The 74
America's Education News Source
Copyright 2026 The 74 Media, Inc
Sign up for The 74 newsletter.
Get stories like this delivered straight to your inbox. Sign up for The 74 Newsletter
Ed tech companies routinely pitch AI tutoring platforms as a way to deliver personalized instruction at a scale that no human teacher can match. But when researchers from Stanford University looked at how much students actually used one major AI platform, something startling happened: Students didn’t use it that much at all.
In the study, published Wednesday, two unnamed school districts carved out dedicated time for hundreds of elementary school students to work with a well-known AI reading tutor, either during class time or after school. Researchers followed about 350 students across two randomized controlled trials. All of the students were expected to log on for at least two 30-minute sessions a week.
They found that of the students assigned to work independently with the AI, just over 60% in the first district and 53% in the second ever logged on to the platform — at all.
Among all students, average weekly usage came to just over two minutes in District A and just over five minutes in District B.
Those who did log on averaged 13.2 minutes a week in District A and 25.8 minutes in District B, using the tutoring for just four to five weeks on average in an “intervention window” that ran from 14 to 31 weeks.
For Carly Robinson, the paper’s lead author and research director for the Stanford SCALE Initiative, the gap between access and use isn’t a shock. “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.”
That’s going to take “intentional design” that appeals to both students and their teachers, who must choose whether to offer access.
“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,” Robinson said in an interview.
But she was careful to note that the study didn’t draw conclusions about AI’s effectiveness, or the degree to which students were interested or uninterested in the bot, saying many factors could be at play. “This is not necessarily the students not engaging,” she said. In the two districts, the AI platform “was likely one of many tools available to teachers.”
For the study, researchers randomly assigned a group of students to work on the platform alongside a few classmates and a human tutor whose job was to support their engagement and motivation and to troubleshoot any problems students might encounter. In District B, the tutors were actually middle-school students who “had a free intervention block in their school day.” A typical session included a short check-in, 15 minutes on the platform and a few minutes of reflection.
Pairing students with a tutor worked, Robinson said — to a point. Usage increased by roughly one minute a week in District A and 4.4 minutes in District B. The number of stories students completed each week jumped 71% in District A and 80% in District B.
What the human pairing didn’t do was move the needle on reading scores: Neither district saw a statistically significant improvement in end-of-year reading achievement. But Robinson said the study wasn’t primarily focused on that. Rather it was looking at the overall impact of adding a human into the equation, someone who provides “accountability, motivation and relationship building.”
Wednesday’s findings mirror recent ones from Khan Academy founder Sal Khan, who in April said that the rollout of his Khanmigo AI chatbot in 2023 was “a non-event” for many students. “They just didn’t use it much.”
Khan said AI tutoring doesn’t necessarily make students motivated to learn, or to fill in gaps in their knowledge needed to ask questions.
The new data also raise an uncomfortable question for educators: 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. So the students who stood to benefit most from extra reading practice were among the least likely to get it.
Robinson said she sees that as a red flag for anyone considering AI tutoring as a quick fix for underserved students: “I think it should give us pause about treating AI tutoring as an equity solution.”
Alex Sarlin, founder of the EdTech Insiders newsletter and a veteran industry watcher, said the new study “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, leading to lower usage among the neediest students, and a faulty assumption that students will jump into new tools without structured guidance.”
The researchers’ approach showcases a promising direction, he said, “as it is increasingly clear that providing access to tooling is not nearly enough to drive usage, let alone outcomes.”
Amanda Bickerstaff, co-founder and CEO of AI for Education, which provides AI literacy training to teachers, said results like these aren’t all that surprising, given what we know about these tools.
All GenAI chatbots, she said, can make mistakes, lack important context about students and how they learn best, and can provide biased outputs. Her group has recommended keeping these tools out of the hands of students through second grade, “and only with significant human oversight and AI literacy training” for students in grades three through five.
“At this stage, there has been little evidence that GenAI chatbot tutors meaningfully impact learning outcomes for students,” she said, “or that they are developmentally appropriate for students in elementary schools.”
Robinson, the study’s lead author, said she sees the usage findings as part of a larger pattern playing out as schools adopt AI tools more broadly. Schools, she said, should consider offering students “different iterations of these things based on what they actually need — and that’s probably a more likely pathway to scale than just saying, ‘Let’s give everyone an AI tutor.’ ”
Historically, personalized instruction has depended almost entirely on human teachers, with the teacher-student relationship central to the experience. But advances in technology — most recently in AI — have changed this dynamic, Robinson and her colleagues write. Now, personalized instruction exists on what they term “a spectrum of relational intensity,” from a consistent one-on-one human tutor to a computer platform that students navigate alone.
AI tutors may approximate human interactions, Robinson said, but students may still benefit from the care and companionship that humans provide. Logging on and sticking with something that might prove to be difficult, she said, is easier with a human in the mix. “There is just this component of accountability that a human can provide, where it’s so easy to look away or check out of something when it gets hard when you’re dealing with a screen.”
Did you use this article in your work?
We’d love to hear how The 74’s reporting is helping educators, researchers, and policymakers. Tell us how
Greg Toppo is a senior writer at The 74.
We want our stories to be shared as widely as possible — for free.
Please view The 74’s republishing terms.
By Greg Toppo
This story first appeared at The 74, a nonprofit news site covering education. Sign up for free newsletters from The 74 to get more like this in your inbox.
Ed tech companies routinely pitch AI tutoring platforms as a way to deliver personalized instruction at a scale that no human teacher can match. But when researchers from Stanford University looked at how much students actually used one major AI platform, something startling happened: Students didn’t use it that much at all.
In the study, published Wednesday, two unnamed school districts carved out dedicated time for hundreds of elementary school students to work with a well-known AI reading tutor, either during class time or after school. Researchers followed about 350 students across two randomized controlled trials. All of the students were expected to log on for at least two 30-minute sessions a week.
They found that of the students assigned to work independently with the AI, just over 60% in the first district and 53% in the second ever logged on to the platform — at all.
Among all students, average weekly usage came to just over two minutes in District A and just over five minutes in District B.
Those who did log on averaged 13.2 minutes a week in District A and 25.8 minutes in District B, using the tutoring for just four to five weeks on average in an “intervention window” that ran from 14 to 31 weeks.
For Carly Robinson, the paper’s lead author and research director for the Stanford SCALE Initiative, the gap between access and use isn’t a shock. “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.”
That’s going to take “intentional design” that appeals to both students and their teachers, who must choose whether to offer access.
“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,” Robinson said in an interview.
But she was careful to note that the study didn’t draw conclusions about AI’s effectiveness, or the degree to which students were interested or uninterested in the bot, saying many factors could be at play. “This is not necessarily the students not engaging,” she said. In the two districts, the AI platform “was likely one of many tools available to teachers.”
For the study, researchers randomly assigned a group of students to work on the platform alongside a few classmates and a human tutor whose job was to support their engagement and motivation and to troubleshoot any problems students might encounter. In District B, the tutors were actually middle-school students who “had a free intervention block in their school day.” A typical session included a short check-in, 15 minutes on the platform and a few minutes of reflection.
Pairing students with a tutor worked, Robinson said — to a point. Usage increased by roughly one minute a week in District A and 4.4 minutes in District B. The number of stories students completed each week jumped 71% in District A and 80% in District B.
What the human pairing didn’t do was move the needle on reading scores: Neither district saw a statistically significant improvement in end-of-year reading achievement. But Robinson said the study wasn’t primarily focused on that. Rather it was looking at the overall impact of adding a human into the equation, someone who provides “accountability, motivation and relationship building.”
Wednesday’s findings mirror recent ones from Khan Academy founder Sal Khan, who in April said that the rollout of his Khanmigo AI chatbot in 2023 was “a non-event” for many students. “They just didn’t use it much.”
Khan said AI tutoring doesn’t necessarily make students motivated to learn, or to fill in gaps in their knowledge needed to ask questions.
The new data also raise an uncomfortable question for educators: 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. So the students who stood to benefit most from extra reading practice were among the least likely to get it.
Robinson said she sees that as a red flag for anyone considering AI tutoring as a quick fix for underserved students: “I think it should give us pause about treating AI tutoring as an equity solution.”
Alex Sarlin, founder of the EdTech Insiders newsletter and a veteran industry watcher, said the new study “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, leading to lower usage among the neediest students, and a faulty assumption that students will jump into new tools without structured guidance.”
The researchers’ approach showcases a promising direction, he said, “as it is increasingly clear that providing access to tooling is not nearly enough to drive usage, let alone outcomes.”
Amanda Bickerstaff, co-founder and CEO of AI for Education, which provides AI literacy training to teachers, said results like these aren’t all that surprising, given what we know about these tools.
All GenAI chatbots, she said, can make mistakes, lack important context about students and how they learn best, and can provide biased outputs. Her group has recommended keeping these tools out of the hands of students through second grade, “and only with significant human oversight and AI literacy training” for students in grades three through five.
“At this stage, there has been little evidence that GenAI chatbot tutors meaningfully impact learning outcomes for students,” she said, “or that they are developmentally appropriate for students in elementary schools.”
Robinson, the study’s lead author, said she sees the usage findings as part of a larger pattern playing out as schools adopt AI tools more broadly. Schools, she said, should consider offering students “different iterations of these things based on what they actually need — and that’s probably a more likely pathway to scale than just saying, ‘Let’s give everyone an AI tutor.’ ”
Historically, personalized instruction has depended almost entirely on human teachers, with the teacher-student relationship central to the experience. But advances in technology — most recently in AI — have changed this dynamic, Robinson and her colleagues write. Now, personalized instruction exists on what they term “a spectrum of relational intensity,” from a consistent one-on-one human tutor to a computer platform that students navigate alone.
AI tutors may approximate human interactions, Robinson said, but students may still benefit from the care and companionship that humans provide. Logging on and sticking with something that might prove to be difficult, she said, is easier with a human in the mix. “There is just this component of accountability that a human can provide, where it’s so easy to look away or check out of something when it gets hard when you’re dealing with a screen.”
Copyright 2026 The 74 Media, Inc
Study: Giving Kids Access to AI Tutors Doesn’t Mean They’ll Use Them – The 74
Leave a Comment
