Sycophantic AI feedback and disconfirming evidence search among business students

Abstract Generative AI may influence learning by shaping which evidence students choose to encounter, even when it does not determine their final answer. This study examines whether sycophantic feedback is associated with reduced search for information opposing an initial business judgement and whether perceived judgement correctness accounts statistically for that relationship. A questionnaire-based decision task conducted in September 2025 yielded 405 retained behavioural records from students at 16 Chinese universities. Participants received one of two assigned feedback conditions and opened four of eight materials concerning a fictional business expansion decision. The sycophantic group allocated 36.7% of its openings to opposing materials, compared with 47.5% in the evidence-based comparison group, an unadjusted difference of − 10.9 percentage points (95% confidence interval − 15.5 to − 6.2). An exploratory conditional choice model that respected the four-of-eight constraint and adjusted for material preferences within initial positions yielded an opposition-related odds multiplier of 0.458 (95% confidence interval 0.326–0.645). In the common 297-participant path-analysis sample, under the principal ordinary-least-squares model, the estimated indirect component through post-feedback correctness was − 0.015 (bootstrap 95% confidence interval − 0.031 to − 0.002). This pathway was sensitive to school clustering and modest hypothetical residual confounding; the auxiliary self-report contrast was not significant. The findings locate a potential cost of affirming AI feedback at the evidence-acquisition stage. They support an account of altered information allocation more strongly than a uniquely identified psychological mechanism or a general decline in critical-thinking ability.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-12
DOI
https://doi.org/10.21203/rs.3.rs-11351646/v1
Primary Topic
Decision-Making and Behavioral Economics
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Sycophantic AI feedback and disconfirming evidence search among business students

Pengfei Li
Research Square
Decision-Making and Behavioral Economics
preprint

Sycophantic AI feedback and disconfirming evidence search among business students

Pengfei Li
preprint en

Abstract

Abstract Generative AI may influence learning by shaping which evidence students choose to encounter, even when it does not determine their final answer. This study examines whether sycophantic feedback is associated with reduced search for information opposing an initial business judgement and whether perceived judgement correctness accounts statistically for that relationship. A questionnaire-based decision task conducted in September 2025 yielded 405 retained behavioural records from students at 16 Chinese universities. Participants received one of two assigned feedback conditions and opened four of eight materials concerning a fictional business expansion decision. The sycophantic group allocated 36.7% of its openings to opposing materials, compared with 47.5% in the evidence-based comparison group, an unadjusted difference of − 10.9 percentage points (95% confidence interval − 15.5 to − 6.2). An exploratory conditional choice model that respected the four-of-eight constraint and adjusted for material preferences within initial positions yielded an opposition-related odds multiplier of 0.458 (95% confidence interval 0.326–0.645). In the common 297-participant path-analysis sample, under the principal ordinary-least-squares model, the estimated indirect component through post-feedback correctness was − 0.015 (bootstrap 95% confidence interval − 0.031 to − 0.002). This pathway was sensitive to school clustering and modest hypothetical residual confounding; the auxiliary self-report contrast was not significant. The findings locate a potential cost of affirming AI feedback at the evidence-acquisition stage. They support an account of altered information allocation more strongly than a uniquely identified psychological mechanism or a general decline in critical-thinking ability.

Research Square
Zhengzhou Business University (CN)
Decision-Making and Behavioral Economics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon