What is an AI hallucination?

“Hallucination” is a common name for an AI-generated claim that is false, unsupported, or made up. It might be a wrong date, an invented quote, a source that does not exist, or a confident explanation built around a false detail.

Many generative AI systems produce an answer by finding likely patterns in data and language. A likely-sounding next sentence is not the same thing as a checked fact. The system may also misunderstand your question, lack current information, or combine details that do not belong together.

Fluency is not proof. Good grammar, a calm tone, and a long list of details can make an answer feel reliable without making it true.

Notice the warning signs

Any AI-generated claim can be wrong, but these clues should make you slow down and check carefully.

Untraceable sources

A paper, book, court case, quote, or website cannot be found where the answer says it should be.

Suspicious precision

The answer gives exact percentages, dates, or names without showing where those details came from.

Changing answers

The core facts shift when you ask the same question in another way or request evidence.

Mismatch with context

The answer uses the wrong country, subject, time period, rules, or definition for your task.

A lack of warning signs does not prove an answer is correct. It only means the error may be harder to notice.

Check the claim, not the confidence

1. Pick out the claims

Break the answer into statements you could test. Focus first on claims that affect your conclusion, mark, money, safety, or another person.

2. Find the original source

Look for the organisation, research paper, official record, or first-hand document that should support the claim. An AI-generated citation is only a lead until you confirm that it exists and says what the answer claims.

3. Read around the evidence

Check the date, author, context, and limits. A real sentence can still be misleading when it is taken from the wrong year or separated from an important condition.

4. Compare independent sources

For important or disputed facts, compare more than one reliable source. Do not count several pages that all repeat the same original claim as independent confirmation.

5. Correct or remove the claim

If you cannot verify a detail, label it as uncertain, replace it with supported information, or leave it out.

A useful follow-up prompt

“Separate this answer into checkable claims. For each claim, tell me what kind of primary source could verify it. Do not invent a citation.”

Choose a source that fits

The right source depends on the claim. Start as close as you can to the people or organisation responsible for the information.

  • Rules and public services: use the responsible government or official body.
  • Research findings: read the original study and note its limits.
  • Product features: check the maker’s current documentation.
  • School tasks: use the sources and referencing rules your teacher requires.
  • News events: compare reputable reporting and check when the event happened.

Search summaries, social posts, and AI answers can help you find a source. They should not quietly replace the source.

Use extra care for high-stakes tasks

Health, legal, financial, safety, and crisis decisions need current, qualified help. An AI response can miss context that changes the right action. Use it to prepare questions or understand basic terms, then check with an appropriate professional or official service.

Your judgement stays in the loop. You are responsible for deciding whether the evidence is strong enough for the way you plan to use the answer.

Sources and further reading