Lesson 3 of 5 · about 5 min
Why AI makes things up, and how to catch it
When you finish, you can explain why hallucinations happen and use three habits that catch most of them.
Stuck? Ask the AI tutorA confident guess is still a guess
People call it a hallucination when an AI states something false as if it were true. It might invent a book, a quote, a court case, a link, or a statistic. The cause is built into how the tool works. It is designed to produce text that fits the pattern, and a believable-sounding detail fits the pattern even when it does not exist.
The model has no inner alarm that rings when it does not know. Often the tool will fill the gap with something plausible instead of saying “I am not sure.” You need to supply the alarm yourself.
Where errors hide most often
Mistakes cluster in certain places. Watch for specific names, exact numbers, dates, quotations, titles of sources, legal and medical details, and anything very recent or very local. The more specific and checkable the claim, the more it deserves a check.
Errors also grow when your question has a false idea inside it. If you ask, “Why did the city of Porto ban bicycles in 2019?” a poorly behaving tool may explain a ban that never happened. When your question assumes something, the tool may go along with it.
Three catching habits
First, ask for uncertainty. Try: “Tell me which parts of your answer you are least sure about.” This does not make the tool perfect, but it often surfaces weak spots. Second, ask for something checkable, such as the name of a source, then look that source up yourself. Third, cross-check any claim that matters with a trusted outside source.
A fourth habit helps with big jobs: split the task. Ask the tool to draft, then ask it in a new chat to critique the draft and list claims that need verification. Different chats give different samples, and disagreement is a useful signal.
A worked example from daily work
Diego, a high-school teacher in Medellín, asks for three quotes from famous scientists about curiosity for a classroom poster. The tool provides three beautiful quotes with names attached. Diego searches each one and finds that one is real, one is a loose paraphrase, and one cannot be found anywhere.
Because he checked, the poster uses only the real quote. A better prompt would have been: “Give me ideas about curiosity in your own words, and only include a direct quote if you are certain it is real and tell me where it appeared.” Even then, he should still check. The lesson is simple: verify exact words before you print them.
Practice now
- Ask an AI tool for “five books about [a hobby you know well].” Look up each title. Count how many are real, and notice how confident the tool sounded about all of them.
- Ask a question with a false idea inside it, such as a made-up event in your town. See whether the tool corrects you or goes along. Then ask: “Is there anything in my question that might be wrong?”
Use any AI assistant you have. Do not paste private, client or patient information into it.
Check yourself
1. Why do hallucinations happen?
2. Which type of claim most deserves a check?
3. What is a good way to use a second chat?
Ask the AI tutor
Ask anything about this lesson, in any language. The tutor is an AI: it can be wrong, and it never needs your personal details.
AI-drafted. Checked by an automated validator and a separate AI reviewer; a person has not read every lesson. Translations and tutor answers are AI-made. Found a mistake? Tell us. AI can be wrong. General information only, not professional advice. Not affiliated with any AI company named. Current University is not accredited.