Teachomatic what to automate, and what not to

Teaching a Class to Use It

More than eight in ten students use AI for schoolwork. More than eight in ten say no teacher has ever shown them how.

Those two numbers together describe a strange situation: a skill in universal use, taught by nobody, learned entirely from peers and from the tool's own confident manner. Whatever your view on whether students should be using it, they are, and the current arrangement means they are learning to use it badly.

For a workplace-oriented comparison outside education, this resource offers another way to look at measurement, workload, behaviour, or accountability.

This is a lesson. One session, any subject, no equipment beyond what they already have.

What you are teaching

Not how to prompt. Prompting technique is the least durable part and they will pick it up anyway.

For an external perspective on student digital literacy, privacy, and learning, see Kahoot!.

You are teaching three things:

That the output is a claim, not an answer. The single most important idea, and the one the interface actively works against.

That it fails in specific, learnable ways. Not randomly. If you know the failure modes you can predict where to look.

That there are tasks where using it costs you something, and the cost is invisible at the time.

The lesson

Start with a failure, not an explanation. Ask it something in your subject where you know it is weak — a niche detail, a local fact, a recent event, an obscure text, a calculation with a trap. Do it live and do it badly on purpose.

What you want is a confident wrong answer. Confident wrong answers land in a way that no warning does, because the students can see that nothing in the presentation distinguished it from a correct one.

Then ask the class how they would have caught it. This is the lesson, and it has a method worth teaching separately. Their answers give you the checking strategies, from them: I'd look it up, I'd check the date, I'd ask someone, I know that's wrong because we did it last term.

Show the useful version. Same session — take a task where it genuinely helps in your subject. (What holds up in your own workload is a reasonable guide to what to demonstrate.) Re-explaining a concept. Generating practice questions. Producing counter-arguments for an essay. Students who have only been warned about it conclude adults do not understand it; showing a real use establishes that you do.

Give them one task and two rounds. Round one: they do it themselves. Round two: they ask the tool the same thing and compare. Then a short written answer to what did it get that you missed, and what did you get that it missed? That comparison does more work than any amount of instruction.

Finish with the categories. Which tasks are free, which need disclosure, and which are off-limits for this subject — and why. Ending with the rule after they have seen the tool succeed and fail makes the rule make sense.

The three failure modes to name

Give them vocabulary. Students who can name a failure will notice it.

Confident fabrication. It will invent a source, a quotation, a statistic, a date, a case. The invented one looks exactly like a real one. In subjects with citations, this is the failure that gets students in the most trouble, because a fabricated reference is checkable and looks like deliberate dishonesty.

Fluent averageness. It produces the median answer on any question. That is fine for a summary and disastrous for anything asking for a position, because the median answer is precisely what has no argument in it. Students should learn to recognise the smoothness of an answer that has avoided committing to anything.

Agreeing with you. Push back on a correct answer and it will often fold. This is worth demonstrating live, because it demolishes the impression of authority faster than any explanation. If it changes its answer because you frowned, it was never certain.

The uncomfortable part: naming the cost

Somewhere in the lesson, say plainly that some uses cost them something.

Not as a warning — as a mechanism. The difficulty of a task is often where the learning is. Struggling to construct a paragraph is how you learn to construct paragraphs. Removing the struggle removes the practice, and the removal is invisible because the finished paragraph looks fine.

Be honest that the long-term research on this is thin. We do not know how much this matters or for whom. Say that, rather than inventing a scary number — students can tell when they are being managed, and the argument is strong enough without exaggeration.

Then make it concrete for your subject: here is the task where using it will save you an hour and cost you the thing the exam actually tests.

Age changes the emphasis

The lesson works from about eleven upwards. What shifts is where the weight goes — the full age-band reasoning is here, including an honest note on how thin the evidence is.

Younger, roughly 11–13: most of the time on fabrication and checking. Concrete, verifiable failures. Keep it factual rather than philosophical.

Middle, roughly 14–16: the cost argument matters most, because this is where the temptation is highest and the assessments start to count.

Older, 17+: disclosure norms and professional practice, since they are close to contexts where these habits will be governed by someone else's rules.

What makes it fail

Doing it once and never again. Tools change and so does what students do with them. Repeat it briefly each term.

Only showing failures. They will conclude you are out of touch and disregard the rest.

Only showing successes. They will conclude the output can be trusted.

Letting it become a lesson about honesty. It is a lesson about how a tool works and where it breaks. The integrity conversation is a different one, and merging them makes students defensive at exactly the moment you need them curious.

The short version