A hallucination is output from a generative model that is fluent, confident and wrong — invented statistics, non-existent citations, fabricated quotes or events that never occurred.
Hallucination arises from how these models work: they predict likely continuations of text rather than looking facts up. A plausible-sounding falsehood can therefore be generated as readily as a true statement, and the model has no internal mechanism for distinguishing them. Grounding and retrieval reduce but do not eliminate the problem.
For publishers the risk runs both ways. Publishing unverified model output can introduce fabricated facts into your own content, damaging credibility and, on YMYL topics, causing real harm. Models may also hallucinate about your brand, attributing products or claims to you that do not exist.
Verify every factual claim, statistic and citation in AI-assisted content before publishing — invented sources are among the most common failure modes. On YMYL topics require review by a qualified person. Publish clear, authoritative information about your own brand so retrieval systems have accurate material to draw on.
Want to learn more? Explore our SEO Glossary or check out our expert guides on SEO strategies.