AI hallucination is what happens when an AI answers with total confidence and perfectly smooth language, but the content is wrong or simply does not exist. The trouble is it gives no hint that it is making things up — the tone sounds exactly the same as when it gets everything right.
What is AI hallucination
The word here has nothing to do with human hallucinations. It means the moment an AI produces an answer that sounds perfectly reasonable but is not backed by real information — citing a book that does not exist, describing a historical event that never happened, or doing a calculation wrong while writing out steps that look entirely convincing.
Why is AI so confident when it is wrong
Because this kind of AI does not "know" an answer the way a person recalls a fact. It works by predicting the next word that naturally follows, based on huge amounts of language it has seen before. When there is no real information to draw on, it keeps predicting anyway, because that is the only task it was trained to do. There is no internal alarm that says "I genuinely do not know this", the way a person usually senses.
Examples children run into often
The one I see most often in class is code that runs but gives the wrong answer. A student asks AI to write a program that calculates something, the code has no errors and runs cleanly, but the number it produces is wrong, because the AI misunderstood the problem slightly. Children who believe "it ran, so it is correct" miss this every time. Another common one: ask AI to summarise a piece of Thai history and it slips in the wrong year or the wrong person's name, smoothly enough that nobody notices.
Teach your child to verify an AI answer in three steps
- Ask for the source — ask it back "where did you get that from". A vague answer, or a source that turns out not to exist, is a warning sign
- Check one fact yourself — pick the single most important point in the answer and look it up somewhere else. There is no need to verify every line
- Test code against a known answer — before trusting a program, run it on an input where you already know the correct result. If it does not match, something is wrong
Code that runs is not the same as code that is correct. It only means there was no error.
Play "catch the AI" with your child
I like using this game with students: "catch the AI out". Ask AI a question your family already knows the answer to — something about your own household, a subject just covered at school, or a sum your child can work out by hand — then compare it with what AI says. If AI gets it wrong, have your child work out why. This turns AI mistakes into something ordinary rather than frightening or magical, and makes verifying answers feel like a fun habit instead of a chore.
If your child already uses AI for homework, read our article on children using ChatGPT for homework. To see how this connects to maths as a subject, read our article on how coding and maths connect. To build the right foundation for understanding AI from the start, the what is AI, and how to use it well course is a good place to begin, and to read code closely enough to catch AI's mistakes, the Python course builds that skill further. See the full list on our courses page.
Frequently asked questions
Does hallucination happen with every AI tool
Yes, to varying degrees, in every text-generating AI. None of them guarantee a correct answer every time, so verification matters whichever one you use.
Can a young child understand this idea
Upper-primary children usually grasp it well with a simple example, like the catch-the-AI game in this article. There is no need for technical language.
If AI gets things wrong this often, should my child still use it
Yes. Learning to verify an answer is a skill they will need for life either way, and avoiding AI altogether does not build that skill.
