What do I do when AI says something wrong about my business?
You can't force an AI assistant to retract an answer — there is no support line for ChatGPT being wrong about you. What you can do is fix the sources engines read: first capture the wrong answer verbatim with the date and the engine, then compare it against your verifiable, published facts, then publish the correct facts where machines actually look — a plain source-of-truth page, matching structured data, and a machine-readable Verified Claims Record on your own domain. Engines re-read; corrected sources are how wrong answers decay.
Why this happens
AI assistants answer from what they can crawl, and from older training data when they can't. If your site doesn't state a fact plainly — or states it differently across pages — the engine guesses, blends you with a similarly named business, or says nothing. When Nymrel ran this check on itself, AI returned no verified information about the company and attributed an unrelated person's biography to the founder. Both failures trace to sources, not malice.
The fix, in order
The repair path is the same whether the answer is wrong, missing, outdated, or someone else's facts wearing your name:
- Record the exact wrong answer, the engine, and the date — verbatim, not from memory.
- Write down the verifiable fact each answer should have matched, and where that fact is published.
- Publish a plain page that states the contested facts in visible text — not buried in a PDF or an image.
- Add structured data (Organization, Person, Service, FAQPage) that matches the visible text exactly.
- Publish a Verified Claims Record at /.well-known/verified-claims.json on your own domain, so agents read your facts directly. The format is open and documented.
- Re-check on a cadence — answers change when sources and models change, in both directions.
What a wrong answer costs
This is not hypothetical damage. In 2024 a Canadian tribunal held Air Canada responsible for wrong information its own website chatbot gave a customer. In May 2026 a German court issued a preliminary injunction treating Google's AI Overview as Google's own content, ordering it to stop repeating false AI-generated statements. Meanwhile Adobe Analytics measured AI-sourced retail traffic growing roughly 393% year-over-year in early 2026 — with those visitors converting about 42% better. Wrong answers now sit directly in the buying path.
Where to go from here
The free checks come first — run them before paying anyone, including us.
ChatGPT Recommendation Check (JalenBuilds Tools)
Free check from the same studio — see whether ChatGPT-style answers surface your business today.
AI Answer Accuracy Check — $299–$499
Nymrel runs the branded query set for you, delivers the dated discrepancy report and fix plan, and can publish the record. Optional monitoring from $99/mo.
Related questions
Can I get an AI engine to remove a wrong answer about my business?
Not directly — the engines don't offer a correction line for business facts. The reliable lever is the sources they read: publish the correct facts plainly on your own domain, in visible text and matching structured data, and re-check until the answer decays. Legal routes exist in extreme cases, but they are slow and jurisdiction-specific.
How do I find out what AI is saying about my business?
Ask the engines the questions your buyers would ask — your name, your service, your area — and record the answers verbatim with dates. A free recommendation check covers the quick version; a full accuracy check runs a structured branded query set across several engines and compares every answer against your verifiable facts.
What is a Verified Claims Record?
A machine-readable, dated, source-backed file published at a stable path on your own domain that lists your verified facts, so autonomous agents and AI answer engines can read the correct information instead of guessing. It fails closed: only facts backed by a source are marked verified. The open format is documented at nymrel.com/verified/methodology.
Will fixing my sources guarantee AI describes me correctly?
No — nobody can guarantee what an engine will say, and anyone promising that is overselling. Correct, consistent, machine-readable sources are the strongest input you control, and re-checking on a cadence catches regressions. Accuracy and machine-readability are what an honest vendor can deliver.