Skip to content
Citable

Definition

Hallucination

Also known as: Confabulation, fabrication

A hallucination is output from a language model that is fluent and confident but false or unsupported: an invented statistic, a citation to a paper that does not exist, a misattributed quote, or a plausible-sounding date. It arises because the model predicts likely text rather than checking facts, and it is the main failure that answer engines use retrieval and citations to contain.

The term was borrowed from psychology and became mainstream in 2023, when a New York federal court sanctioned lawyers who had filed a brief containing case citations that ChatGPT had invented. The fabricated citations looked exactly like real ones and were checkable in minutes.

Hallucination is not a bug that a patch removes. A 2025 paper from OpenAI researchers, Why Language Models Hallucinate, argues that it persists partly because training and benchmarks reward a confident guess over an admission of uncertainty. The engineering response is to change what the model is asked to do: an answer engine retrieves passages and generates from them — retrieval-augmented generation with grounding and visible citations, so that a reader can verify each claim against a source.

The defence is editorial, not technical: state the number, the unit, the date and the full entity name inside the same paragraph, and keep each answer self-contained. The method is in How to write answer-first content.

Frequently asked questions

Can the way I write cause a hallucination about my own content?

Yes. When a retrieved passage is vague — a number without a unit, a claim without a date, a product referred to as 'it' — the model fills the gap with whatever is statistically likely, and the invented detail is then attributed to you. Self-contained passages with explicit names, figures, units and dates give the model nothing to guess.

Guides that use this term

  • Content

    E-E-A-T for answer engines: authors, dates and sources that machines can verify

    E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) is Google's description of a reliable source, and it is not a ranking factor. For answer engines it translates into signals a machine can verify: a named author with a profile page and Person markup, visible published and updated dates that match the structured data, outbound links to primary sources with a citation list, first-hand evidence, and an About page that says who is responsible.

    8 min read

  • Content

    How to write answer-first content that LLMs can quote

    Answer-first content opens every page and every section with a self-contained answer of roughly 40–70 words, then explains, qualifies and expands. The first sentence names the main entity, each heading carries exactly one question, and no passage depends on a pronoun or a paragraph elsewhere. Written this way, a passage still makes sense after an answer engine splits the page into chunks and retrieves one of them in isolation.

    8 min read