Build AI With Us

Giving Your Kid an AI Tutor Isn't Enough — Here's What a New Stanford Study Actually Found

Written by
Build AI With Us
Reviewed by
Muddsar Iqbal
Published
Last updated
7 min read

A new Stanford study tracked 355 students using an AI reading tutor and found something parents assume is wrong: just giving a kid access to an AI tool barely changes how much they use it. Here's what actually moved the needle, and what the study honestly didn't find.

Here's an assumption a lot of parents make without ever testing it: if an AI tutoring app is genuinely good, a kid will just use it. Buy the subscription, hand over the login, and the learning follows. A new Stanford study, published in June 2026, put that assumption to a real test with real students — and found it mostly doesn't hold.

What the study actually did

Researchers from Stanford's SCALE Initiative ran two randomized controlled trials with 355 elementary students (grades 1-5, most from economically disadvantaged backgrounds) using an AI reading tutoring platform. Every student in the study had the same access to the same AI tool. The only difference between groups: some students also got light human support — not instruction, just someone helping them actually use the platform and stay engaged with it.

The question wasn't "does AI tutoring work." It was narrower and more useful: does simply having access to a good AI tool mean a kid will actually use it?

The answer: not really, on its own

In the group with AI access but no added human support, usage was low — students logged an average of about 2 to 5 minutes per week on the platform. Not per session. Per week, total. Nearly half of the students in the comparison group never logged in at all.

When human support was added — someone checking in, helping students get started, keeping the habit alive — weekly usage rose in both districts: by about 46% in one (from roughly 2 to 3 minutes a week) and by about 85% in the other (from roughly 5 to 10 minutes a week), and the number of stories read per week rose 71-80%. That's a real, measurable effect, and it's the headline finding: access alone doesn't produce engagement. A person alongside the tool does — even if, as the numbers above show, the totals involved are still small in absolute terms.

The part that keeps this honest

Here's where it would be easy to oversell this study, so it's worth being direct about what it didn't find: even with the human-support boost, this particular trial found no measurable gain in reading test scores in either district. The engagement went up. The achievement scores, in this study, did not move in a statistically meaningful way — and the direction of the (non-significant) effect actually trended slightly negative, not positive. That's not evidence the approach backfired; with effects this small and samples this size, a non-significant negative trend and a non-significant positive trend are both consistent with "no real difference detected" — but it's one more reason not to oversell this as proof that human support drives better test scores.

A few reasons that matters, rather than undermines the finding:

Why this matters if you're considering an AI tool for your kid

The practical lesson isn't "AI tutoring doesn't work." It's narrower and more useful: don't judge a tool by its features — judge it by whether your child will actually, consistently use it, and plan for the fact that most won't without someone involved.

A few ways this shows up in real decisions:

If you've bought an AI tutoring subscription your child barely touches, this study suggests that's not unusual, and it's not necessarily a sign the tool is bad. It's a sign the tool is missing the human layer that gets a habit started and keeps it going.

"Self-paced" and "unsupervised" are not the selling points they sound like. A tool marketed on independence — your child can use this anytime, no scheduling needed — is describing exactly the condition this study found produces near-zero usage. Flexibility without accountability tends to mean the app sits unopened.

The person matters more than the platform, at least at the start. This doesn't mean a parent needs to sit through every session. It means some consistent human touchpoint — checking in, asking what was learned, noticing when the habit lapses — does more for actual use than any feature the AI tool itself offers.

Where this fits with project-based, mentored learning

This is the exact gap 1:1 mentorship is built to close — not by replacing a good AI tool, but by being the consistent human presence this study found makes the difference between a tool that gets used and one that quietly goes untouched. A mentor doesn't just introduce a project; they show up for it, which is precisely the variable this study isolated as the one that actually moves engagement. If you want to see what that structure looks like in practice rather than relying on an app alone, our guide on why 1:1 mentorship works better than apps alone walks through the difference.

A simple starting checklist

If you want a clearer sense of what consistent, mentored AI learning actually looks like for your child, book a free, no-pressure assessment at book a Free Trial Class.

FAQs

Q: Does this study mean AI tutoring doesn't work? A: Not quite — it means AI tutoring access alone often goes unused, which is a different and earlier problem than whether the tutoring itself is effective. The study didn't find achievement gains either way, but usage in both groups was likely too low to expect them.

Q: Is this study strong enough to base a big decision on? A: Treat it as a useful, concrete signal rather than settled proof — it's an unpublished working paper with modest sample sizes, not a large, peer-reviewed, replicated study. The engagement finding (access alone isn't enough) is the most solid part; the achievement findings are inconclusive, which the researchers themselves are upfront about.

Q: My child has an AI tutoring app and barely uses it. What should I actually do? A: This study suggests the fix is less about finding a better app and more about adding a consistent human touchpoint — checking in regularly, sitting with them occasionally, or structuring a specific time to use it rather than leaving it fully self-directed.

Q: How much should my child actually be using an AI tutoring tool? A: There's no universal number, but this research area points to roughly 30 minutes a week as a commonly cited threshold vendors associate with measurable gains — useful as a rough benchmark, not a guarantee.

Q: Isn't "kids need supervision to use AI tools" just common sense? A: It might sound obvious in hindsight, but it directly contradicts how a lot of AI education tools are marketed — as something a child can use independently, anytime. This study is useful specifically because it tests that marketing claim and finds it doesn't hold up for most kids.

Q: Does this apply to older kids and teens too, or just the young children in this study? A: The study itself only covered grades 1-5, so it can't be extended directly to teens. But the broader pattern — that access without a human layer tends to produce low engagement — shows up consistently enough across AI-learning research that it's a reasonable caution to apply more broadly, even if this specific data is about younger kids.

researchai-tutoringparent-guide2026mentorship

Ready to see if 1:1 AI mentorship is right for your child?

Book a Free Trial Class. Meet a mentor and get a personalized path.

Book a Free Trial Class

Or compare our live AI classes for kids and 1:1 AI tutoring for kids.