Definition
An AI hallucination is an answer from an AI system that sounds confident but is false or not supported by its sources, such as an invented fact, quote, citation or product detail. It can be a made-up detail from memory or a faithful repeat of a page that was wrong.
Why do AI hallucinations happen?
AI hallucinations happen because language models are trained and scored in ways that reward a confident guess over admitting uncertainty. That is the core argument of the 2025 paper Why Language Models Hallucinate: when a test gives no credit for "I don't know", a model learns to guess.
Search adds a second way to get it wrong. When an AI answer is built from pages fetched live, a bad page produces a bad answer even if the model repeats it faithfully.
| Kind of wrong answer | Where it comes from | What you can do about it |
|---|---|---|
| From memory | The model fills a gap with a plausible guess and cites nothing | Make the correct facts about you easy to find and consistent across the web |
| From a bad source | The model faithfully repeats a page that was wrong, joking or out of date | Correct the source page, or publish a clearer one the engine can fetch instead |
Google gave a public example of the second kind in May 2024. It explained that an AI Overview recommending eating rocks traced back to satire republished on another site, and a glue-on-pizza answer traced back to forum content.
Why do AI hallucinations matter for your brand?
AI hallucinations matter because an assistant can state a wrong price, a discontinued product or an invented policy about your business to someone deciding whether to buy. It does not show up in your analytics, because nobody clicks through to correct it.
How do you check what AI says about you?
You check it by asking the assistants your customers use the questions they would ask about you, then comparing each answer against the facts. Note whether the answer cites a source. A wrong answer with a source points to a page to fix; a wrong answer without one points to a gap in what the web says about you.
Sources
- Kalai, Nachum, Vempala and Zhang, Why Language Models Hallucinate (arXiv 2509.04664, 2025)
- Google: AI Overviews, about last week (30 May 2024)
Last checked against these sources on 17 September 2026.