Differentiation with AI: the same topic for learners at different levels

The biggest obstacle to differentiation has rarely been skill. It has been time. Three versions of the same material have in practice meant three times the preparation, and the Finnish teachers’ union OAJ has noted that with scarce resources, differentiation often remains “a pious hope”. AI changes that equation: the versions take minutes, and the teacher’s time moves to where it is really needed, judging whether the levels fit and reviewing the result.

This guide covers what AI actually helps with in differentiation, what it needs to know, and what remains the teacher’s decision. The method itself, three skill levels from the same topic, is described in our guides on differentiation in language teaching and in every subject.

What AI helps with in differentiation

AI’s strength is varying the same content. When you have one good text or exercise, AI produces versions of it for learners at different levels: a lighter one with shorter sentences, vocabulary explained and intermediate steps visible, and a more demanding one where the questions ask for reasoning and application.

The same goes for the rest of differentiation’s manual work. Opening up vocabulary next to a text, adding guiding questions, splitting an exercise into steps and adapting a text for a language learner have all been possible before. What is new is that they take minutes instead of an evening at the kitchen table.

Upward too, not only as support

Differentiation with AI works in both directions. In a University of Turku study, the students who were furthest along often found their extra exercises to be at the same level as the basic ones, so they did not move their skills forward (Lahdenperä 2018). You can give AI a precise brief: make a version of this that requires reasoning, application to a new situation, or source-critical evaluation. The advanced version then lands at a genuinely different level, not just longer.

What AI needs to know

A plain “differentiate this” produces a generic result. A well-fitting level version needs context: the grade or skill level, the goal, what the group has already practiced and where it usually stumbles. The best starting point is your own material that has already worked with the group. There is a separate guide on giving context: how to make AI exercises fit your class.

What remains the teacher’s

Three things do not transfer to AI. The first is how well the levels fit: only a teacher who knows the group can see whether the lighter version is genuinely accessible and whether the demanding one challenges the right students. The second is review. The Finnish National Agency for Education’s AI recommendations state that staff evaluate AI outputs before they are used (Finnish National Agency for Education 2025). The third is data protection: students’ personal data or support needs do not belong in a general-purpose AI whose storage practices you do not know.

A timely reason to get this method in place is the learning support reform: as support is increasingly given within the student’s own group, differentiated material is the tool that support happens with in practice.

How Bruukki helps

Bruukki is built for this method. You bring in your own text or exercise set and get a suggestion for a lighter and a more demanding version, in languages along CEFR levels. Bruukki takes into account the goals of the curriculum you have chosen, so the versions stay anchored to the same core content. No student data is needed, and AI suggests, you decide who gets what.

The data-protection concern from the previous section is taken care of in Bruukki. Bruukki is a Finnish service: teachers’ and materials’ data is stored in the EU, and Bruukki does not use your content to train language models. Schools get the data-protection documentation, such as the data processing agreement, ready-made from us, so assembling it does not fall on the teacher or the principal.

Where to start

Pick one exercise set you are making this week anyway, and ask for a supported and an advanced version as described. Compare the suggestions against your group and adjust the levels. Once the method is familiar, a three-level set takes the same preparation as a single version used to.

If you want to try how your own material bends to different skill levels, try Bruukki for free. We are glad to receive teachers’ feedback, so tell us if something does not work.

Sources

• Finnish National Agency for Education (2025). Tekoälysuositukset varhaiskasvatukseen, opetukseen ja koulutukseen julkaistu. (in Finnish)
• Lahdenperä, E. (2018). Lahjakkaan oppilaan opetuksen eriyttäminen. Master’s thesis, University of Turku.
• “Lahjakkaat oppilaat tunneilla tyhjänpantteina? Oppilaiden väliset erot kasvaneet valtavasti.” MTV Uutiset, 22 Nov 2025 (in Finnish).

Read also: differentiation without double the work, in every subject too and the learning support reform: what it means for teaching materials.