Students Offered a Free AI Tutor Scored About 4 Points Lower in the Same Course — and Barely Opened Their Course Materials
"Every child will get an AI tutor" sounds like good news. A new randomized trial suggests the question isn't whether a tutor is available, but what it does to the way a student studies once it is. In this one, the students who were offered the tutor did worse, not better.
The number to hold: 4.12 points. That is the grade gap, on a 100-point scale, the University of Maryland researchers estimated for students offered an AI tutor versus classmates in the same course who weren't.
Source: "The Effects of Course-Integrated AI Tutoring on Student Performance and Engagement: A Randomized University Trial," EdWorkingPaper 26-1598 (October 2026), as reported by EdTech Innovation Hub. It is a working paper, not yet peer-reviewed, and studied university students, not school-age children.
What the Study Does, and Doesn't, Say
We want to be straight with you about the limits, because this story is easy to oversell. The tutor was a study assistant built on GPT-4o and grounded in each course's own materials. Only about 15% of students offered it ever used it. Across the entire sample, the researchers did not find a statistically significant grade effect; the 4-point drop shows up when comparing sections within the same course. And the authors say the results describe this tool as set up, not every AI tutor.
Availability changes habits. Having an easy answer machine nearby can quietly replace the slower work of reading, practising and getting stuck.
Nobody was watching the habit. Instructors saw grades at the end. Parents see even less.
Younger students have less practice studying alone. If adults at university can slide, a 13-year-old with a free chatbot is likely at least as exposed.
"Offering a tutor and teaching a child to learn with one are not the same thing."
It Fits a Wider Pattern
Earlier this month a RAND survey found two-thirds of students said heavy AI use was hurting their own thinking, and most kept using it anyway. Other research has shown a large gap between what children do with AI for homework and what parents think they do. A tutor that quietly does the thinking is the same pattern, wearing a friendlier name.
Not every result points the same way: some other AI tutoring trials, with more structure built in, have reported gains. Design appears to decide the outcome, which is exactly why it is worth asking how your child's tool is set up.
What to Do This Week
A tutor is only as good as the habits around it. The gap in this study opened quietly, long before any grade showed it.
Start a free 14-day Blaick trial — no credit card required.
Blaick helps you see which AI tools your child uses and how often, so you can guide the habit before it becomes a crutch.
Start free trial →