When an AI answer cites you but gets you wrong

By the SEOCells team, OmCells R&D Labs LLP · Published 8 November 2026

An AI answer can cite your page and still misstate it: part of a claim, a claim your page never makes, or one it contradicts. That is a reputation risk. Vague pages invite the model to fill gaps, often wrongly. Write explicit, specific, checkable claims, one per short passage, and check what the answers say.

Key points

A citation is not an endorsement of accuracy. The link can be right while the sentence next to it is wrong.
Misstatements come in three kinds: partly supported, unsupported, and contradicted by your page.
Vague wording, scattered facts and out-of-date pages make wrong gap-filling more likely.
One clear, dated claim per short passage is easier to quote correctly.

How can an answer cite me and still be wrong?

An AI answer is written by a model that reads several sources and then writes in its own words. The link at the end of a sentence says “this source was used”, not “this sentence is exactly what the source says”. Research on generative search has found for some time that a fair share of cited statements are not fully supported by the sources they cite. We see it in our own scans too.

For a business, that can look like this. An online shoe shop's returns page says “free returns within 30 days on unworn shoes”. The answer says “free returns on all orders”, cites the shop, and a customer later argues about worn trainers. The citation was real; the claim was not.

What kinds of mistakes happen?

Three ways an answer can misstate a cited page
Kind	What it looks like	Example
Partly supported	The answer keeps part of your claim and drops a condition	“Free returns” without “within 30 days, unworn”
Unsupported	The answer says something your page never says	“Open on Sundays” when the page does not mention Sundays
Contradicted	The answer says the opposite of your page	“No parking” when the page describes the car park

Why does this matter so much?

Because the reader sees your name next to the claim. If an answer says your clinic offers a treatment you do not, or quotes last year's price, the disappointment lands on you, not on the AI. It can mean awkward phone calls, unhappy reviews, or in regulated fields such as health and finance, claims you would never make yourself.

Why do vague pages get misquoted more?

When a page is vague, the model has to fill the gap with something, and it reaches for what is typical for businesses like yours. “Competitive prices and flexible hours” invites the model to invent the prices and the hours. “Check-ups from £45; open Monday to Saturday, 8:00 to 18:00” leaves little room.

Scattered facts cause the same trouble. If your opening hours are on the contact page, the holiday hours in a news post from last December and the Saturday exception in a footnote, the model may stitch together a version that never existed.

How do you write claims that are hard to get wrong?

One claim per short passage. Put the condition in the same sentence as the claim.
Use names and numbers, not “we”, “it” and “competitive”.
Date anything that changes: prices, offers, opening hours, team members.
Keep one canonical page for each important fact, and link to it rather than repeating it in slightly different words.
Say what you do not do, when people often assume you do (“we do not offer same-day delivery”).
Remove or update old pages that contradict current facts, including old blog posts and PDFs.
Keep the same facts in your structured data, map listing and directory profiles.

Is it only my own site that matters here?

No. Answers often blend your page with what other sites say about you. An old directory entry with your previous address, or a review site that lists a service you stopped offering, can feed the same mistake. Once a year, search for your business name and look at the first few pages of results with fresh eyes. Anything out of date there is a likely source of trouble.

What can you do when an answer is already wrong?

There is rarely a form to fill in. What you can do is remove the confusion at the source. Find which page the answer cites, and check whether the wrong claim comes from your own vague or outdated wording, or from another site. Fix your page first. If a third-party listing is wrong, ask the site to correct it. Then check again in a few weeks, several times, since answers vary between runs.

Most assistants also have a feedback button on answers. Using it is worth the ten seconds, though there is no promise of what happens next.

How do you keep an eye on it?

Ask the questions customers ask about you, especially about price, availability and policies, and read the answers rather than just counting citations. A SEOCells Deep Scan checks whether answers that cite your page represent it faithfully and flags the ones that do not, so you know which pages to tighten first.

Questions answered

Why is ChatGPT giving wrong information about my business?

The model writes in its own words from several sources and may fill gaps with what is typical. Vague, scattered or outdated information on your site and elsewhere makes this more likely.

Can I ask an AI company to correct an answer?

Most assistants have a feedback option on answers, but there is no guaranteed correction process. The more reliable fix is to make the correct information clear and consistent at its sources.

Does a citation mean the AI checked the claim?

No. A citation shows which source was used, not that every sentence matches it. Cited statements are sometimes only partly supported, or not supported at all.

What kind of pages get misquoted most?

Pages with vague claims, facts spread across several places, or out-of-date information that contradicts newer pages.

Sources

Search engines and AI assistants change often; check the linked documentation for the latest details. Nothing in this post is a promise of rankings, citations or results.

Check whether AI answers that cite you say what your page actually says.

How the AEO Report works