Students barely used AI tutors when left alone, Stanford study finds
A Stanford study found students left alone with an AI tutor used it 2–5 minutes a week; human support raised use but not end-of-year reading scores.
Photo by Katerina Holmes on Pexels
Giving elementary students access to an AI tutor does not mean they will use it. That is the finding of a Stanford University study of AI tutoring published on 17 June 2026, which tracked about 350 children across two randomized controlled trials. When students were left to work on a widely used AI reading tutor on their own, average weekly use came to just over two minutes in one district and just over five in another — far below the two 30-minute sessions a week the schools had set aside.
The study, “Access Is Not Enough: Human Support Improves Engagement in AI Tutoring,” was published through the Stanford National Student Support Accelerator and led by Carly Robinson, research director for the Stanford SCALE Initiative. As reported by The 74’s coverage of the Stanford AI tutoring study, the work set out to test not whether the tool works, but whether students use it at all — and what changes when a person is added to the loop.
How much students used the AI tutor
Across both districts, students assigned to work alone with the platform logged on rarely. Just over 60% of those students in the first district (District A) ever logged on; in the second (District B), the figure was 53%. Among all students assigned to independent use, average weekly use was just over two minutes in District A and just over five minutes in District B. Even counting only the students who logged on at least once, weekly use averaged 13.2 minutes in District A and 25.8 minutes in District B. On average, students used the tutor for only four to five weeks during an intervention window that ran from 14 to 31 weeks.
The schools had set aside time for at least two 30-minute sessions per week — about 60 minutes — so actual use fell well short of the design.
Why AI tutoring access did not equal use
For Robinson, the gap between access and use was not a surprise. “As we’re talking about bringing AI tools into the classroom, the challenge isn’t just building good AI tools,” she said. “It’s getting students to use them and engage with them effectively.” Reaching students, she said, takes “intentional design” that appeals to both learners and the teachers who decide whether to offer the tool at all. Robinson cautioned that the study did not measure how effective the tutor is, or whether students were interested in it, noting the platform “was likely one of many tools available to teachers.”
| Measure | District A | District B |
|---|---|---|
| Share who ever logged on (worked independently) | just over 60% | 53% |
| Average weekly use — all assigned students | ~2 min | ~5 min |
| Average weekly use — students who logged on | 13.2 min | 25.8 min |
| Added weekly use when paired with a human tutor | +~1 min | +4.4 min |
| Increase in weekly stories completed with a human tutor | +71% | +80% |
Source: Stanford NSSA study “Access Is Not Enough” (2026), as reported by The 74.
What a human in the loop changed
The researchers also tested a second condition: students worked on the platform in small groups alongside a human tutor whose job was to support motivation and engagement and to troubleshoot problems. In District B, those tutors were middle-school students who had a free intervention block in their school day. A typical session paired a short check-in, about 15 minutes on the platform, and a few minutes of reflection.
Adding a person helped — to a point. Weekly use rose by roughly one minute in District A and 4.4 minutes in District B, and the number of stories students completed each week climbed 71% in District A and 80% in District B. What the human support did not do was lift reading results: neither district recorded a statistically reliable improvement in end-of-year reading achievement. Robinson noted that measuring reading gains was not the study’s main aim; the focus was the effect of adding “accountability, motivation and relationship building.”
The equity question
The data raised a further concern. 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 — meaning the children who stood to gain the most reading practice were among the least likely to get it. “I think it should give us pause about treating AI tutoring as an equity solution,” Robinson said. The pattern echoes earlier evidence that AI in classrooms can widen rather than close gaps; the OECD has reported that AI can raise grades but harm learning when tools substitute for the effort of learning.
Outside researchers said the results fit a wider pattern. Alex Sarlin of the EdTech Insiders newsletter said the study points to “low usage rates that don’t meet dosage recommendations” and “a faulty assumption that students will jump into new tools without structured guidance.” Amanda Bickerstaff, chief executive of AI for Education, said her group recommends keeping generative-AI chatbots out of the hands of students through second grade, and using them in grades three to five only with close human oversight and AI-literacy training.
Background
AI tutoring tools are routinely promoted as a way to deliver one-to-one, personalized instruction at a scale no human teacher can match. Evidence on whether students actually use them has lagged behind the marketing. Khan Academy founder Sal Khan said in April 2026 that the 2023 rollout of his Khanmigo chatbot was “a non-event” for many students, who “just didn’t use it much.” Robinson and her colleagues frame personalized instruction as a “spectrum of relational intensity,” running from a consistent one-on-one human tutor to a platform a student navigates alone, and argue that the human relationship still carries weight a screen cannot.
That places the new findings within a longer debate over the role of people in AI-assisted classrooms, a theme Winss Solutions has tracked in coverage of how AI should support, not replace, teachers and of the OECD’s work on the digital transformation of school education and the divide in access it can leave behind. The Stanford study adds a measured data point: even a capable AI tutor does little if students never open it, and adding a person raises engagement without, on its own, moving test scores.
Sources: Stanford National Student Support Accelerator; The 74
Featured image: photo by Katerina Holmes on Pexels (free Pexels license).
Become a Sponsor
Our website is the heart of the mission of WINSS – it’s where we share updates, publish research, highlight community impact, and connect with supporters around the world. To keep this essential platform running, updated, and accessible, we rely on the generosity of you, who believe in our work.
We offer the option to sponsor monthly, or just once choosing the amount of your choice. If you run a company, please contact us via info@winssolutions.org.
I specialize in sustainability education, curriculum co-creation, and early-stage project strategy. At WINSS, I craft articles on sustainability, transformative AI, and related topics. When I’m not writing, you’ll find me chasing the perfect sushi roll, exploring cities around the globe, or unwinding with my dog Puffy — the world’s most loyal sidekick.
