LLM scaffolding level is associated with the student's next dialogue act, and the association persists within broad request categories¶

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

Subclaims¶

q2 i? The distribution of students' next dialogue acts differed across scaffolding levels (χ2 = 1,092.6, df = 28, p < .001, V = .137): after Solving responses students most often made another writing request (29.7%), after Explaining responses conceptual questions were most common (40.5%). → Suhyeon Lee 2026 q2 i? The association remained statistically significant within all four broad current student dialogue-act categories examined (all p < .01), and persisted after collapsing Explaining and Solving, though the effect size decreased from V = .137 to V = .116. → Suhyeon Lee 2026 (2)

Evidence¶

Suhyeon Lee 2026¶

Suhyeon Lee, Juneha Baek, Jaehyeong Park, Donghyuk Shin. (2026). LLM Pedagogical Behavior in AI Tutoring Interactions. https://arxiv.org/abs/2608.22993 doi:10.48550/arxiv.2608.22993

q2 · i? · associational · r2

Turn-level chi-squared analysis of 14,637 paired turns linking each response's scaffolding level to the student's next dialogue act; the article notes the turn-level analyses do not model within-student dependence.

"The distribution of students’ next dialogue acts differs across scaffolding levels ( χ2 = 1,092.6 ,df= 28,p<.001,V=.137; Table 3)."

Suhyeon Lee 2026 (2)¶

Suhyeon Lee, Juneha Baek, Jaehyeong Park, Donghyuk Shin. (2026). LLM Pedagogical Behavior in AI Tutoring Interactions. https://arxiv.org/abs/2608.22993

q2 · i? · associational · r2

Within-dialogue-act stratified analyses (Table 10) repeating the association test inside each request category; the largest contrast was provide-context turns after 16.4% of L4 versus 7.3% of L3 responses among writing requests.

"The association remains statistically significant within all four broad current student dialogue-act categories examined (all p < .01). The pattern therefore persists beyond broad differences in the type of request preceding the LLM response."

Discussion¶

Related Claims¶

• A general-purpose LLM used as a tutor without pedagogical prompting overwhelmingly provides high assistance, with more than 95% of responses classified as Explaining or Solving — related
• The relationship between scaffolding level and students' subsequent provide-context turns is not monotonic — related
• LLM scaffolding level varies systematically across student request types and across assignments, but high-assistance responses remain dominant — related
• Contingent tutoring held delegation at the level of an answer-on-request assistant while delivering more artifacts, one decision at a time — related
• Within a single LLM session, students' help-seeking shifts toward answer- and verification-seeking while explicit self-regulation modestly increases — related
• SPED-adaptive prompting shifts tutor behavior toward We-Do scaffolding, naturalistic language support, and participation-affirming feedback across three codebooks — related
• Direct student requests for answers significantly increased the AI's provision of direct answers (26.1% vs. 12.5% baseline; r = 0.14), though the AI maintained scaffolding in most cases — related
• Educators redefined AI's educational role from answer provider to learning partner and tutor supporting conceptual understanding — related