We set reports because they claim something specific: that a student did a piece of work, made sense of it, and can account for it. The report was never meant to be evidence of report-writing. Yet its predictable structure is exactly what an AI tool produces best. Nikolic et al. (2023) found AI tools could pass the introduction, method and reflection of a report, and a year later the same team found that, given a student’s own data, a newer model could produce tables and figures ready to paste into the final report (Nikolic et al., 2024).
That undoes a comfortable assumption. Giving each student their own data protects the doing, not the write-up. Nor does a more authentic topic help: Kofinas et al. (2025) found that authenticity made little difference to whether markers could spot AI use.
What an AI tool still can’t supply is the student’s account of what actually happened. So mark the decisions, not the format: weight interpretation and anomalies over structure, ask for one raw trace of the work, add a five-minute conversation on one result, and set fewer full reports. An environmental science vignette shows the redesign at work.
Reviews
There are no reviews yet.