Tutors with Tutor CoPilot access use more understanding-fostering strategies (e.g., prompting students to explain) and rely less on giving away answers than control tutors¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r3 · q3
Currency · newest study 2025 · checked against its sources 2026-10-20 (automated, 2 of 2 entries) · not reviewed by a person

Subclaims¶

q3 i? Treatment tutors used strategies fostering student skills and comprehension significantly more; prompting the student to explain and asking questions to guide thinking appeared roughly 2 standard deviations more frequently on a log-odds scale. → Rose E. Wang 2025 q3 i? Control tutors tended to rely on strategies that directly lead students to the solution or provide passive support, such as giving away the answer or providing the solution strategy. → Rose E. Wang 2025 (2)

Evidence¶

Rose E. Wang 2025¶

Rose E. Wang, Ana T. Ribeiro, Carly D. Robinson, Susanna Loeb, and Dora Demszky. (2025). Tutor CoPilot: A Human-AI Approach for Scaling Real-Time Expertise. arXiv:2410.03017v2. https://arxiv.org/abs/2410.03017 doi:10.48550/arxiv.2410.03017

q3 · i? · causal · r3

Classifier-based analysis of 550,000+ chat messages from the RCT sessions: fine-tuned RoBERTa classifiers (F1-scores 0.65–0.90) labeled strategies, and z-scored log-odds ratios compared usage across conditions (Figure 3). The article prints the ~2 standard deviation log-odds difference but no standardized effect size.

"For example, "prompting the student to explain" and "asking questions to guide thinking" were used approximately 2 standard deviations more frequently on a log-odds scale in treatment sessions compared to control sessions."

Rose E. Wang 2025 (2)¶

Rose E. Wang, Ana T. Ribeiro, Carly D. Robinson, Susanna Loeb, and Dora Demszky. (2025). Tutor CoPilot: A Human-AI Approach for Scaling Real-Time Expertise. arXiv:2410.03017v2. https://arxiv.org/abs/2410.03017

q3 · i? · causal · r3

Same message-classifier comparison across treatment and control sessions (Figure 3). The article characterizes control tutors' strategies as solution-focused and passive, which help students complete tasks but do less to develop deeper understanding. No effect size is printed for this contrast.

"In contrast, control tutors tended to rely on strategies that focus on directly leading students to the solution or providing passive support."

Discussion¶

Related Claims¶

• Tutor Copilot Raises Exit Ticket Mastery 4Pp
• Tutor CoPilot's benefit is largest for lower-rated and less-experienced tutors (9 p.p. and 7 p.p.), diminishing as tutor effectiveness increases — related
• Tutors perceive Tutor CoPilot as helpful for on-the-spot explanations but flag suggestions that are not grade-level appropriate and 'too smart' — related
• Human tutoring has a medium effect over no tutoring, and in one small study students learned as well when tutors only prompted them as when tutors also explained and gave feedback — a broader claim this one bears on
• Offloading work to AI costs learning: students given answers did well in practice but scored lower on an unassisted exam — a broader claim this one bears on
• Limiting verbose Math Agent guidance sharply reduced answer giveaway but decreased cognitive engagement — related
• Providing the tutor with student mastery and practice-history context improved engagement and next-item correctness — related
• More than 70% of teachers' help giving to students with ASD consisted of providing direct feedback, explaining, or instructing — related