AI in schools: only 20 solid causal studies, warns the US department's research laboratory

The Northeast & Islands OER has a very thin body of evidence on the real effect of AI tools on learning, and proposes a series of safeguards.

In a note published on August 10, 2026 by the Institute of Education Sciences, Jessica Mercer Young and Kirsten Peterson, from the REL Northeast & Islands Regional Laboratory, state a sober point: rigorous causal studies of the effect of current AI tools on learning remain extremely rare. Their review only identifies about 20 of them.

Three trends emerge from this limited corpus. The AI mediated by the teacher — i.e. integrated into an educational sequence designed and led by an adult — has positive effects. The tools put directly into the hands of the students give contrasting results. And the use of generalist assistants, not designed for the school, can be detrimental to learning.

The authors draw a series of recommendations: to use AI to support students' reflection rather than to replace it, to preserve the pedagogical relationship, to evaluate each tool on criteria of learning, confidentiality and accessibility, to teach critical thinking in response to the responses generated, and to invest in the training of educational teams.