AI Reading Tutor Study: Stanford Finds Kids Barely Used It

The plan was an hour a week. The reality was closer to two minutes. In an AI reading tutor study led by Stanford’s SCALE Initiative and reported in June 2026, elementary schools set aside two 30-minute sessions a week for students to use a well-known AI reading platform. Across 355 children in two districts, roughly half never logged on at all, and average use came to just over 2 minutes a week in one district and about 5 in the other. Adding a human helper boosted engagement but did not move reading scores. If your child’s school just announced an AI tutor, this is the study to read first.

Key Takeaways

• Two randomized trials with 355 students in grades 1–5 found that about 60% (District A) and 53% (District B) of students assigned to use an AI reading tutor on their own ever logged on.
• Average weekly use was just over 2 minutes and about 5 minutes, against a schedule of 60; even students who logged on averaged 13 and 26 minutes a week.
• Pairing students with a human tutor raised stories completed by 71–80%, but reading achievement did not improve significantly in either district.
• Students who used the tool on their own tended to be higher-achieving and less likely to receive special education services, so the kids who needed practice most got the least.
• For parents: an AI tutor works only if a person builds the habit around it. Treat it like piano practice, not like a vending machine.

What the Stanford AI reading tutor study tested

The study is a working paper titled “Access Is Not Enough: Human Support Improves Engagement with AI Tutoring,” led by Carly Robinson, research director of the Stanford SCALE Initiative, and posted through the National Student Support Accelerator at Stanford. It has not yet been through peer review, which is normal for education working papers but worth knowing.

The design was two randomized controlled trials in two unnamed school districts. Students in grades 1 through 5 were randomly assigned to one of two conditions:

• Independent use: work on an AI literacy platform on their own during dedicated time, either in class or after school.
• Human-supported use: work on the same platform in a small group with a human tutor whose job was not to teach reading, but to encourage participation, troubleshoot technology, set group norms and build accountability. A typical session had a short check-in, about 15 minutes on the platform and a few minutes of reflection. In District B, the tutors were middle schoolers with a free period.

Both groups were scheduled for two 30-minute sessions a week. The platform was not named in the paper or in The 74’s coverage, which described it only as “a well-known AI reading tutor.”

The numbers: how far use fell short

Measure	District A	District B
Independent group: share who ever logged on	just over 60%	53%
Average weekly use, all independent students	just over 2 minutes	just over 5 minutes
Average weekly use, students who did log on	13.2 minutes	25.8 minutes
Weeks of actual use, on average	4–5 weeks within a 14–31 week window	4–5 weeks
Effect of human support on weekly use	about +1 minute	about +4.4 minutes
Effect of human support on stories completed	+71%	+80%
Effect on end-of-year reading achievement	not significant	not significant

Two things jump out. First, even the students who used the tool did so for a few weeks and then stopped. Second, the human helper changed how much work got done during the minutes kids were on the platform far more than it changed how many minutes they spent there.

In an August op-ed in The 74, Robinson and co-author Meghan McCormick of the Overdeck Family Foundation, a funder of tutoring research, noted that use “remained far below the 30-minute-per-week dosage recommended to reap academic benefits. At these levels, students likely did not receive enough of the intervention for gains in reading achievement to emerge.”

That last point matters for interpretation. The study does not show the AI reading tutor is ineffective. It shows that, in real schools, kids didn’t get enough of it to find out.

Why access doesn’t become use

Robinson was careful to say this was “not necessarily the students not engaging,” since the platform “was likely one of many tools available to teachers.” Several mechanisms are probably at work, and each has a fix.

Self-directed tools demand skills young kids are still building. An AI tutor that waits for a child to log on, pick a story and keep going is relying on executive function: planning, starting a task and sticking with it when it gets hard. Those skills develop through childhood. A 7-year-old is not a small, less-motivated adult. Robinson put it simply: “it’s so easy to look away or check out of something when it gets hard when you’re dealing with a screen.”

Asking for help is itself a skill. Khan Academy’s chief learning officer, Kristen DiCerbo, told Chalkbeat in April 2026 that AI can only respond to what students ask, and “students aren’t great at asking questions well.” Sal Khan described his own AI tutor as “a non-event” for many students: “Some will; most won’t.” We look at that admission in detail in our article on what Sal Khan said about Khanmigo.

Competition for time. In a school day, the AI tutor competes with every other activity a teacher could choose. Without someone making it part of the routine, it’s the easiest thing to skip.

The equity trap. Among students who used the platform on their own, those who logged on tended to be higher-achieving and less likely to receive special education services. Robinson’s conclusion: “it should give us pause about treating AI tutoring as an equity solution.” Self-serve tools tend to serve the kids who already serve themselves.

What makes kids actually engage: the human factor

The one intervention that clearly worked was a person. Not a reading teacher, just someone who showed up, checked in, fixed the login problem and noticed who wasn’t working. That person raised stories completed by 71–80%.

This fits decades of tutoring research. A 2020 NBER meta-analysis by Andre Nickow, Philip Oreopoulos and Vincent Quan of preK–12 tutoring experiments found “consistent and substantial positive impacts on learning outcomes,” with a pooled effect of 0.37 standard deviations. Effects were strongest for teacher and paraprofessional tutors, in earlier grades, and during the school day. Human tutoring works partly because of the relationship and accountability, the very things a screen can’t supply.

The research also points to a promising flip: use AI to help human tutors rather than replace them. Stanford SCALE’s 2026 review of the evidence on AI in K–12, by Lily Fesler, Susanna Loeb and colleagues, found that “AI tools that provide regular, automated feedback and diagnostics to human tutors can improve instructional quality and student outcomes,” especially for less experienced tutors.

That same review is a useful reality check on the whole field. Of more than 800 papers on AI in K–12 education, only 20 produced strong causal evidence, and the authors “did not identify any high-quality causal studies in K-12 settings in the U.S. for students.” Most AI tutor marketing is running ahead of the science.

What this means for parents

Ask how the school will make sure kids use it

If your child’s school adopts an AI tutor, the useful question isn’t “is it a good tool?” It’s “what’s the plan to make sure kids actually use it, and how will you check?” Usage data exists on almost every platform. Ask whether teachers look at it.

At home, be the human in the loop

The Stanford finding translates directly. A child left alone with a learning app will often drift. A parent who sits nearby for the first few minutes, asks “what story are you on?”, and talks about it afterward is doing what the study’s tutors did. You don’t need to know how to teach reading. Research on parent involvement in tutoring points the same way.

Schedule it like practice, and keep it short

Two short, predictable sessions at the same time each week beat “use it whenever.” Put it on the calendar. Five minutes of focused use is better than a 30-minute slot that turns into two.

Watch who the tool is serving

If your child is a confident reader who loves the app, great. If your child struggles with reading, don’t assume an AI tutor fills the gap. The struggling readers were the least likely to use it on their own. They are also the kids for whom a human tutor or reading specialist has the best evidence. Our comparison of AI tutors and human tutors lays out the trade-offs.

What not to do: count logins as learning

A child who “has access” to an AI tutor has nothing yet. A child who logs on but taps through stories has very little. The question is whether they can read something new better than last month. Check that directly: have them read aloud to you from a book they haven’t seen.

What to Watch For Over the Next 3 Months

• Week 4: Ask your child to show you the AI tutor and what they did on it this week. If they can’t, it isn’t being used. Ask the teacher what usage looks like.
• Month 2 red flags: Your child calls the tool “boring” or “annoying,” avoids it, or rushes through it to finish. A struggling reader who is being steered to the app instead of to a person.
• Month 3 self-check: Has your child’s reading improved on a book they haven’t seen before? Does the school have usage data and look at it? Is there a human checking in, at school or at home?

Frequently Asked Questions

Do AI reading tutors work?

We don’t know yet. In the Stanford trials, students used the AI tutor far too little to test its effect: averages of 2 to 5 minutes a week against a planned 60. Reading scores didn’t improve, but researchers say kids likely didn’t get enough dosage. Stanford’s 2026 review found no high-quality causal studies of AI tools for US K–12 students.

Why didn’t kids use the AI tutor?

Researchers don’t have a single answer. Likely factors include young children’s still-developing ability to start and stick with a task alone, difficulty asking good questions, competition from other classroom activities, and the tool not being built into teachers’ routines. Students who struggled most were least likely to use it on their own.

Did adding a human help?

Partly. A human helper who encouraged students and fixed tech problems raised the number of stories completed by 71–80%, but weekly minutes rose only slightly and reading achievement still didn’t improve significantly. Human support made the time more productive; it didn’t make kids use the tool enough.

Is this study peer-reviewed?

Not yet. It is a working paper from the Stanford SCALE Initiative, posted through the National Student Support Accelerator and the Annenberg Institute’s working paper series. Working papers are common in education research, but findings can change after review. The basic usage numbers are direct measurements, which makes them less likely to change.

Should I sign my child up for an AI reading app at home?

If you do, plan to be involved. The study’s lesson is that a person who checks in and builds routine makes a large difference to engagement. For a child who struggles with reading, a human tutor or reading specialist has much stronger evidence than any AI tool today. Talk to your child’s teacher first.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Robinson, C., et al. (2026). “Access Is Not Enough: Human Support Improves Engagement with AI Tutoring.” Working paper, Stanford SCALE Initiative / National Student Support Accelerator. https://nssa.stanford.edu/studies/access-not-enough-human-support-improves-engagement-ai-tutoring
2. The 74. (2026, June 17). “Study: Giving Kids Access to AI Tutors Doesn’t Mean They’ll Use Them.” The 74. https://www.the74million.org/article/study-giving-kids-access-to-ai-tutors-doesnt-mean-theyll-use-them/
3. McCormick, M., & Robinson, C. (2026, August 26). “Even With Human Help, Kids Need Motivation to Use AI Tutors. The Question Is What.” The 74. https://www.the74million.org/article/even-with-human-help-kids-need-motivation-to-use-ai-tutors-the-question-is-what/
4. Fesler, L., Martinez Claeys, J. P., Agnew, C., & Loeb, S. (2026). “The Evidence Base on AI in K-12: A 2026 Review.” Stanford SCALE Initiative. https://scale.stanford.edu/sites/default/files/The%20Evidence%20Base%20on%20AI%20in%20K-12%20Report.pdf
5. Nickow, A., Oreopoulos, P., & Quan, V. (2020). “The Impressive Effects of Tutoring on PreK-12 Learning: A Systematic Review and Meta-Analysis of the Experimental Evidence.” NBER Working Paper 27476. https://www.nber.org/papers/w27476
6. Barnum, M. (2026, April 9). “Why Sal Khan’s AI revolution hasn’t happened yet, according to Sal Khan.” Chalkbeat. https://www.chalkbeat.org/2026/04/09/sal-khan-reflects-on-ai-in-schools-and-khanmigo/