AI tutoring can safely and effectively support students: An exploratory RCT in UK classrooms

Summary

An exploratory 165-student trial retains message-by-message human supervision and uncertain transfer estimates.

Five UK schools compared static hints with interactive tutoring; tutoring sessions were randomized between humans and human-supervised LearnLM. Roughly 76% of drafted messages needed no substantive edits.

The next-topic advantage was estimated at 5.5 percentage points, with a 95% credible interval of −1.4 to +12.4. This supports further study, not established autonomous superiority or long-term transfer.

Vendor-involved exploratory study; short-term platform outcomes and message-level supervision, without long-term independent testing.