When AI says something wrong about your brand

First, find out where the claim came from

Ask the assistant the question again and look at whether it cited sources. This single check decides everything that follows:

• It cited sources. Open them. The wrong claim is on one of those pages, and one of them is probably yours. This is the good case: it is fixable, and often within a day.
• It cited nothing. The answer came from the model’s memory of its training data, which is months or years old and cannot be edited by you, by us, or by the vendor on request.

Cause 1: your own page still says the old thing

The most common cause by a distance, and the most embarrassing. The engine is quoting you accurately; you are the one who is out of date. Check, in this order:

• Old pricing on a page you forgot, or a comparison page from two years ago.
• A blog post announcing something you have since discontinued.
• A “coming soon” that has been soon for eighteen months.
• Stale numbers in your Organization or Product markup, which nobody ever re-reads.
• A staging or duplicate subdomain that is still indexed.

Fix it by correcting the page, dating the correction visibly, and saying explicitly on the current page that the old thing is no longer true. An engine that reads two of your pages saying different things will pick one, and an explicit “this replaced that on [date]” resolves it in your favour.

Cause 2: a third-party page the engine trusts

A roundup listing your old price. A directory entry nobody has touched since 2023. A review of a product you have rebuilt. These are harder because you do not control them, but they are far from hopeless.

1. Identify the exact page. Not the domain — the URL the engine cited. Usually there are only one or two.
2. Email the author with the correction and a source. Publishers update factual errors far more often than people expect, particularly listicles that are re-published annually.
3. Claim the directory or profile listing. Most of these have an owner claim process nobody uses.
4. If the page will not change, out-publish it. Make your own page the most direct, most current answer to the exact question. A dated, specific page on your own domain competes well: in run A of our own audit, 74 of 90 citations went to pages whose URL carried a word from the question (75 of 90 in run B), and companies’ own sites were the large majority.

Cause 3: name confusion

The engine is describing somebody else. Another company with your name, a product with a similar one, or an acronym that means something else entirely in the market where the question was asked.

We have this problem ourselves and it is instructive. In the Philippines, “AEO” usually means Authorized Economic Operator — a customs-accreditation programme run by the Bureau of Customs. Ask an AEO question in a Philippine context and the sources returned are about customs brokerage, not marketing. That is not a bug in the engine; the customs meaning is genuinely the dominant one here.

What works against entity confusion:

• Be consistently disambiguated. Use the same full name in the same form everywhere — site, profiles, directories, markup. Half your web presence saying “Acme” and half “Acme Technologies Inc.” is two entities as far as a machine is concerned.
• Say what you are next to your name, in text, on your own pages. “Bungad AEO, an answer engine optimization subscription” does more work than any markup.
• Name the clash explicitly. A page that says “AEO also means Authorized Economic Operator; this is not that” gives a retrieval step something unambiguous to grab.
• Keep your sameAs links accurate across your profiles, so the entity is linked rather than inferred.

Cause 4: it is stale memory and there is no source to fix

No citations, and the claim is simply what the model absorbed. Three honest options, in order of how much they help:

1. Make the correct answer trivially retrievable. A clear, current, question-shaped page on your own domain gives the search step something better to find than the memory. This is the only lever with real leverage.
2. Use the vendor’s feedback control. Most assistants have a thumbs-down or report option. It is worth thirty seconds and it is not a fix — treat it as reporting a fault, not filing a correction.
3. Wait. Models are retrained. A claim from an old training set fades as newer material accumulates, provided the newer material exists — which loops back to option one.

What to do if the answer is defamatory rather than merely wrong

A different problem from visibility and outside what we do. Keep dated screenshots and the exact prompt that reproduces it, use the vendor’s reporting channel, and take legal advice. Do not let anyone sell you AEO as a remedy for it; the honest answer is that we cannot help with that and you should not buy a marketing subscription hoping it does.

How to check whether you fixed it

Re-ask the same question, in a fresh logged-out session, a week later and a month later, and record the sources each time. Do not ask a follow-up in the same conversation — the assistant is reading your correction from the chat, not from the web, and you will convince yourself of a fix that has not happened. The full method is on how to check whether AI mentions your brand.

Common questions