A working paper released June 17, 2026, by Carly Robinson at Stanford's SCALE Initiative asked a simple question: if you hand kids a free AI tutor, will they actually use it (Stanford SCALE, 2026)? Two school districts serving lower-income communities took part. One offered the tool after school, the other built it into class time, and some students got extra encouragement from a human tutor to log on while others were left to try it on their own.
The real finding was about usage, not learning. Kids using it after school logged, on average, about 2 minutes a week on their own, and about 3 minutes with a tutor cheering them on. In class, that rose to just over 5 minutes independently and nearly 10 minutes with support. When students did log on, sessions ran about 13 minutes after school and almost 26 minutes during class. Reading scores barely differed between groups either way. A lot of kids never opened the tool at all.