I asked ChatGPT about my own company and it got the details wrong
Published · Updated
You cannot edit what ChatGPT says about you directly. There is no dashboard, no correction form, and anyone offering to "get you fixed on ChatGPT" for a fee is selling something that does not exist.
What you can change is what it reads. That is slower, it is less satisfying, and it is the only thing that works.
Why has it got us wrong?
Because it is assembling an answer from whatever it can find, and if what it can find is thin, old or contradictory, it fills the gaps.
Three causes cover most cases. Your own site does not state the fact clearly, so there is nothing authoritative to draw on. The information exists in several places and they disagree, an old address here, a former trading name there, so the engine picks one. Or the fact is simply absent everywhere and the model has generalised from businesses that look like yours.
That last one is the uncomfortable category, because it is the model reasoning rather than repeating. It is also the most fixable, because supplying the fact clearly usually resolves it.
What actually changes the answer?
Making the correct version easy to find, consistent everywhere, and machine-readable.
On your own site, state the fact plainly in ordinary prose, not only in a logo or a graphic. If your registered name, trading name, address or founding date matters, write it out somewhere a crawler will read it. Add structured data describing your organisation, so a machine is not inferring your company details from marketing copy.
Everywhere else, make it match. Companies House, your directory listings, LinkedIn, review platforms, trade bodies. Inconsistency across sources is a common cause of an engine picking the wrong version, and it is tedious rather than difficult to fix.
Where you are eligible, structured public data sources carry disproportionate weight. Wikidata is the obvious one and it is open to edit, though only if your business genuinely meets a notability bar. Do not fabricate one.
How long does it take?
Weeks for retrieval, much longer for anything baked into the model.
That distinction matters. When an engine searches live and cites sources, changes to those sources show up reasonably quickly. When it is answering from training data, you are waiting for a future model. You cannot tell from the outside which is happening on any given answer, so the sensible approach is to fix the sources and re-test rather than to wait.
How do I check whether it is actually fixed?
Ask again in a fresh session with no account signed in, and ask more than once.
Signed in, you get answers shaped by your own history, which will mislead you into thinking things are better than they are. And these systems are not deterministic, so a single good answer is not proof. Ask three or four times across a couple of engines before concluding anything.
When is this not worth chasing?
If it is a one-off oddity on a question nobody actually asks, leave it.
The test is whether a real buyer would plausibly ask that question during a real buying decision. "What does this company do" and "who owns it" matter. An obscure detail the engine garbled once, on a question no customer has ever asked, is not worth a project. Fix the things that would embarrass you in front of a prospect and let the rest go.
Where to start
If you want to know how widely the wrong version has spread, the AI Visibility Audit at the AI Visibility Audit page tests the queries your buyers actually use across four engines and shows you what each one says and where it got it. £995, delivered in five working days.
For background, our guide on how AI decides who to recommend explains the mechanics of source selection, and our guide to measuring AI search visibility gives you a method you can run yourself.
Frequently Asked Questions
Can I report an error to OpenAI?
There is a thumbs-down and feedback mechanism in the interface, and it is worth using, but treat it as a nudge rather than a correction. It does not update an answer on request.
Will adding a correction page to our site fix it?
On its own, rarely. A page saying "contrary to what you may have read" is weak. Stating the correct fact clearly in the natural place on your site is stronger.
Does this affect Google's AI Overviews too?
Often yes, since the underlying problem is usually inconsistent or absent source information, and that affects every engine drawing on it.
Is it worth paying someone to fix this?
Only if the wrong information is appearing on questions your buyers actually ask. Anyone who cannot show you that evidence first is guessing.
How often should we check?
Quarterly is enough for most businesses, unless something has changed, such as a rebrand, a move, or an acquisition.
Your next step
Take the free two-minute Quick Check for an AI readiness score, see what the £995 AI Visibility Audit covers, or book a free 45-minute discovery call. Definitions for the terms used here are in the AI glossary and the Claude glossary.
Loading the full page. If it doesn't load, JavaScript is required.