When AI describes your company wrongly

An answer that describes your company wrongly is repeating what the material it can find says about you. None of the six systems takes corrections, so the fix is the material: your own pages first, then the directory entries, association listings and old coverage still carrying the outdated version.

This is usually the fastest problem in this subject to improve, and the reason is worth understanding. Where a company is absent, it has to earn a place in the answer. Where it is misdescribed, it already has a place and the wrong thing is sitting in it. Replacing wrong information is easier than winning attention.

Whether the description is right is Entity Accuracy, one of the things we measure, defined on Baseline.

Can you report it to them?

No, and it is the first thing everyone tries.

There is no form, no listing to claim and no account manager. This is not a directory with an entry you own — it is an answer composed from documents at the moment the question is asked. Change the documents and the next answer changes. Leave them and no amount of correspondence will help.

You cannot edit the answer. You can only edit what it is made from.

Where the wrong version comes from

Almost always from something you published, or allowed to stand, at some point.

A trade directory entry written when you joined in 2014. An association listing with a product range you have since narrowed. A press release about a contract you no longer service. A page on your own site, three clicks down, that nobody has read since it went live. None of it is malicious and all of it is retrievable.

Occasionally the cause is worse: you have been confused with a company of a similar name, and the description belongs to them. That is harder, and it is the case where structured data earns its keep, because an identifier tied to your registration number states plainly which company you are.

The four kinds of wrong

Out of date. Described by a product line you stopped making, or a market you left. The commonest by far, and the easiest to fix.

Wrong category. Placed among distributors when you manufacture, or among general suppliers when you specialise. This one costs you enquiries quietly, because you are excluded from the questions you would have won.

Wrong scale or reach. Described as a local supplier when you export to thirty countries, usually because nothing outside your own site says otherwise.

Confused with someone else. Rare and the most damaging, because the answer is confidently about a different company.

How to fix it, in order

Start with what you control and work outward.

Your own site first. One description of the company, on the about page, in the footer, in the schema. If three pages of yours disagree, nothing further along the chain can be fixed.

Then the listings you can edit. Directories, association pages, marketplace profiles, your own social and company profiles. Same words, not a paraphrase.

Then the ones you have to ask about. Trade bodies, distributor pages, old articles. This is slower because it runs on other people's schedules, and it is where most in-house attempts stop.

Then re-test. Ask the same questions again a month later and see whether the description has changed. Deciding what a machine must be able to verify about you, and making every source agree, is entity mapping, defined on Map.

How long it takes

Faster than the rest of this work, and still not instant.

Pages you edit have to be recrawled. Third-party corrections land when the third party gets to them. In our experience the description shifts before anything else does, which makes this the sensible thing to fix first — you get a visible result while the slower work is still running.

What it does not do is make you appear in answers you were absent from. Accuracy and presence are different readings. A correctly described company can still be named in two answers out of forty, which is a question of AI Visibility rather than accuracy, and set out in why your competitors appear in ChatGPT answers and you do not.

When it does not move

Three situations, all worth recognising early rather than late.

The wrong version is better corroborated than the right one — five sources say the old thing and only your website says the new one. The fix is more sources, not more edits.

The correction has not been recrawled yet, which is a waiting problem rather than a strategy problem.

Or the description is not actually wrong, only unflattering. A system summarising you as a mid-sized supplier of one product line is describing what the evidence supports. That is not an accuracy problem; it is a coverage problem, and it is fixed by publishing rather than correcting.

What if the information is damaging?

Then it is worth taking advice rather than tactics. Where content about your company is defamatory or infringes a right, that is a legal question with legal routes, and it belongs with your solicitor rather than with anyone selling search work — us included.

What we can say is that the practical fix and the legal one are separate. Correcting the sources changes what the answers say. It does not resolve whether something should have been published, and we would not pretend otherwise.

What this article cannot tell you

It cannot tell you which source is producing the wrong description in your case. The systems do not disclose their weighting, and identifying the likely culprits is investigation rather than knowledge.

It cannot promise every correction lands. Some third parties never update, and some old material outlives the company that published it.

And AI systems are probabilistic and change continuously. A description that is right this week may be assembled differently next month, which is why accuracy is something you re-test rather than something you finish. How answers are assembled at all is set out in how AI builds supplier shortlists.

What to do this week

Ask each of the six systems to describe your company, in a logged-out session, and write down what comes back. Then read it as a buyer would rather than as its owner.

If the description is wrong, list every place online that says the wrong thing, and start at the top of the list above. If you would rather have it measured properly — what all six say, how it differs, and which sources it appears to rest on — that is part of the AI Visibility Audit.