AI search · Glossary
What is AI hallucination?
AI hallucination is when an AI model such as ChatGPT gives an answer that sounds confident but is false, such as a wrong date, an invented fact or a citation to a source that doesn't exist. OpenAI's help center describes hallucinations as responses that are not factually accurate.
Written by Noor KhanWhy AI hallucination matters
Hallucinations reach customers. When an assistant states a wrong price, a closure or a service a business doesn't offer, people act on it. In July 2026, Google's AI Overview told searchers a bakery in Ontario was closing permanently after confusing it with a business of the same name in Illinois, as CBC News reported.
How AI hallucination works
- ChatGPT answers from patterns in the data it was trained on, and OpenAI says this can produce incorrect or misleading answers, sometimes stated with confidence.
- OpenAI researchers showed in September 2025 that training and evaluation reward guessing over admitting uncertainty, and that models get facts wrong more often when those facts appear rarely in their training data.
- Web search reduces the problem but doesn't remove it: OpenAI says search results and citations can be incomplete, outdated or incorrect.
How to apply it
- Ask ChatGPT, Google's AI and Perplexity about your business with web search on, and open every source they cite.
- Correct the outdated or wrong pages and listings the answer relies on.
- State key facts plainly on your own site, one fact per sentence.
- Report wrong answers with each assistant's feedback button after the source is fixed.
Example
A clinic finds ChatGPT listing a service it stopped offering. The answer cites an old directory listing, so the clinic corrects the listing, adds a sentence about the change to its services page, and re-tests a few weeks later.
Common mistake
Treating the thumbs-down button as the fix. The feedback goes to the AI company, but answers can keep repeating the error while the wrong source is still online.
put this into practice