SearchestraHallucination Rate: When AI Invents Things About Your Brand
A hallucinated mention is one an AI fabricates: an association, attribute or characterization of your brand that has no basis in any underlying source. Hallucination rate is the frequency of these mentions as a share of total mentions in a query set. It is a Portrayal metric with brand-safety stakes, because a confident, false claim can misrepresent your brand to a buyer with no way for you to see it.
Hallucination versus factual inaccuracy
These look similar but trace to different causes. A hallucination is an AI platform error, the model invents something with no source. A factual inaccuracy is when the AI accurately reflects a source, but the source itself is wrong. The remedies differ: one is a model problem, the other a content problem.
Why it must be surfaced, not hidden
Hallucinated mentions inflate mention rate, distort share of voice and corrupt sentiment analysis. Beyond data integrity, they carry reputational risk. A responsible provider surfaces flagged hallucinations to you rather than silently excluding them, and reports the rate per engine rather than as a blended average, because rates vary significantly across models.
What to monitor
| Item | Why it matters |
|---|---|
| Per-engine rate | Hallucination rates differ across models |
| Trend over time | Model updates can shift rates in either direction |
| Flagged examples | You need to see the specific false claims |
| Detection method | Absence of a documented method is itself a red flag |
The Searchestra view
Searchestra treats accuracy signals as part of Portrayal and reads them alongside sentiment, because positive tone attached to a hallucinated claim is a risk, not a win. Monitoring hallucination trend matters because a single model update can materially change how your brand is described.
Hallucination rate is a brand-safety metric; demand per-engine reporting, trend monitoring and flagged examples, and never read it apart from sentiment.
Frequently asked questions
What is an AI hallucination about a brand?
A fabricated association, attribute or characterization with no basis in any source. It is a model error, distinct from a fact that is simply outdated or wrong at the source.
How is hallucination different from factual inaccuracy?
A hallucination has no source; a factual inaccuracy faithfully reflects a source that is itself wrong. They need different fixes.
Why report hallucination rate per engine?
Because rates vary significantly across models. A blended average hides that one engine may misrepresent your brand far more than another.