a process, and it teaches you which sources actually move answers in your category.
What if the wrong answer persists after the fixes?
Give the correction time to be re-read: live-browsing answers can update within weeks, while answers from trained knowledge may lag until a model update. Keep retesting monthly with the same question, and use the platforms' feedback buttons on the wrong answer, which exist for exactly this case. If the error involves something serious, a confusion with another company or a claim that damages you, document everything with dated screenshots; the paper trail matters if you ever need to escalate. This is also why monitoring your AI visibility beats discovering errors from a confused customer. An AI Visibility Audit captures every current answer about your brand, wrong ones included, with the screenshots already organized.