Voice and text tutoring produced statistically equivalent weekly mastery within students, while voice transformed the interaction at a 2.8× cost premium¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r2 · q2–q3
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q3 i? Within-student week-level mastery was statistically equivalent between voice and text modalities, with equivalence bounds reported so the null reads as affirmative evidence of equivalence. → Shihao Yang 2026 q2 i? Voice's process transformation carried a 2.8× delivery-cost premium, so programs can buy engagement with voice or equivalent measured weekly mastery with text at a third of the cost. → Shihao Yang 2026 (2)

Evidence¶

Shihao Yang 2026¶

Shihao Yang, Marshall Van Alstyne, and Chrysanthos Dellarocas. (2026). When AI Tutors Speak: Evidence from a Randomized Field Experiment. https://arxiv.org/abs/2609.23958 doi:10.48550/arxiv.2609.23958

q3 · i? · causal · r2

Within-student weekly alternation analysis (student and week fixed effects, clustered standard errors) among AI-arm students; the design offsets noisy three-item subscores with 145 student-weeks and reports equivalence bounds alongside point estimates.

"Week-level mastery is statistically equivalent betweenmodalities,whiletheinteractionitselfchangessubstantially(roughlytwice the dialogue turns per minute and 2.4×the question-asking on equal engaged time)."

Shihao Yang 2026 (2)¶

Shihao Yang, Marshall Van Alstyne, and Chrysanthos Dellarocas. (2026). When AI Tutors Speak: Evidence from a Randomized Field Experiment. https://arxiv.org/abs/2609.23958

q2 · i? · causal · r2

Conversation-level cost accounting from provider billing over the study's 256 tutor conversations and 1,880 metered voice minutes, pricing the voice-versus-text configuration choice for program deployment.

"Voice's process transformation carries a 2.8×cost premium, so programs can buy engagement with voice, or equivalent measured weekly mastery with text at a third of the cost."

Discussion¶

Related Claims¶

• Voice modality roughly doubled dialogue turns and question-asking on equal engaged time within the same students — related
• Perceived task load and germane load did not differ between the contingent tutor and either baseline — related
• The talking head contributes little to AutoTutor's learning gains; the content of the conversation matters — related