Is AI feedback true but useless?

Summary

A large redrafting project makes uptake of feedback visible while leaving learning unresolved.

Nearly 9,000 Year 6 pupils in 211 schools wrote independently, then revised after mixed human–AI feedback. The company reports an eight-point gain, about 0.2 effect size; 64% scored higher.

The important questions concern understanding and acting on feedback. Improvement in a revised artifact neither isolates AI’s contribution nor demonstrates independent transfer.

A vendor before/after study of human–AI feedback, without randomization or delayed unaided tests.