Differentiation is one of those things every teacher knows they should do more of, and almost nobody has time to do properly. Writing three versions of the same task is three times the work for one lesson. This is the job AI is genuinely good at, provided you give it the right instructions, and provided you check what comes back.
The method below takes about ten minutes and produces something you can actually put in front of a class.
Start with a worksheet you already trust
Don't ask AI to invent the task. Ask it to adapt one. The difference matters: a worksheet you have already taught from carries your judgement about what the content should be, what the misconceptions are, and how the questions build. That judgement is the part AI cannot supply, and the part you don't want to outsource.
Pick something that worked but that you know left part of the class behind, or bored the top end. That is the worksheet worth tiering.
Name the three levels before you prompt
The most common mistake is asking for "an easier version and a harder version". You will get an easier version that just has fewer questions, and a harder version that is the same questions with bigger numbers. Neither is differentiation. Both are the same task with the volume changed.
Decide what actually varies. Usually it is one of these:
- Scaffolding: how much structure the student is given. Sentence starters, a worked example, a partially completed table.
- Cognitive demand: whether the student is recalling, applying, or evaluating. Same content, different thinking.
- Reading load: how much text stands between the student and the actual task.
- Openness: whether there is one right answer or a defensible range.
Choose one or two of these to vary, and hold everything else constant. If you vary all four at once you end up with three unrelated worksheets rather than three levels of one.
Three tiers of one task should feel like the same lesson. If a visitor could not tell which students were on which version, you have tiered it well.
Write the prompt properly
A vague prompt produces a generic worksheet. Be specific about the year level, the subject, what varies, and what must stay the same.
I teach Year 9 Science. Below is a worksheet on energy transfer that I already use. Produce three versions of it.
Keep the same six questions, the same context, and the same key vocabulary in all three versions. Vary only the scaffolding:
Version A: include a worked example before question 1, sentence starters for the written responses, and a labelled diagram.
Version B: the worksheet as written, no changes to structure.
Version C: remove the diagram, replace two recall questions with questions that ask students to justify or predict, and add one open question with no single right answer.
Do not simplify the science in Version A or make Version C longer. Return all three with the same question numbering so I can mark them together.
Notice what the prompt does. It states the year level and subject. It names exactly what varies and what must not. It forbids the two failure modes: dumbing down the content, and making the extension version merely longer. And it asks for consistent numbering, which is the difference between three worksheets you can mark side by side and three you cannot.
Check the three things AI usually gets wrong
Read the output before it goes anywhere near a classroom. In our experience three problems come up again and again.
1. The "support" version has quietly taught something different
Scaffolding is supposed to change the route, not the destination. AI often lowers the content along with the difficulty, so the students who most need to reach the same understanding are the ones taken somewhere else. Read Version A and ask: if a student completes this successfully, have they learned the same thing as a student who completed Version B?
2. The extension version is longer without being deeper
Extra questions are not extension. Check that Version C actually demands different thinking, and that a fast student is not simply being punished with more work.
3. The examples are not local
Models default to American contexts, imperial units, and northern-hemisphere seasons. Expect to correct spelling, currency, measurement, and any example that assumes a context your students don't have.
- Could all three versions be marked against the same criteria?
- Does the support version reach the same learning, by a different route?
- Does the extension version demand different thinking?
- Is every example, unit and spelling right for an Australian classroom?
- Would a student be able to tell which version they were given?
That last one is worth taking seriously. Students know. If the tiers are visibly labelled "easy", "medium" and "hard", the differentiation stops being a support and starts being a message. Give the versions neutral names, use consistent formatting, and hand them out without commentary.
What this actually saves
It does not save the thinking. You still decide what the lesson is for, which worksheet is worth tiering, what should vary, and whether the output is any good. That is most of the professional judgement and none of it moves.
What it saves is the typing: the mechanical work of producing three formatted documents from one. That is genuinely an hour of your evening, and getting it back is worth ten minutes of careful prompting.
The teachers who get the most out of this already know what good differentiation looks like, and are simply using a faster tool to produce it.