Glossary
hallucination
In short: A hallucination is a confident but false statement produced by an AI model, such as a wrong price, address or service attributed to a business.
Hallucination is the term for an AI model producing fluent, plausible text that is not true. For a business this shows up as an assistant stating a price you do not charge, an address you left years ago, a service you do not offer, or merging your details with those of a similarly named company.
The causes are usually mundane. Training data has a cut-off, so older facts persist. Retrieval may surface a stale directory listing and the model repeats it. Or the model has no clear information about you and fills the gap by pattern-matching from similar businesses. Ambiguity (two trading names, inconsistent NAP data, prices stated differently on different pages) makes all three more likely.
Reducing hallucinations about your business is mostly a matter of removing ambiguity and feeding the engines current, corroborated facts:
- Keep name, address, phone and hours identical everywhere.
- Publish prices or a clear pricing model on a crawlable page.
- Use LocalBusiness or Organization schema with the same values.
- Fix third-party listings, which grounded answers often draw on.
Seoptist detects hallucinations by comparing claims in sampled answers against your fact card. Mismatches become accuracy issues and, on Growth and above, trigger alerts that show which engine said what.