Does ChatGPT have wrong information about your business? How to find out

When an assistant states something false about your business (a service you dropped, a price you never charged, an address you left two years ago), there is no "suggest an edit" for the answer itself. The claim came from somewhere, and the fix has to happen there. But the order matters, and the first step is the one most businesses skip: find out exactly what it is saying today, in writing, because "wrong" and "I heard from a customer that it said something weird" are different problems.

Why does ChatGPT say wrong things about my company?

Because it is quoting a source, not describing you. The pipeline repeats whatever published a parseable fact about your business — a directory entry, an old listing, someone else's comparison page — with the same confident tone it would use for your own site. The fix is upstream, at the source; the sections below show which sources those are, and which of them you can correct directly.

What wrong looks like when measured

In the freshly measured published sample, a med spa in Nashville, the assistants produced 14 outright wrong claims about the business — out of 333 checked: services it does not offer, attributes it does not have, even answers that could not confirm it exists. Alongside those, 53 price quotes — 36 matching the business's own published prices exactly, 17 with no price on the site to check them against — numbers assembled from third-party directories, even though 25 of the 26 pages on the business's own site display a price a human can read.

That is the typical shape of the problem: the assistant is not making things up out of spite. It is quoting whoever published the fact in a form it could parse — and that was usually not you.

The scam pattern to avoid first

There is a wave of outreach that targets exactly this worry: an email or contact-form message claiming "we found problems with how AI sees your website," urgent and unspecific, with a paid fix behind it. The tell is always the same: the findings are not shown. A real audit hands you the exact questions asked and the exact answers that came back, with sources; the sample report does this, and so does every paid report. Its price is stated, not discovered on a call. If someone's evidence for what AI says about you cannot be read, it does not exist.

What you can fix directly

Roughly in order of leverage:

1. Your own pages, in machine-readable form. Plain-text prices next to service names, Offer/PriceSpecification markup, current service lists, consistent name/address/phone. This is the source most likely to be cited for your business — make it the version you stand behind.
2. Your Google Business Profile. Hours, categories, services. Assistants and AI Overviews lean on it heavily for local questions.
3. llms.txt at your root. A small file that tells AI crawlers which of your pages are the canonical facts. We publish a working guide with real benchmark data.
4. The third-party directories you can edit. Stale aggregator pages are the hardest to fix and the most worth fixing — each one is a wrong answer waiting to be cited.

What you cannot fix directly, and how long it takes

You cannot correct the assistant, only the sources it reads. After a fix, propagation is not instant: it follows the crawl and refresh cycles of whichever engine cited the stale source — typically days to a few weeks. That is why the last step is not "fix" but re-check: ask the same questions again, compare against the written baseline, and see whether the wrong claims actually cleared. The before/after comparison, not the fix itself, is what tells you the work landed.

The first step, free

Before fixing anything, get the baseline: our free check reads your site and shows you the exact questions we'd ask — all 30, in full. No account, no card. The full measurement — 30 questions, five AI surfaces, every claim graded and sourced — follows the published method and costs $4.99 once.