Hallucination
A fluent model output that is partly or wholly unsupported, wrong or invented.
Hallucination is a failure mode in which an AI system produces an answer that sounds plausible but is not adequately supported by the available evidence. In language models, this can mean invented facts, false citations, distorted summaries, made-up names or confident causal claims that do not follow from the source material.
The term is useful, but it should be used carefully. A model hallucination is not a human hallucination. The model is not perceiving, believing or imagining in a clinical sense. It is generating output from learned patterns, prompt context and probability. The problem is not that the model “sees something.” The problem is that the answer can look epistemically finished before it has been checked.
Why it matters
Hallucination matters because fluency can lower vigilance. A smooth answer can feel complete, especially when it uses the right tone, format and vocabulary. In everyday work, that can turn into wrong summaries, false references, weak decisions or misplaced trust.
The practical response is not panic. It is process: source checks, narrower prompts, better context, explicit uncertainty and human review when the output affects other people.
Human-AI angle
Hallucination is a test of the human side of AI. People do not only evaluate facts. They also respond to confidence, rhythm and apparent coherence. The more natural the answer feels, the easier it is to confuse presentation with reliability.
Related concepts
Sources and context
- Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023). Survey of hallucination in natural language generation. ACM Computing Surveys, 55(12), Article 248. https://doi.org/10.1145/3571730
- National Institute of Standards and Technology. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0). https://www.nist.gov/itl/ai-risk-management-framework
Related reading
Continue exploring
- TopicTrust in AI
A guide to sources, uncertainty, verification and calibrated reliance.
- ArticleTrust in the age of ready-made answers
A research-informed essay on what disappears when a fluent answer compresses sources, uncertainty, disagreement and responsibility.
- ConceptOverreliance
Relying on a system beyond what its performance or the situation warrants.
- ConceptAI literacy
Competence in understanding, evaluating and using AI with appropriate judgment.
- PracticeHow to check whether an AI answer has sources
A repeatable source check for claims that arrive without a visible evidence trail.
© 2026 Feliks Mamczur / Prompted Psyche. All rights reserved. Short quotations are permitted only with attribution to the author, the title and a link to the source.