How to Motivate Students to Use AI Tutors: Study Findings

How to Motivate Students to Use AI Tutors: Study Findings

Schools have started handing students AI reading tutors the way they once handed out workbooks, assuming access alone would get the job done. Two randomized trials of roughly 350 elementary students say otherwise. Just over 60% of independently assigned students in one district and 53% in the other ever logged on to a well-known AI reading platform at all, according to The 74.

Schools had built in two 30-minute weekly sessions. Actual independent use came out to just over two minutes a week in one district and just over five in the other, The 74 reported. Carly Robinson, the Stanford researcher who led the study, put it plainly: "having access to this AI tutor isn't the same as using it," she said, according to Chalkbeat.

This article looks at what the research actually shows: why students skip AI tutors even when time is set aside for them, whether adding a human to the mix fixes the problem, and what that means for how to motivate students to use AI tutors in a real classroom or after-school program. None of this settles whether AI reading tutors help kids read better. It's a diagnosis of an adoption problem that has to be solved before that question can even be tested.

The findings apply most directly to elementary and middle school educators, tutoring program coordinators, parents supervising at-home use, and district staff evaluating AI tutoring platforms. It's not a verdict on every AI education tool, and results from one reading platform in two districts shouldn't be treated as universal.

Why students skip AI tutors even with dedicated time

Researchers gave elementary students specific windows, in class or after school, to work with the AI reading platform over intervention periods lasting 14 to 31 weeks. A large share of students simply never logged in during that entire stretch, The 74 found.

Among the students who did log on, the pattern was brief and inconsistent rather than steady. Those who used the platform averaged 13.2 minutes a week in one district and 25.8 minutes in the other, but they only logged in for four to five weeks total out of a much longer intervention window, according to The 74. In other words, even the students who tried the tool mostly didn't stick with it.

Robinson connected this to a broader shift in how schools adopt technology: "Having these tools available, even if they're really good, doesn't necessarily mean they're going to get used if they're not being embedded into kids' learning experiences," she told The 74.

Something similar played out with Khan Academy's Khanmigo chatbot. Founder Sal Khan described its 2023 rollout as "a non-event" for many students, saying "they just didn't use it much," The 74 reported. Two different platforms, different districts, and the same result: giving students a login doesn't create a habit.

For schools evaluating a new AI tutoring tool, that's a specific question to bring to a vendor or an ed-tech coordinator before rollout: what does actual usage look like in comparable implementations, not just what the platform claims it can teach.

Can human support increase AI tutor engagement?

Stanford's researchers built a second layer into the study to test a common assumption: that pairing an AI tutor with a person to provide encouragement would close the usage gap. Some students worked with the platform alongside a trained adult, afterschool program staff in one district, and strong-reading middle schoolers acting as peer tutors in the other. Their job was to motivate students and troubleshoot problems, Chalkbeat reported.

The lift was real, but small. Human support raised weekly use by about one minute in the afterschool district and 4.4 minutes in the in-class district, moving average weekly time from roughly two to three minutes in one setting and from about five to nearly ten minutes in the other, according to Chalkbeat and The 74.

The more telling detail is what didn't move. Human support did not increase the odds that a student would log on in the first place, Chalkbeat found. It added a few minutes for students who were already showing up, but it didn't get new students in the door.

There was one bright spot: the number of stories completed each week rose 71% in one district and 80% in the other when a human was involved, The 74 reported. That's a meaningful jump in output among the sessions that did happen, though the research doesn't say why, and it shouldn't be read as evidence that human support made each minute more efficient. It's simply more completed work from students who were already engaging.

For schools weighing whether to assign a teacher, aide, or peer tutor to support AI tutor use, this data point matters for budgeting that role correctly. The evidence here doesn't support asking that person to focus on getting reluctant students to start; the Stanford intervention didn't move that number at all. What it does support is using that person to sustain and deepen engagement among students who are already logging on, which is a narrower and more realistic job description.

Why this study can't answer whether AI tutoring works

The platform provider stated that at least 30 minutes of weekly use was associated with reading improvement, a claim made by the company itself rather than an independently verified threshold, Chalkbeat noted. The study's own target was 60 minutes a week. Actual use ranged from about two to ten minutes depending on the district and condition, nowhere close to either number.

Neither district showed a statistically significant improvement in end-of-year reading scores, and there was no meaningful score difference between students who worked independently and those who had human support, according to both Chalkbeat and The 74.

Robinson was direct about what that does and doesn't mean: "We never really got close enough to the dosage needed to find out" whether the tool helps reading, she said, per Chalkbeat. The study also never compared students who used the platform against students who didn't use it at all, which limits what it can say about the tool's actual effect, Chalkbeat reported.

That distinction is worth holding onto whenever a new AI tutoring headline crosses your feed: low usage that prevents a fair test is a different finding than a tool proven not to work. Conflating the two is the easiest way to misread a study like this one.

Alex Sarlin, founder of the EdTech Insiders newsletter, framed the results as part of a familiar pattern rather than a one-off failure. The study, he said, "shines a light on several of the most persistent challenges in ed tech implementation: low usage rates that don't meet dosage recommendations, differential technology usage based on prior student achievement... and a faulty assumption that students will jump into new tools without structured guidance," per The 74.

How to motivate students to use AI tutors: start with participation data

The clearest lead on structured motivation comes from outside this particular study. A quasi-experimental study of 110 middle schoolers in a hybrid human-AI tutoring program, published last year, tested what happened when students set a weekly goal, earned a reward for hitting it, and got regular check-ins from a human tutor about their progress. Weekly practice time rose about 25%, and skills mastered per week rose about 40%, with the gains holding steady over the following weeks, according to the ERIC-published study.

That's a promising signal, not a proven fix for the Stanford population. The middle schoolers in that study weren't the same age group, and the design was quasi-experimental rather than a randomized trial, which means it can't rule out other explanations for the improvement as cleanly as a randomized study could.

There's also an equity dimension worth sitting with before assuming any fix will reach the students who need it most. In the Stanford trials, students who logged on independently tended to be higher-achieving and less likely to receive special education services, Chalkbeat and The 74 both reported. The 74's analysis of that pattern points to an uncomfortable possibility: the students who might gain the most from extra reading practice were among the least likely to get it. That's an interpretation of the data, not a settled causal finding, but it's a reason to check participation numbers by student group rather than assuming a platform is reaching everyone equally.

Robinson's own read on the broader problem lines up with a design-first response rather than a bigger-budget one: "The challenge isn't just building good AI tools," she said. "It's really getting students to use them, and that seems to take the same type of intentional design that we've learned matters with other ed tech interventions and tutoring," per Chalkbeat.

Not everyone thinks the answer is better design at all ages. One advocacy researcher cited by The 74 argued there's currently little evidence that generative-AI chatbot tutors meaningfully affect learning outcomes for elementary students, or that they're developmentally appropriate at that age. Her recommendation: keep these tools away from students through second grade entirely, and limit use in grades three through five to settings with significant human oversight and dedicated AI literacy instruction.

Given how thin the evidence base still is, the more responsible move for a school or after-school program isn't to roll out goal-setting and rewards at scale and hope it works. It's to pilot it and measure. A workable version of that test: pick a specific weekly goal for each student (minutes logged, stories completed, whichever the platform tracks), have an adult do a two-minute check-in on that goal once a week, and track two things side by side, log-in rates and minutes used, broken out by achievement level and special education status. Reassess after four to six weeks before deciding whether to expand the approach or scrap it.

For a parent supporting at-home use, the same logic scales down: set one specific weekly target with your child rather than an open "use this when you have time," and check in on it once, briefly, rather than monitoring every session.

What to check before adopting or continuing an AI tutor

Across two randomized trials of about 350 elementary students, access to an AI reading tutor didn't translate into meaningful use, and adding trained human support raised usage by only one to 4.4 minutes a week, far short of the 60-minute target the study was designed around. That doesn't prove AI reading tutors fail to help students. Usage never came close to a level that could test that question in the first place, a distinction Robinson emphasized directly.

What the evidence does point to is a design problem with a possible, still-unproven solution: structured goals, rewards, and check-ins moved the needle in a different, smaller study, and it's worth testing carefully rather than assuming it or dismissing it.

Before adopting or continuing an AI tutoring tool, ask the platform vendor or district technology office for actual usage data, not projected benefits, and confirm what dosage threshold any learning claim is actually based on. Then check whether the implementation plan includes structured goals, regular check-ins, and a way to track participation by student group, the elements the evidence points to, rather than treating scheduled access as sufficient on its own.