ChatGPT is wrong about your business

A customer forwards you a screenshot: ChatGPT just told them your company offers something you stopped doing years ago, sent them to the wrong location, or described you as a different business entirely. There is no button to correct it — but every wrong fact has an origin, and finding it takes five minutes. This guide runs in the useful order: the check first, then the diagnostic that tells you where the wrong fact came from, and only then the fix — which is never inside the chatbot, but in the sources it reads.

Almost everyone does the opposite. They see the error, panic, and rewrite half a page "for AI" without knowing where the wrong fact even came from. Sometimes it works by accident; almost always it's wasted effort. There is no button to edit what ChatGPT says about you — no editable listing the way Google Business Profile has one — but that doesn't mean you're stuck. Let's take it in order: first why it gets things wrong, then the diagnostic, and finally the right fix for your case.

There are two ChatGPTs, and only one reads your website

When ChatGPT answers you, the information can come from two very different places, and the difference changes everything.

The first is its training memory: the model learned from an enormous snapshot of the web taken months ago. If your price, your team or your description changed after that snapshot, the model repeats the old version because it's the only one it "remembers." It's not reading your website right now; it's reciting from memory.

The second is live search. Since OpenAI added search to ChatGPT, the assistant can query the web in real time, pull in current results and show you the sources it cited — per its own documentation, with links you can hover over or open in the "Sources" panel beneath the answer. That ChatGPT does read current pages. And it's the only one of the two you can influence quickly.

The whole strategy for correcting an error depends on which of the two failed you. That's why the diagnostic comes before any fix.

The two-minute diagnostic: with search and without

Open ChatGPT and ask the same question about your business two ways. First, as is, without asking for anything special. Then repeat it adding something like "search the web and tell me your source." Compare the two answers.

• Wrong without search but right with search: the error lives in the model's training memory, and the current web already has the correct fact. Good news: your job is mostly making sure live search always finds the right version, and waiting for retraining to catch up.
• Wrong both ways: the wrong fact is also on the web ChatGPT consults. Here the fix is concrete and in your hands: there's a public source saying something incorrect and it needs correcting.
• Right both ways: you may have seen an old answer, a cached one or a one-off. Test again in a few days before moving anything.

This two-minute test saves you weeks of optimizing the wrong thing. It's exactly the same logic serious guides use for measuring your mentions in ChatGPT: measure first, act second.

If the error also shows up with search: fix the sources

This is the most actionable case. There's public information — yours or someone else's — saying something incorrect, and ChatGPT is pulling it in. You attack it on three fronts, from the inside out.

1. Get your own website in order first

Your site is the source you control most. Check that the correct fact is written explicitly and literally, not implied. AI doesn't read between the lines: if your page says "founded a few years ago," it doesn't know the year; if it says "Founded in 2021," it does. Put the current price, the current address, the current description in plain text, on the page where it belongs (About, contact, product). Remove old pages with stale facts that are still indexed: each one is a source competing against the good one.

2. Align your external profiles

ChatGPT trusts what several sources confirm more than one page vouching for itself. If your website says one thing and your LinkedIn, your Google Business Profile, Crunchbase or your directories say another, you're handing the AI contradictory versions and it picks whichever it likes. Consistency of information across sources — the same address, the same name, the same description everywhere — is what the search world calls entity consistency, and it's half the battle. We unpack it in our guide to the brand SERP.

3. Check that OAI-SearchBot can get in

You can have a perfect website and still be invisible to ChatGPT's search because of one technical detail. The crawler that surfaces your site in ChatGPT's search answers is called OAI-SearchBot — distinct from GPTBot, the training one. According to OpenAI's documentation, if your robots.txt blocks OAI-SearchBot, your website can't appear as a cited source, and ChatGPT is left with whatever others say about you. It's a two-minute check that resolves more cases than you'd expect. The full bot-by-bot map is in our guide on blocking (or not) AI bots.

If the error lives in the model's memory: corroborate and wait

When the failure shows up without search but disappears with it, the wrong fact is buried in the model's training. There's no button to rewrite that memory, and there won't be one. What does work is twofold.

First, make sure live search always wins: the clearer, more consistent and better-linked the correct version is on the current web, the more likely ChatGPT pulls it in via search instead of reciting the old one from memory. Second, seed corroboration: media mentions, third-party profiles and fresh content repeating the correct fact. Every time the model gets retrained, it learns from a more recent snapshot of the web; if by then the correct version is the dominant one, the memory corrects itself over time. This is slow — measured in months, not days — and it's the same authority-and-trust logic that decides whether a model cites you at all: we cover it in E-E-A-T in generative search.

The thumbs-down: what it does and doesn't do

Under every ChatGPT answer there's a feedback button. When you flag an answer as incorrect and spell out the specific false fact, that report reaches OpenAI. It's not a magic button: it doesn't rewrite the answer on the spot or guarantee anything, and it doesn't replace fixing the sources. But it takes ten seconds, it's a direct signal to the provider and it has no downside. Do it on every platform where you see the error — Perplexity, Copilot and Gemini have similar mechanisms — and keep at the underlying work, which is what actually moves the needle.

Where to start tomorrow

If you did just one thing, it would be the diagnostic: take the three or four questions a customer would ask ChatGPT about you — price, location, what you do, who you get compared with — and ask them with search and without. Note where it fails and what kind of failure each one is. That five-minute map tells you exactly what to fix and in what order, and saves you from rewriting your entire website to correct a fact that actually lived in some forgotten directory. Fix your own sources first, then third parties', and give retraining time to handle the rest. Here too, consistency wins.

Frequently asked questions