How AI answers are made
AI hallucination
Also called: Hallucination
An AI hallucination is an answer that sounds confident and believable but is wrong or not supported by any source.
In plain words
Language models write the most likely next words. Sometimes the most likely-sounding answer is simply not true.
A 2025 research paper, "Why Language Models Hallucinate", explains why. Training and testing reward a model for guessing, instead of for admitting that it is not sure.
Example
Imagine asking an AI tool which payment methods a small web shop accepts. If it finds no clear answer, it may still list a few common ones. The list sounds right, but it is a guess.
Why it matters for AI search
When an AI tool talks about your business, it can get your prices, address or features wrong.
Clear, checkable facts on your own pages give a grounded system something real to quote instead of something to guess.
What you can do
- State key facts in plain text on your own site: prices, opening hours, locations and product names.
- Keep one clear, up-to-date page per important fact instead of several pages that disagree.
- Ask AI tools about your brand now and then, and write down what they get wrong.
- Use the free AI prompt generator for ready-made questions to try.
Often confused with
- An outdated answer
- An answer based on old information is not always a hallucination. It can be right for the date of the model’s knowledge cutoff.
Sources
Facts and sources last checked on 19 October 2026.
Related terms
How does your page do in AI search?
GEO-Score checks one page on 22 signals, from AI crawler access to structured data, and shows what to fix first.
Analyze a page