Home / Features / Hallucination detection
At every scan we check whether what an AI claims about you matches what your site actually says. An outdated price, a service you no longer offer, an area you do not cover: a red flag appears on the question concerned.
Check what is said about me 14 days, no card requiredWhen a search engine shows outdated information about you, the link to the source is visible. The user clicks, lands on your page, and sees the discrepancy for themselves. The error corrects itself in the reader's mind.
In an AI answer that mechanism disappears. The information is stated inside fluent text, without qualification, often without a link. The reader has no reason to doubt it, and you have no way of knowing it is circulating. A prospect can decide not to contact you because an AI quoted a price you stopped charging two years ago, and you will never hear about it.
The first problem with hallucinations is therefore not how serious they are, it is how invisible they are. Detection starts by making them visible.
For each question asked during a scan, we record the factual claims the AI makes about your brand: prices, services, areas covered, availability, product characteristics. Each one is then checked against what your own site states.
That choice of reference matters. An external database would be outdated by design, and human verification would not hold at scale. Your site is the only authoritative source on your brand, and it is the one AI platforms are supposed to consult. Comparing the two therefore measures the gap between what you publish and what circulates.
Not all inaccurate information is fixed the same way. The distinction completely changes what you have to do.
The AI is right.
The claim matches what your site states. Nothing to do, and it is useful information in itself: it confirms AI platforms are reading the correct version of your data.
The AI is quoting an old version of you.
The information does exist on your site, but it is no longer the reference value. An old price in a blog post, an archived offer, a page nobody ever updated.
This is the most common case, and the easiest to fix: the problem comes from your side.
The information exists nowhere.
The stated value appears on no page of your site. The AI inferred it from context, picked it up from a third party, or invented it.
This is the trickiest case, because the fix does not happen on your site but upstream, in the sources.
There is no extra report to open. In your question tracking, a red flag appears on the questions where inaccurate information was detected. You open it, you see what the AI claimed, on which platform, and what your site says.
The check runs at every scan. An error that disappears from one scan to the next confirms your fix was picked up, which is the only genuinely usable proof.
Let us be direct: you cannot ask an AI to correct itself. There is no rectification form, no guaranteed turnaround, no contact. Any tool promising otherwise is selling you something that does not exist.
What works sits upstream. For outdated information the fix is direct: update or unpublish the page carrying the old value, and make sure the reference value is easy to find and unambiguous on your site. AI platforms eventually reload.
For hallucinated information the work is on third-party sources: identify the ones carrying the error, get them corrected where possible, and above all make the correct information available somewhere AI platforms actually consult. It is the same mechanism as for visibility, applied to accuracy.
In both cases, measurement tells you whether it worked. That is what separates a managed correction from a hoped-for one.
A hallucination almost always comes from a source. These modules tell you which one.
Factual, verifiable claims the AI makes about your brand: prices, services offered, areas or countries covered, offer characteristics, availability details. Value judgements and subjective wording are out of scope, since they cannot be checked against any reference data.
Because they are not fixed in the same place. Outdated information exists on your site: you correct it yourself, in minutes. Hallucinated information exists nowhere on your side: the fix goes through third-party sources and takes far longer. Conflating the two wastes time on the easy cases and underestimates the hard ones.
No flag appears, which is good news in itself. Be careful not to read it as proof that no error exists anywhere: the check covers claims made in response to your tracked questions. An AI asked about a completely different angle could produce an error outside that scope. The more broadly your questions cover your activity, the more solid the coverage.
Allow 3 to 8 weeks, with variation between platforms. A fix on your own site is generally picked up faster than one obtained from a third party. An on-demand scan lets you check without waiting for the weekly scan.
No, and nobody can. There is no rectification procedure comparable to a search engine's. The only effective action is to make the correct information more available, clearer and more accessible than the error, both on your site and in the sources AI platforms consult.
Yes. It runs at every scan, across all your tracked questions and all three platforms, in the Solo, Consultant and Agency plans. No feature is locked behind a higher tier, only the volumes of projects and questions change.
Create your project, run a first scan, and find out whether the information circulating about your brand matches reality.
Check what is said about me 14 days, no card required, no commitmentPRODUCTS
RESOURCES
COMPANY
