The research in 60 seconds
Researchers followed 26 students and 13 teachers across five Norwegian secondary schools trialling an AI platform for formative feedback. Students liked how fast, specific and available the feedback was. Teachers found it saved time and helped some students write better. But the feedback was sometimes vague or hard to follow, and teachers struggled to use it well without solid planning across the term.
So what?
The real finding isn't that AI can generate feedback, that part's obvious by now. It's that the teaching around it still matters just as much as ever. The system in this study drew on curriculum goals, teacher planning, student work and expert-designed prompts, and teachers who'd planned their assessment coherently across the term got far more out of it. Students, meanwhile, mostly valued not having to wait for the teacher.
None of this proves AI improved learning. It's a small qualitative study, built mainly on interviews plus some unstructured classroom observation, looking at one platform rather than AI feedback as a category. The researchers themselves argue AI should complement teacher judgement, not replace it. The practical takeaway: treat AI feedback as an assessment-design problem first and a technology problem second.
For teachers
Give the AI a narrow job. Rather than asking for general comments on a whole piece, point it at one learning goal or success criterion. Students in the study valued feedback that was concrete and subject-specific; broader feedback often came across as vague or confusing.
Use it while the work can still change. Let students ask for feedback mid-draft, then have them revise something before moving on. What students valued wasn't more comments, it was comments fast enough to actually act on.
Keep the decision with the student. Ask them to decide whether to use, adapt or ignore each suggestion. The study found examples of feedback that was irrelevant, inconsistent or pitched too hard, so treating every AI suggestion as correct would undercut the whole point of formative assessment.
For school leaders
Start with the assessment planning, not the tool. Before rolling AI feedback out further, check whether learning goals, assessment points and feedback expectations are clear enough for the system to actually work with.
Bring real student work into professional learning. Have teachers compare AI-generated feedback against their own judgement directly. Teachers in the study rated the technology, but were firm that the teacher gets the final say.
Watch what happens to the time it frees up. Faster feedback only matters if that time goes somewhere useful. Several teachers used it to circulate more, catch students who needed support and spend more time on the parts of learning that need a human.
