What Is a Hallucination? Definition and Example

A hallucination is a confident statement from an AI model that isn't true, or isn't supported by the sources it cites. Language models generate text from learned patterns, so they can produce plausible details, such as a wrong price, a feature that doesn't exist or an invented quote. For brands, hallucinations are an accuracy risk that GEO monitoring helps catch.

Hallucination vs a sourced error

It's worth separating two kinds of wrong answer, because the fixes differ.

Example

Illustrative. An assistant says a project management tool "includes built-in invoicing". There's no citation, and the vendor never offered invoicing. Likely cause: the model blended it with competitors that do. The fix is an explicit, crawlable features page and a comparison page that says what the product does and doesn't include.

How do you detect hallucinations about your brand?

1. Keep a fact sheet: pricing, plans, features, locations, founding facts, each with its official URL.
2. Ask branded questions in each engine monthly.
3. Compare answers line by line; for each error, open the citations.
4. Log whether the error is sourced or unsourced, and track it until fixed.

Why it matters for GEO

Grounding answers in retrieved sources reduces reliance on memory (Lewis et al.), but doesn't eliminate errors. Your best defence is making correct facts easy to find and quote, on your own site and on the third-party pages engines trust.

Common mistakes

• Assuming every error is a hallucination and missing the outdated source behind it.
• Fixing only your own site when the wrong fact lives on a third-party page.

Related

• Grounding
• Knowledge cutoff
• Training data