Did the AI
get your facts right?
Paste an AI answer about your brand and the facts it should have gotten right. The tool checks each fact against the answer — a number or negation mismatch is a hallucination signal. Deterministic, explainable, no model in the loop.
What this checks
When an AI engine answers a question about your brand, it can hallucinate — state something plausible that happens to be wrong. The only way to catch it without trusting the same model to check itself is a deterministic comparison against your own facts. This tool runs the same check the IndexHalo FactBase uses in its GEO workspace: for each fact, it finds the answer sentences that share a subject phrase, then checks whether they agree on numbers and negation.
Two kinds of mismatch are flagged. A number mismatch means the answer cites a different figure than your fact (your fact says 2019, the answer says 2021). A negation mismatch means the answer affirms what your fact denies, or vice versa. Both are mechanical, explainable, and false-positive resistant — the check only fires when the fact and the sentence share a subject.
What this does and does not measure
The contradiction check is measured — the same answer and facts produce the same verdict every time, with the specific sentence and mismatch reason shown. What it does not do is monitor answers continuously across engines over time. That is the FactBase monitoring in the paid GEO workspace, which archives every answer and alerts you when a new one contradicts a fact.
Nothing you paste is stored, and no external API is called. The analysis runs entirely in the request.
Frequently asked questions
What does this tool actually check?+
You paste an AI answer about your brand and the facts it should have gotten right. The tool splits the answer into sentences, finds the ones that share a subject with each fact, and checks whether they agree on numbers and negation. A mismatch is a hallucination signal — the engine said something about your brand that your own facts contradict.
How is this different from asking an AI to check itself?+
An AI checking its own answer is a model evaluating a model — circular and unreliable. This check is deterministic: the same answer and facts produce the same verdict every time, with the specific sentence and the mismatch reason shown. No model in the loop.
What counts as a contradiction?+
Two kinds: a number mismatch (your fact says 2019, the answer says 2021) and a negation mismatch (your fact affirms something the answer denies, or vice versa). Both are mechanical, explainable, and false-positive resistant because the check only fires when the fact and the sentence share a subject phrase.
What if my fact isn't mentioned at all?+
That is reported as 'unmentioned' — the answer doesn't reference the subject of your fact. That is not a hallucination, but it is a signal that the engine is ignoring your source for that topic, which is a different problem worth knowing about.
Is anything stored?+
No. The analysis runs entirely in the request. Nothing you paste is stored or sent to a third party.