Student-AI exchanges in an undergraduate research methods course are empirically classifiable into three interaction paths, with Passive Review most frequent (46%) and Strategic Dialogue 36%¶

Claim · All claims
Evidence · 1 study (3 entries) · qualitative r2 · q2
Currency · newest study 2025 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Qualitative content analysis of 50 double-coded messages classified exchanges into the three typological paths, with R1 final classifications of 46% Path 1, 18% Path 2, and 36% Path 3. → Zahra 2025 q2 i? Path-level inter-rater agreement between two human coders was 68% (κ = .48), rated moderate, with 80% agreement on Strategic Dialogue classification. → Zahra 2025 (2) q2 i? A three-coder majority-vote distribution (P1 = 48%, P2 = 16%, P3 = 36%) closely mirrored the primary researcher's classifications, corroborating the proportions. → Zahra 2025 (3)

Evidence¶

Zahra 2025¶

Zahra, F. T. (2025). Three Pathways of Student-AI Interaction: Constraint-First Design for Higher-Order Thinking. Manuscript under blind review; framework first presented at EAAI-25, Philadelphia, PA. https://arxiv.org/abs/2610.00338 doi:10.48550/arxiv.2610.00338

q2 · i? · qualitative · r2

Qualitative content analysis of 50 randomly sampled student-AI messages from a corpus of 580 produced by 14 undergraduates across seven assignments. The reported distribution is "46% of coded exchanges" for Path 1, 18% Path 2, 36% Path 3; no effect size is printed.

"Using R1’s final classifications, Path 1 (Passive Review) accounted for 46% of coded exchanges, Path 2 (Direct Question) for 18%, and Path 3 (Strategic Dialogue) for 36%."

Zahra 2025 (2)¶

Zahra, F. T. (2025). Three Pathways of Student-AI Interaction: Constraint-First Design for Higher-Order Thinking. Manuscript under blind review; framework first presented at EAAI-25, Philadelphia, PA. https://arxiv.org/abs/2610.00338. doi:10.48550/arxiv.2610.00338

q2 · i? · qualitative · r2

Two human coders independently coded all 50 user messages against a ten-label codebook; the article reports "68% (κ = .48)" path-level agreement, moderate by Landis and Koch. Coders agreed on Path 3 in 80% of cases (40/50). No effect size is printed.

"Path-level inter-rater agreement between R1 and R2 was 68% (κ = .48), which Landis and Koch (1977) classify as moderate."

Zahra 2025 (3)¶

Zahra, F. T. (2025). Three Pathways of Student-AI Interaction: Constraint-First Design for Higher-Order Thinking. Manuscript under blind review; framework first presented at EAAI-25, Philadelphia, PA. https://arxiv.org/abs/2610.00338. doi:10.48550/arxiv.2610.00338

q2 · i? · qualitative · r2

Supplementary triangulation computed a majority-vote distribution across R1, R2, and GPT-5, which the article says "closely mirrors R1's classifications" (46/18/36), offered as corroboration that proportions are not an artifact of a single coder.

"the majority-vote distribution across all three coders (R1, R2, and GPT-5) was computed separately: P1 = 48%, P2 = 16%, P3 = 36%."

Discussion¶

Related Claims¶

• A reflection-structured assignment produced predominantly Strategic Dialogue content, offering preliminary, non-causal evidence that instructional framing may shift path distribution — related
• Without explicit scaffolding, the two lower-agency paths (Passive Review and Direct Question) accounted for 64% of student-AI interactions in the primary sample — related
• Undergraduate LLM reliance classifies into four types — Strategic (34.3%), Instrumental (30.9%), Dialogic (30.4%), and Dependent (4.5%) — using a threshold-based classification rule — related
• Students engaged deeply with LLM feedback, with coded evidence of reflection, rebuttal, and acceptance in the large majority of responses — related
• Students most often report hybrid interaction modes with generative AI, with passive task delegation and verification-only use as smaller patterns — related