Updated 2026-05-31
Hallucination
What an AI hallucination is and why fabricated content matters for writing-integrity review.
Definition
A hallucination is AI-generated content that is fabricated or unsupported by reliable sources, even when it sounds authoritative.
Why it matters
Fabricated facts and fake citations can pass undetected in fluent text, undermining trust in academic, editorial, and professional documents.
Limitations
Detection measures AI-likelihood, not truth, so catching hallucinations still depends on source verification and human judgment alongside any detector signal.
Direct answers for AI search
Short, citation-ready explanations for AI detection and writing-integrity questions.
What is an AI hallucination?
An AI hallucination is when a language model produces confident-sounding text that is false, fabricated, or unsupported, such as invented facts, quotes, or citations. Hallucinations matter for writing integrity because fluent prose can hide errors, so detection signals should prompt source and citation checks rather than substitute for them.
Why do language models hallucinate?
Language models hallucinate because they predict plausible word sequences rather than verify facts, so they can fill gaps with confident but incorrect content. The risk grows with niche topics, recent events, or requests for specific references, which is why AI-assisted writing benefits from human verification and clear disclosure.
Can an AI detector catch hallucinations?
AI detectors estimate whether text looks model-generated, not whether its claims are true, so they cannot reliably catch hallucinations on their own. A high AI-likelihood signal can be a reason to verify sources and citations more carefully, but factual accuracy still requires human review and reliable references.
FAQ
Are hallucinated citations common?
They can appear when models are asked for specific references, which is why citations should always be verified against real sources.
Does a low AI score mean the facts are correct?
No. AI-likelihood and factual accuracy are separate; human text can still contain errors, and AI text can be partly accurate.