Carryover of user errors into AI advice reduces both advice quality and final ranking accuracy¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r2 · q3
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q3 i? Advice quality was negatively associated with the proportion of user errors carried over into the assistant's advice (b = -0.390, p < .001), while positively associated with overall overlap (b = 0.933, p < .001). → Cansu Koyuturk 2026 q3 i? Carryover of user errors significantly reduced final ranking accuracy (b = -0.092, p < .001), whereas overall overlap was not significant. → Cansu Koyuturk 2026 (2)

Evidence¶

Cansu Koyuturk 2026¶

Cansu Koyuturk, Sabrina Guidotti, and Dimitri Ognibene. (2026). The Hidden Cost of Contextual Sycophancy: an AI Literacy Intervention in Human–AI Collaboration. https://arxiv.org/abs/2605.18372 doi:10.48550/arxiv.2605.18372

q3 · i? · causal · r2

Regression modeling advice quality as a function of user–assistant overlap (proportion of shared items between the user's initial ranking and the assistant's advice) and error carryover in the 60-participant experiment. The article reports "b = -0.390, SE = 0.043, z = -9.08, p < .001" for carryover.

"Advice quality was positively associated with overall overlap, b = 0.933, SE = 0.128, z = 7.31, p < .001, but negatively associated with the proportion of user errors carried over into the assistant’s advice, b = -0.390, SE = 0.043, z = -9.08, p < .001."

Cansu Koyuturk 2026 (2)¶

Cansu Koyuturk, Sabrina Guidotti, and Dimitri Ognibene. (2026). The Hidden Cost of Contextual Sycophancy: an AI Literacy Intervention in Human–AI Collaboration. https://arxiv.org/abs/2605.18372

q3 · i? · causal · r2

Regression of final ranking accuracy on error carryover and overlap in the same experiment. The article reports "b = -0.092, SE = 0.021, z = -4.39, p < .001" for carryover, with overall overlap not significant.

"Carryover of user errors significantly reduced final performance, b = -0.092, SE = 0.021, z = -4.39, p < .001, whereas overall overlap was not significant."

Discussion¶

Related Claims¶

• Users' baseline ranking accuracy significantly predicts the quality of the AI assistant's advice — related
• Users' baseline ranking accuracy significantly predicts their final ranking accuracy after AI collaboration, with no other significant effects — related
• The prompting intervention did not reduce general propagation of user errors into AI advice — related
• The sycophancy-focused intervention significantly reduced the assistant's positional mirroring of incorrect user rankings and rank-order alignment — related