Hallucination Watch
Somewhere today, an AI may tell a prospect something false about your brand: an old price, a plan you no longer sell. Hallucination Watch checks every collected answer against the facts you recorded and flags each problem, with the answer it came from.
The 2 kinds of flag
| Flag | What it means |
|---|---|
| Contradicts your fact sheet | The answer gives a price, a number, or a yes-or-no answer that differs from the one you recorded. |
| Answered without reading your site | The answer stated one of your facts, but the model looked nothing up first. It spoke from what it learned earlier, which is how old information survives. |
Each flag carries the severity you gave the fact: Critical or Warning.
What to do with one
- Read both sides. Decide whether the AI is wrong.
- If it is wrong and it matters, fix the source: make the correct fact obvious on your own website, in the words a buyer would use. That is what the models read.
- If the AI was close enough (different wording for the same thing), add that wording to the fact’s aliases, so it stops flagging.
- If a flag is not useful, dismiss it. It goes quiet but is kept.
If the page is emptyNo flags is good news, but check you have facts recorded at all. With an empty fact sheet there is nothing to check against, and the page will always look clean.
How the check works
The comparison is a direct one: your recorded value against what the answer said. No AI judges another AI. Prices, numbers, and yes-or-no facts can be checked that way. Text and list facts cannot, so they never raise a contradiction. Silence beats a confident wrong opinion.
Next: Answer Drift →Privacy policy · Terms of service · Legal notice · Cookie choices