Hallucination
A hallucination is a confident, fluent statement produced by an AI system that is factually wrong or unsupported by any source.
For businesses this manifests as invented pricing, discontinued services described as current, wrong locations, or attributes belonging to a similarly-named company.
The usual cause is an information gap. When authoritative current information is missing or unreadable, the model fills the space from whatever is nearest in its training — including stale or conflated material.
Why it matters
Hallucinations about your business are actively harmful and largely preventable. Clear, current, retrievable information is the fix, because it removes the gap the model was filling.
An example
An assistant describes a small consultancy as offering payroll services it stopped providing two years ago, and quotes a price it never charged. The consultancy's current site lists its services only inside a JavaScript carousel, so there was nothing current for the assistant to retrieve — it filled the gap from older and partly unrelated material.
Common mistakes
- Assuming hallucinations are random and therefore unfixable. Many trace back to missing or unreadable information.
- Correcting the wrong thing. Changing a page no crawler can read does not change what is retrieved.
Questions
- Why does ChatGPT say wrong things about my business?
- Usually because current, authoritative information about you is missing, inconsistent, or unreadable to its crawler, so the answer draws on older or conflated sources. Making accurate facts easy to retrieve is the most reliable correction.
- Can I report a hallucination about my business?
- Most assistants have a feedback control on each answer, and it is worth using. The more durable fix is on your side: clear, current, retrievable information, and consistent descriptions of you on independent sites.