What is an AI hallucination?
In one line
Ages 8 to 10: An AI hallucination is when an AI says something wrong or made up but sounds sure about it.
Ages 11 to 13: An AI writes the words that usually fit, so it can produce a sentence that sounds right but isn't true.
Ages 14 to 18: Hallucination (also called confabulation) is a natural result of generating statistically likely text; the model's confident tone is not a measure of its accuracy.
An AI hallucination is when an AI system states something false or made up as if it were true, such as a wrong fact, a person who doesn't exist or a fake source. It happens because the AI predicts likely-sounding words instead of checking facts. Always check important answers with a trusted source.
Last reviewed: 20 October 2026 6 min read
The idea in one picture
Imagine a friend who has heard thousands of book reports and is great at sounding like one. Ask them about a book they never read and they might still give a smooth, confident report, with a made-up ending. They are not trying to fool you; they are filling the gap with what usually sounds right. An AI hallucination is like that.
Worked example. Ask an AI, "Name a book about owls written for kids, with its author." It may give a real book. It may also give a title and author that sound perfect but do not exist together. The only way to know is to check: a library catalog, a bookstore or a trusted adult.
How it works, step by step
- You give the AI a prompt.
- A large language model predicts a likely next piece of text, called a token, based on patterns from its training.
- It adds that piece and predicts again, building a fluent answer.
- Nothing in this process checks the answer against the real world unless the system is connected to a reliable source, and even then it can slip.
- If the likely-sounding words are wrong, you get a confident, false answer: a hallucination.
A little deeper (ages 11 to 13)
"Hallucination" is a nickname. The AI does not see things the way a person hallucinates. Some experts prefer the word "confabulation", meaning confidently stated false content. Hallucinations are a particular risk with open-ended questions that need long answers, and in subjects that need specialized, context-specific knowledge. An AI can even invent sources or steps of reasoning that make a wrong answer look well supported.
Going further (ages 14 to 18)
Generative models produce outputs that approximate the statistical patterns of their training data. That can produce accurate text, and it can also produce text that is wrong or contradicts itself. Models are often poorly calibrated: how sure they sound does not reliably match how often they are right. Connecting a model to search or documents can help, but it does not guarantee correctness. The biggest risk is when people act on false content because it sounds confident, so important uses need human checking.
Common mix-ups
- Mix-up: "The AI is lying."Fact: Lying means meaning to deceive. The AI produces likely text without knowing whether it is true.
- Mix-up: "If it gives a source, it must be right."Fact: AI systems can invent sources too. Open and check the source itself.
- Mix-up: "Only bad AI hallucinates."Fact: Hallucination is a known risk for generative AI in general.
- Mix-up: "Asking again fixes it."Fact: A second answer can be different and still wrong. Check outside the AI.
What to do about it
- Use AI answers as a starting point, not the final word.
- Check important facts, names, numbers and quotes in a trusted source, such as a book, an official website or a teacher.
- If the AI gives a source, open it and see whether it really says that.
- For health, safety, money or schoolwork that will be graded, ask a trusted adult or the teacher.
Try it (no device, 5 minutes)
The confident guesser. One person is the "AI" and must answer every question in one confident sentence, even when they don't know. The others ask a mix: some easy ("What color is the sky on a clear day?"), some they know but the "AI" probably doesn't ("What did I eat for breakfast?"). After each answer, the group votes: true, false or can't tell? Then check. Which answers sounded the most sure? Were they the most correct?
Quick check
What is an AI hallucination?
Answer: A false or made-up answer that an AI states as if it were true.
A false or made-up answer that an AI states as if it were true.Why do AI systems hallucinate?
Answer: They predict likely-sounding words instead of checking facts.
They predict likely-sounding words instead of checking facts.(Ages 14 to 18) Does adding a source link make an AI answer reliable?
Answer: No. The source can be invented or misquoted; open it and check.
No. The source can be invented or misquoted; open it and check.Questions
An AI hallucination is when an AI states something false or made up as if it were true. Examples include a wrong date, a person who doesn't exist or a source that was never written.
Generative AI predicts likely-sounding text based on patterns in its training data, rather than checking facts. That usually produces sensible text but can also produce confident, false statements.
You often can't tell from the tone, because false answers can sound just as sure as true ones. Check important facts, names, numbers and sources in a trusted place, and open any source the AI gives.
No. Lying means meaning to deceive, and the AI has no such intention. It produces likely text without knowing whether it is true, which is why checking matters.
Connecting an AI to search or documents can reduce some errors, but no method guarantees correct answers. For anything important, a person should check the answer.
Mechna Academy is a city of machines for curious children aged 8 to 18, opening first to ages 8 to 10. In each mission, a child guesses how a machine decides, then opens its Peek Hatch to check, changes one thing and tests it on a case it has never seen. Children learn how AI works this way, and they never chat with an AI.
For grown-ups: join the waitlistPlaces are limited. The waitlist is for parents and guardians 18 and older.