Prompt-based personalization of an LLM/RAG teaching assistant produces measurable variation in response expression, length, and complexity across personalization conditions¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · checked against its sources 2026-10-20 (automated, 1 of 1 entry) · not reviewed by a person

Subclaims¶

q2 i? Across 97 prompt configurations for 30 student questions, learner profiles were associated with measurable differences in LLM-generated response characteristics. → Saptarshi Basu 2026

Evidence¶

Saptarshi Basu 2026¶

Saptarshi Basu, Sandeep Kakar, Ashok Goel. (2026). A Prompt-Engineering Approach to Develop Scalable, Flexible, and Real-Time Hybrid Micro-Level Personalization in a General Purpose AI Teaching Assistant. https://arxiv.org/abs/2609.03402 doi:10.48550/arxiv.2609.03402

q2 · i? · design · r2

Automated NLP analysis of 2,910 responses generated with GPT-4.1 (temperature 0) for 30 real KBAI course questions under 96 personalized profiles plus one baseline, evaluated on lexical similarity, semantic similarity, complexity, and verbosity. The article reports "measurable vari- ation in response expression, length, and complexity".

"Overall, the NLP analyses demonstrate measurable vari- ation in response expression, length, and complexity across personalization conditions."

Discussion¶

Related Claims¶

• Higher self-assessment, verbosity, reflective processing, and technical abstraction are associated with greater response complexity, and Evaluation and Analysis questions tend to yield more complex responses — a narrower finding that bears on this claim
• Personalized responses remain semantically grounded in the shared knowledge base while varying in surface-level expression and structure — a narrower finding that bears on this claim
• Verbosity preference is associated with systematic variation in LLM response length — a narrower finding that bears on this claim
• In human evaluation, abstraction preference was associated with perceived response quality, and more complex questions generally received lower ratings — a narrower finding that bears on this claim
• Abstraction and processing preferences, but not self-assessment, were significant predictors of perceived response complexity, abstraction, and processing style — a narrower finding that bears on this claim
• Implicit demographic cues produce larger and less predictable effects in open-ended tasks, including longer responses for laptop users and education-related length effects — related