Source: Robinson, C. D., Gormley, D., Ribeiro, A. T., & Loeb, S. (2026). Access is not enough: Human support improves engagement with AI tutoring (EdWorkingPaper No. 26-1451). Annenberg Institute at Brown University. https://doi.org/10.26300/pz7p-p388
LEARN Brief Credits: Dr. Jeannie Haubert
Overview
Two randomized controlled trials in elementary schools tested whether pairing students with an in-person human tutor — focused on engagement, not direct instruction — increased their use of an AI literacy platform, compared to students using the platform independently. Even with dedicated session time, nearly half of students in the independent-use group never logged on, and those who did averaged just 2–5 minutes per week. Adding a human tutor raised usage and engagement significantly, but overall use remained far below the threshold needed to move the needle on reading achievement.
Key Insights
- Access alone doesn’t drive use. Roughly half of students with scheduled platform time and open access never used the AI tutor at all; average weekly use was only 2–5 minutes — far below the ~30 minutes/week the platform provider links to reading gains.
- Human tutors moved the needle on engagement. Pairing students with a tutor focused on motivation, check-ins, and troubleshooting (not instruction) increased weekly platform usage by 1–4.4 minutes and boosted story completion by 71–80%, even though tutoring sessions left less time for platform use.
- Gains were real but small in absolute terms. Even with a tutor, total added usage amounted to less than two additional hours of platform time over the entire intervention (14–31 weeks) — not enough to produce measurable reading achievement gains in either district.
- Struggling students were least likely to engage. Among students working independently, those receiving special education services and lower-achieving students were the least likely to use the platform — raising equity concerns that AI tools may disproportionately benefit students already positioned to succeed.
- Consistency of support matters. Effects varied widely by site — some saw large usage gains from tutoring, others saw little to none — suggesting local implementation quality, not just the presence of a tutor, determines impact.
Action Steps
- Don’t treat AI tutoring as “set it and forget it.” Budget for human staffing (tutors, paraprofessionals, or trained aides) alongside any AI platform purchase — access without support is unlikely to generate use or outcomes.
- Monitor usage data, not just licenses purchased. Track weekly minutes and completion rates by student subgroup; treat platforms with <30 min/week average use as a red flag requiring intervention.
- Prioritize support for struggling and special education students. Since these students are least likely to self-engage, ensure human check-ins and small-group tutoring explicitly target them rather than defaulting to independent use.
- Pilot before scaling, and evaluate by site. Because effects varied substantially by implementation site, test any human-support model in a few schools first and compare usage/engagement data before a district-wide rollout.
- Set realistic expectations with staff and boards. Human support increases engagement, but this study found no achievement gains at the dosage levels achieved — frame AI tutoring as a multi-year engagement-building investment, not a quick fix.
Full Study
edworkingpapers.com/sites/default/files/ai26-1451.pdf
