Access is Not Enough: Human Support Improves Engagement with AI Tutoring

Synthesis: Robinson, Gormley, Ribeiro & Loeb (2026) ran two RCTs showing that AI tutoring's binding constraint is take-up, not capability: despite dedicated session time, nearly half of students never used the platform and users averaged only 2–5 minutes per week. An in-person engagement tutor (not direct instruction) raised usage by 1–4 minutes/week and engagement by 71–80% — but dosage stayed far below the level needed for reading gains, and achievement did not improve.

The two RCTs

• Districts A and B, elementary students, AI literacy (reading) platform; treatment = in-person tutor focused on engagement (10-min check-in + 15 min platform + 2–5 min reflection), control = platform alone with the full ~30 min
• Pre-registered outcomes: minutes/week (usage) and stories read/week (engagement); spring ELA as achievement
• Interventions lasted 14–31 weeks

Take-up is the story

• Only 60.7% / 53.3% of control students ever used the platform
• Average weekly usage: 2.18 / 5.23 minutes; even users-only averages were 13.2 / 25.8 minutes
• Students used the platform in only 4–5 of 14–31 weeks
• Platform users skewed higher-achieving and less likely to receive Special Education services — an equity flag: the students who need adaptive support most are least likely to engage

What human support did

• Usage: +1 min/week (A), +4.4 min/week (B) — significant, but a drop against the ~30 min/week the provider recommends for measurable reading gains
• Engagement: +0.20 stories/week (A, +71%) and +0.92 stories/week (B, +80%)
• Total added dosage: ~22 minutes (A) and ~98 minutes (B) across the entire intervention
• No reading achievement gains in either district (negative, non-significant) — consistent with the achieved dosage
• Strong site-level variation: some sites saw little effect, others meaningful gains — local implementation conditions matter

What this means for practice

• Learners. Show up for every scheduled session rather than treating access as use: control students used the platform in only 4–5 of the 14–31 weeks and averaged 2.18 minutes per week in District A and 5.23 in District B.
• Instructors. Give the AI time a person whose job is engagement, not instruction: a 10-minute check-in, 15 minutes on the platform, and a 2–5 minute reflection raised usage by 1 minute per week in District A and 4.4 in District B, and engagement by 71% and 80%.
• Instructors. Do not expect engagement support to buy achievement: neither district showed significant reading gains, and the estimated spring ELA effects were negative and non-significant, consistent with the dosage actually achieved.
• Administrators. Check take-up before renewing a platform, especially for the students who need it most: users skewed higher-achieving and less likely to receive special education services, so access alone may widen participation gaps.
• Administrators. Fund and staff the human layer: the tutors here were after-school program staff in District A and middle school students in District B, and the authors note that human-intensive models require additional personnel, training, and coordination.

Limitations

• Take-up limits what the trials can test: only 60.7% (District A) and 53.3% (District B) of control students ever used the platform, and the engagement gains are measured against a baseline that was already near zero.
• Group sizes were small and unequal across two sites: 84 treatment against 90 control students at five after-school sites in District A, and 36 treatment against 145 control students at two K–8 schools in District B, with the districts differing in grades, setting, and tutor type.
• No background information was collected on the human tutors, and District B's tutors were middle school students, so the intervention's active ingredient is not characterized.
• The achievement estimates are null and negative in sign in both districts, and the authors caution that even the highest-usage subgroup figure (18.3 minutes per week) should be read cautiously given the small sample size.

Citation

Robinson, C. D., Gormley, D., Trindade Ribeiro, A., & Loeb, S. (2026). Access is Not Enough: Human Support Improves Engagement with AI Tutoring. EdWorkingPaper No. 26-1451, Annenberg Institute at Brown University. DOI