School Districts Publish First Rigorous Results From AI Tutoring Pilots

The first large, independently evaluated results from AI tutoring in public schools arrived this week, and they resist both the hype and the backlash. Across 140 schools in eight districts, students using AI tutors gained the equivalent of two additional months of math learning on average — with effects ranging from substantial to zero depending on implementation.

The study, run by a university consortium with district cooperation, randomized access at the classroom level and tracked 38,000 students over a full school year. The headline average conceals the finding researchers consider most important: in classrooms where teachers integrated the tutor into instruction — assigning it deliberately and reviewing its dashboards — gains tripled the average. Where it was deployed as unsupervised screen time, gains vanished.

Reading results were weaker than math across every model and district, echoing decades of education-technology research. And the study documented real costs: teachers reported five to ten additional hours of setup and monitoring monthly, and districts spent $40 to $120 per student on licenses.

Privacy provisions held up better than critics feared. Participating districts required vendors to disable data retention for model training and submit to independent audits — contract terms the researchers published as a template, and which several state education departments have since adopted.

The consortium’s conclusion reads less like a verdict on artificial intelligence than a finding about schools: technology amplifies the instructional culture it lands in. Districts with strong math teaching got stronger; districts hoping software would substitute for it got a well-documented null result.