Hallucination

A hallucination is generated content that is unsupported, inaccurate, or invented while often remaining fluent and plausible.

Plan for this page

Understand one concept in plain language and connect it to nearby ideas.

You will leave with: It can involve fabricated facts, citations, quotations, events, explanations, or confident details. Causes and labels vary, but the practical issue is that generation optimizes plausible output rather than guaranteeing external support.
Time: 3–8 min
Before you begin: No prerequisite. Keep one real example in mind.
Do this now: Read the explanation, then compare the example with the common misunderstanding.

Prepared next stepLearn a verification strategy

HallucinationA hallucination is generated content that is unsupported, inaccurate, or invented while often remaining fluent and plausible.

Large language model
Grounding
Retrieval-augmented generation
Verification

Plain explanation

It can involve fabricated facts, citations, quotations, events, explanations, or confident details. Causes and labels vary, but the practical issue is that generation optimizes plausible output rather than guaranteeing external support.

Example

A chatbot gives a realistic journal title, author list, and link for a paper that does not exist.

Why it matters

Fluency can hide missing evidence, especially in legal, medical, financial, academic, or operational use. Verification effort should rise with the claim’s consequence and uncertainty.

Common misunderstanding

Misconception: hallucinations are easy to spot because they sound strange. Correction: they may be detailed, consistent, and professionally written.

Related concepts

Follow relationships, not only the alphabet

Large language modelA large language model, or LLM, is a parameter-rich model trained on language sequences to predict and generate tokens in context.GroundingGrounding connects a model’s output to selected external information, observations, rules, or evidence relevant to the task.Retrieval-augmented generationRetrieval-augmented generation, or RAG, retrieves relevant material and places it in context for a generative model to use.VerificationVerification is the process of checking whether a claim, output, action, identity, or result is adequately supported for its intended use.

Learn a verification strategy

Sources and scope

What supports this object

Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence ProfileU.S. National Institute of Standards and Technology

Uses the related term confabulation for confidently stated erroneous or false content; terminology and measurement vary, and not every model error is a hallucination.

Open source

Loading page content...

…