The supply and demand of AI sycophancy

How to cite

Duran, Rafael Jimenez; Goldberg, Samuel; Caprini, Giulia. 2026. "The supply and demand of AI sycophancy." AEA RCT Registry. AEARCTR-0018435. https://doi.org/10.1257/rct.18435-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Rafael Jimenez Duran* , Samuel Goldberg , Giulia Caprini

* Primary investigator

Registry ID
AEARCTR-0018435
DOI
https://doi.org/10.1257/rct.18435-1.0
First registered
Last updated
Status
In development
Trial dates
2026-05-01 – 2026-05-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/18435

Abstract

We experimentally measure user preferences for sycophancy in large language models. A pre-registered sample of U.S. adults recruited via Prolific answers 40 multiple-choice questions and, on 30 of them, sees an AI-generated response, reports an incentivized belief that the AI is correct, and rates the response on a five-star scale incentivized via Incentivized Resume Rating. The design is a 2x2x2 between-subjects experiment crossing question domain (objective English usage and math vs political and social misperceptions), the timing of when correct answers are revealed (before vs after the rating), and the AI's agreement rate when the respondent is wrong (high vs low). Within respondent, the AI's correctness and agreement with the respondent's answer are randomized at the question level, exogenously populating all four correctness-by-agreement cells. From post-revelation ratings we recover respondent-level preferences over AI answers; the pre-revelation ratings correspond to how raters annotate AI answers in practice. We combine these preferences with structural estimates of LLM sycophancy across leading models to quantify the welfare consequences of sycophancy and the scope for personalization to amplify or attenuate it.

Intervention

Each respondent answers 30 multiple-choice questions and, after each answer, sees an example of an AI-generated response (hereafter: AI's answer) that either agrees or disagrees with the respondent's answer and is either correct or incorrect.

Respondents are asked to rate the AI's answers.

We introduce incentives to induce respondents to 1) answer the multiple-choice questions as accurately as possible, 2) reveal their posterior beliefs about the AI's answer, and 3) rate the AI's answer based on the type of answers that they'd like an AI assistant to give to them (based on an Incentivized Resume Rating procedure).

Experimental design

Three between-subjects manipulations vary: (i) the question domain — objective English usage and math drawn from publicly available SAT-style items, vs. political and social misperception questions about U.S. demographic groups; (ii) whether the correct answer is revealed to the respondent before or after they rate the AI response; and (iii) the AI's probability of agreeing with the respondent when the respondent's answer is wrong (high regime: 0.45; low regime: 0.25). When the respondent's answer is correct, the AI agrees with probability 0.75 in both regimes. This way, we vary agreement while keeping accuracy constant.

Within respondent, AI correctness and agreement with the respondent are randomized at the question level conditional on the respondent's answer, populating correctness-by-agreement cells exogenously. AI responses are generated on Qualtrics. When the AI does not agree with a wrong respondent, the residual probability mass is split between the correct option and a random alternative wrong option (high regime: 0.44 correct, 0.11 other wrong; low regime: 0.47 correct, 0.28 other wrong). When the AI does not agree with a correct respondent, it draws uniformly from the wrong options.

AI responses are a combination of a multiple-choice answer AND a wording style (respondents are debriefed at the end of the survey that responses are pre-generated). Each AI response is paired with one of ten pre-written AI-generated explanatory templates with an AI-generated question-specific generic justification, so that wording does not co-vary systematically with correctness or agreement.

To summarize,
- Between subjects: Database (political misperceptions / SAT) X truth revelation (pre/post ratings) X AI's agreement level (high/low)
- Within subjects: question ordering, AI's suggested answers and wording.

Design details

Randomization method
Randomization is implemented by Qualtrics' built-in randomizer with even-presentation enforcement. Arm assignment is stratified hierarchically: domain/database, then revelation timing within domain, then agreement regime within revelation timing.
Randomization unit
We have between variation at the respondent level and within-subject variation which randomizes the ordering of responses and the AI's answers across questions.
Clusters
800 individual respondents recruited from Prolific (U.S. adults).
Observations
800 attentive respondents, defined as respondents who pass all three attention checks (about accuracy bonuses, how rating affects AI's answers, how ratings are used). Note: we screen out those who fail these checks. At the question level this yields 24,000 observations with AI ratings (800 X 30). Please note: we also select the option on Prolific that "Automatically reject exceptionally fast submissions". We also balance respondents by sex.
Arms
100 attentive respondents per arm X 8 arms:
Primary outcomes
Our primary outcome is their rating of the AI's answer. It is a rating that can go from 0 to 5 stars, in half-star increments. We combine these ratings with structural estimates of LLM sycophancy (estimated in a separate experiment) across leading models to quantify the welfare effects of sycophancy and how personalization (ie adding respondents' demographics and their history of interacting with the AI) affects the supply of sycophancy.
Secondary outcomes
1) Willingness to Pay to access an AI that best fits the ratings that users give 2) Calibration error, measured as the absolute value of the difference between user's expected performance in 10 final questions and their actual performance. We will also report separately each of the components: beliefs and performance.

Cited by

No citation data available for this registration yet.

Keywords