AI Hallucination and Classroom Verification: A Lesson for Teachers and Students

Last updated: 2026-07-24

AI hallucination is when generative AI gives content that is fluent and confident yet does not match the facts. The trickiest thing about it is not that it gets things wrong, but that it does so without a flicker of doubt. To use AI in class, both teachers and students first have to learn this lesson: how to verify what it produces.

What is AI hallucination

AI hallucination is when generative AI gives content that is fluent and confident yet does not match the facts: made-up quotations, references that do not exist, the wrong person credited, miscalculated numbers. AI does not add 'I am not sure about this'; it writes a guess and a fact in the same tone, and a reader can hardly tell them apart at a glance. This is exactly what most needs guarding against when it is used in class.

Why it cannot be cured

To understand why hallucination is so hard to root out, you have to go back to how a large language model works. When it generates text, what it does is predict the next most likely word, aiming for an answer that resembles what a person would write rather than one that is correct. When it lacks reliable material, or the question involves obscure or very recent information, it will still produce an answer that looks reasonable rather than admit it does not know.

In other words, hallucination is not an occasional fault but an inherent feature of this technology. Tools will get things wrong less and less often, but as long as their nature is to predict text, the responsibility to verify will not go away.

A simple classroom verification routine

Rather than vaguely warning students to 'be careful with AI', give them a set of steps they can follow.

One, mark what can be checked: read the AI's output and circle every specific fact, figure, name, date and quotation, as these are the points that may be wrong and need verifying
Two, find independent sources: for each marked point, check it against sources other than AI, such as textbooks, official government or institutional sites, reliable news and original documents, and note the source
Three, decide what stays: keep what checks out, and cut or correct what does not or what has no evidence behind it, rather than letting it through just because it reads reasonably
Four, keep a record: hold on to the process of checking, not just the final answer, so that the verification itself becomes part of the learning

A fact-checking activity you can run straight away

Turn the routine into an activity and students learn faster. A simple design: the teacher picks a topic students know well (the school's history, a local landmark, a figure in the syllabus), asks students to put questions to AI about it, then uses the four steps above to check the answers. Because the topic is familiar, students find it easier to spot the flaws and are more sure of finding the right information. When a student sees with their own eyes that AI has the founding year of their own school wrong, that impression runs deeper than ten times of a teacher saying 'do not take AI on trust'.

A more advanced version can turn it around: have students use AI to generate a passage with errors deliberately mixed in, and hand it to their peers to check. Whoever sets the task has to know how to verify better than whoever answers it, and with that switch, the ability to verify goes from passive to active.

Verification is at the heart of AI literacy

The EDB's AI for Empowering Learning and Teaching Funding Programme asks schools to hold at least two activities that build students' AI literacy. What sets literacy apart is whether you can verify: being able to use AI is only the entry point, while being able to test what it produces is what counts as literacy. Fix the verification routine as a classroom habit, add one or two fact-checking activities, and you both answer the programme's requirement for student activities and lay a foundation for academic integrity. Once students are in the habit of checking before believing, they naturally bring a bit more caution to AI-generated work, information and images.

Common questions

Since AI gets things wrong, should students simply not be allowed to use it?

A ban is hard to enforce and misses an educational chance. Students will use AI outside school anyway, and the safer approach is to teach them to verify in class. Teaching openly that AI gets things wrong actually leads students to learn not to take it on trust, which is exactly the effect literacy education is meant to have.

Does verifying take too much time and slow the lesson down?

It is a little slower at first and quick once students are used to it. And verification is itself learning content: telling true information from false, tracing it to its source and judging reliability are abilities every subject wants to build anyway. Rather than treat it as an extra burden, treat it as a teaching goal met along the way.

Is there any way to avoid AI hallucination completely?

No. You can reduce hallucination by choosing newer tools, asking AI to give its sources, and putting questions in specific terms, but as long as the nature of generative AI is to predict text, zero errors cannot be guaranteed. Verification, then, is not a stopgap but a habit that is needed for the long term.

Sources

EDB Circular Memorandum 221/2025 (Chinese)

Want to know your school's next step?

Tell us where your school is with the funding programme. We help with evidence, from choosing tools to filing the reports.

Book a free school consultation Request a proposal