Qualitative transcript analysis showed the KBD robot used Socratic questioning, slide-grounded sourced claims, and explained feedback far more often, while the control robot produced false statements and never cited sources¶

Claim · All claims
Evidence · 1 study (3 entries) · causal r2 · q2
Currency · newest study 2026 · checked against its sources 2026-10-20 (automated, 3 of 3 entries) · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? Teachy Mini linked 81.5% of its claims to the lecture slides while the control robot never provided sources. → Stephan Vonschallen 2026 q2 i? The control robot made ten factually wrong statements (6.1% of its messages) whereas Teachy Mini provided no false information. → Stephan Vonschallen 2026 (2) q2 i? Teachy Mini gave corrections or affirmative statements in 61.5% of messages versus 9.8% in the control condition. → Stephan Vonschallen 2026 (3)

Evidence¶

Stephan Vonschallen 2026¶

Stephan Vonschallen, Karim Kaufmann, Dominique Oberle, Friederike Eyssel, Theresa Schmiedel. (2026). Teachy Mini: Development and Preliminary Evaluation of a Knowledge-Based Generative Social Robot for Higher Education. arXiv. https://arxiv.org/abs/2607.22345 doi:10.48550/arxiv.2607.22345

q2 · i? · causal · r2

Qualitative content analysis of three KBD and three control interaction transcripts (655 coded segments; 164 KBD vs 163 control robot messages), coded by one researcher with consensus review by a second.

"In addition, while Teachy Mini explicitly linked 81.5% of its claims to the lecture slides, the robot in the control condition, which did not have this contextual information, never provided sources."

Stephan Vonschallen 2026 (2)¶

Stephan Vonschallen, Karim Kaufmann, Dominique Oberle, Friederike Eyssel, Theresa Schmiedel. (2026). Teachy Mini: Development and Preliminary Evaluation of a Knowledge-Based Generative Social Robot for Higher Education. arXiv. https://arxiv.org/abs/2607.22345

q2 · i? · causal · r2

Same transcript analysis: the authors report the control robot "made ten statements that were factually wrong" while Teachy Mini "did not provide any false information itself", though in five of 15 wrong-answer instances it should have corrected more assertively.

"On the other hand, the robot in the control condition made ten statements that were factually wrong (i.e., errors in 6.1% of all messages), which highlights the danger of LLM hallucination and false information in educational contexts"

Stephan Vonschallen 2026 (3)¶

Stephan Vonschallen, Karim Kaufmann, Dominique Oberle, Friederike Eyssel, Theresa Schmiedel. (2026). Teachy Mini: Development and Preliminary Evaluation of a Knowledge-Based Generative Social Robot for Higher Education. arXiv. https://arxiv.org/abs/2607.22345

q2 · i? · causal · r2

Same qualitative manipulation check: roles reversed across conditions — in the KBD condition students answered Socratic questions first, while in the control condition students mainly asked and the robot explained, which the authors call generally less effective.

"Relatedly, as the student in the KBD condition provided more answers, Teachy Mini also gave more feedback by providing corrections or affirmative statements (KBD: in 61.5% of all messages; CTRL: in 9.8%)."

Discussion¶

Related Claims¶

• Students perceive a KBD-configured generative social robot tutor as more aligned with responsible tutoring behavior than an identical non-KBD control robot — related
• Participants required GSR context-knowledge of learning materials, educational strategies, and course information, with optional physical-environment perception — related
• Twelve semi-structured interviews with university students and lecturers identified twelve knowledge-based design requirements for tutoring GSRs across three knowledge types — related
• KBD configuration produced no significant between-condition differences in system acceptance, intrinsic motivation, or subjective learning effectiveness in a single tutoring session — related
• An exploratory path model suggests perceived responsible behavior mediates the association between KBD and both learner motivation and acceptance — related