Hallucination, in AI, means the AI says something false or made-up as if it were true. The answer sounds confident and fluent, but the fact, quote, name or source behind it does not exist or is wrong.
Picture a student in an exam where a blank answer scores zero. When they do not know, they write a confident guess anyway, because guessing can only help. AI is often in the same position, so it guesses too, and the guess reads exactly like a real answer.
In a sentence: “The chatbot hallucinated a book that was never written, so now I open every source it gives me before I use it.”
What the word means
The plain point: a hallucination is a confident mistake, not a lie. The AI has no intention to deceive. It is producing the answer that fits, and sometimes the fitting answer is false.
NIST, the US government’s standards agency, prefers the word confabulation, and notes that “hallucination” is the everyday name for it.1
A 2025 research paper on why this happens describes the cause the same way the comparison above does: the model guesses when it is unsure.2
A simple example
The plain point: hallucinations look exactly like correct answers. That is what makes them risky.
- Made-up sources: you ask for three books on a topic. Two are real. The third has a believable title and a real author, but the book was never written.
- Wrong details: a birth date that is off by a few years, or a quote put in the wrong person’s mouth.
- Invented facts about you or your business: an opening time, a product you do not sell.
Confidence is not evidence. An AI sounds just as sure when it is wrong as when it is right.
Why it matters to you
The plain point: anything that matters, check before you use it.
- Check names, numbers, dates and quotes. A small wrong detail is easy to miss when the sentence around it reads well.
- Open every source it gives you. Make sure the page exists and actually says what the AI claims. How to check a citation an AI gave you takes one minute.
- Be extra careful with health, money and law. NIST gives the example of a wrong summary of patient records leading to a wrong diagnosis.3
- Give it the facts. When you paste in the document, the AI can answer from it instead of from memory. That approach is called RAG.
Where the word comes from
Computer scientists were using the word long before chatbots. Dictionary.com, which made hallucinate its Word of the Year for 2023, traces one of the first documented uses in computing to “a 1971 research paper on training computers to accurately ‘read’ handwriting”, and says the term “began to appear in the context of machine learning and AI by the 1990s.”4 The same page notes that the AI sense describes not how AI works but “one of the ways it can malfunction”.
Often confused with
- Lying: lying means knowing the truth and hiding it. The AI has no such intent.
- Being out of date: an AI that does not know about last week’s news is not hallucinating; it simply was not trained on it. It becomes a hallucination when it invents an answer instead of saying it does not know.
- Bias: a lean in the answers, for example towards one point of view. Bias can be true-but-unfair; a hallucination is simply false.
Questions people ask
Why does AI hallucinate?
Mostly because it is rewarded for guessing. A 2025 research paper argues that “language models are optimized to be good test-takers, and guessing when uncertain improves test performance.”5 Why AI makes things up goes through the causes.
Can AI hallucinations be stopped?
They can be reduced, not stopped. Anthropic’s guide for Claude suggests giving the AI permission to say “I don’t know” and asking it to quote its sources, then warns that “while these techniques significantly reduce hallucinations, they don’t eliminate them entirely.”6
Do all chatbots hallucinate?
Yes, all of today’s chatbots can. The same Anthropic guide opens by saying that “Even the most advanced language models, like Claude, can sometimes generate text that is factually incorrect or inconsistent with the given context.”6 That is why the checks above matter whichever chatbot you use.
Learn more on this site
- The full explainer: Why AI makes things up covers the four causes and what actually reduces it.
- Checking the work: Auditing AI-generated work.
- The basics: AI explained for complete beginners and the AI glossary.
- Related: Asking AI About Your Health.
- Related: How to Make a Meal Plan With AI.
- Related: How to Plan a Trip With AI.
- Related: What Is a Vision-Language Model? VLM Meaning in Plain English.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What is an AI hallucination?
- The AI says something false or made-up as if it were true
- The AI refuses to answer a question
- The AI gives an answer that is out of date
Show the answer
Answer: The AI says something false or made-up as if it were true. A hallucination is a confident mistake: the fact, quote, name or source behind the answer does not exist or is wrong.
2. Why does AI often guess instead of saying it does not know?
- It wants to trick the person asking
- Like a student in an exam, guessing can help while a blank answer earns nothing
- It is copying answers from other users
Show the answer
Answer: Like a student in an exam, guessing can help while a blank answer earns nothing. The page compares it to an exam where a blank scores zero, so a confident guess seems the better bet.
3. What should you do with a source an AI gives you?
- Trust it if the answer sounds confident
- Use it only if the AI gives more than one
- Open it and check it exists and says what the AI claims
Show the answer
Answer: Open it and check it exists and says what the AI claims. Confidence is not evidence, so open every source and check names, numbers, dates and quotes before using them.
Related terms: Deep learning · Generative AI · GPT · LLM (large language model) · Machine learning · Narrow AI (weak AI)
- 1NIST, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1), July 2024, read at source 1 Oct 2026“The production of confidently stated but erroneous or false content (known colloquially as ‘hallucinations’ or ‘fabrications’)”. ↑
- 2Kalai et al., Why Language Models Hallucinate, arXiv:2509.04664, 4 Sep 2025, read at source 1 Oct 2026“large language models sometimes guess when uncertain, producing plausible yet incorrect statements instead of admitting uncertainty.” ↑
- 3NIST, NIST AI 600-1, July 2024, read at source 1 Oct 2026“a confabulated summary of patient information reports could cause doctors to make incorrect diagnoses”. ↑
- 4Dictionary.com, Dictionary.com’s 2023 Word of the Year Is…, 12 Dec 2023, read at source 12 Oct 2026“One of its first documented uses (in noun form) comes from a 1971 research paper on training computers to accurately ‘read’ handwriting and output it” and “The term, including the verb form, began to appear in the context of machine learning and AI by the 1990s.” ↑
- 5Kalai et al., Why Language Models Hallucinate, arXiv:2509.04664, 4 Sep 2025, read at source 12 Oct 2026“language models are optimized to be good test-takers, and guessing when uncertain improves test performance.” ↑
- 6Anthropic, Reduce hallucinations, undated documentation, read at source 12 Oct 2026“Explicitly give Claude permission to admit uncertainty. This simple technique can drastically reduce false information.” and “Remember, while these techniques significantly reduce hallucinations, they don’t eliminate them entirely.” ↑a“Even the most advanced language models, like Claude, can sometimes generate text that is factually incorrect or inconsistent with the given context.” ↑b
Kalai et al., Why Language Models Hallucinate, arXiv:2509.04664, 4 Sep 2025, read at source 1 Oct 2026.
The comparison on this page (the exam-taker who always guesses) is this site’s own way of explaining, though the paper above uses a similar image.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK