Hallucination
Definition
A hallucination is a false statement an AI model presents as fact, such as a made-up statistic, quote or source. It happens because language models are trained to produce likely-sounding text, and likely-sounding text is sometimes wrong.
By Ronan Liedmeier, fractional head of growth and course author · Published 17 October 2026 · Last updated 17 October 2026
Also called: AI hallucination, confabulation
A figure in a client deck or a clause in a contract needs checking before it leaves your hands.
How it works
A large language model writes by predicting likely next words. Patterns that repeat constantly in its training text, like grammar and well-known facts, come out right. Rare, specific facts, like one company's revenue or a person's birthday, can't be worked out from patterns, so the model fills the gap with something plausible. OpenAI's 2025 research adds a second cause: training and benchmarks mostly score right answers and give nothing for "I don't know", so models learn to guess.
The risk clusters around numbers, dates, citations and niche topics, and around anything after the model's knowledge cutoff.
Example
An analyst writing a compliance memo asks which article of an EU regulation covers data retention. The assistant names an article number with full confidence. She asks it to search the web and link the official text, opens the link, and finds the rule sits in a different article. The memo gets the correct reference.
Common mistake
Checking only when an answer feels off. Hallucinations are plausible by nature, so gut feel catches few of them. Let the stakes decide instead: anything you'll publish or sign gets a source you have actually opened and read, and a check that the figure is still current.
Learn it in AI Academy
- Fact-Checking & Spotting HallucinationsBeginner · 10 min
- How AI Actually Works (The 20% You Need)Beginner · 12 minFree
Related terms
- Knowledge cutoff
- A knowledge cutoff is the date after which an AI model has no information from its training. Anything that happened later, such as news, price changes or new product releases, is unknown to the model unless it searches the web or reads a document you give it.
- Large language model (LLM)
- A large language model (LLM) is an AI system trained on huge amounts of text to predict which words should come next. At scale, that one skill lets it answer questions, summarize, translate and write in a way that reads like a person wrote it.
- Retrieval-augmented generation (RAG)
- Retrieval-augmented generation (RAG) is a way of getting an AI model to answer from your own documents. A search step first finds the passages that match the question, then the model writes its answer using those passages.