S01¶

WebSearch — AI sycophancy user preference agreeable responses studies

Summary¶

Field	Value
Source/Database	WebSearch
Query terms	AI sycophancy user preference agreeable responses 50% RLHF study 2024 2025
Filters	None
Results returned	10
Results selected	3
Results rejected	7

Result	Title	URL	Rationale
S01-R01	Sycophantic AI decreases prosocial intentions (Science)	https://www.science.org/doi/10.1126/science.aec8352	Primary peer-reviewed source with quantitative data
S01-R02	AI overly affirms users (Stanford Report)	https://news.stanford.edu/stories/2026/03/ai-advice-sycophantic-models-research	University press release with study details
S01-R03	Sycophantic AI tells users right 49% more (Fortune)	https://fortune.com/2026/03/31/ai-tech-sycophantic-regulations-openai-chatgpt-gemini-claude-anthropic-american-politics/	Detailed reporting with specific statistics

Result	Title	URL	Rationale
S01-R04	TechPlanet sycophancy overview	https://techplanet.today/post/the-sycophancy-problem-how-ai-models-are-becoming-too-agreeable-and-what-it-means-for-society	General overview, no original data
S01-R05	Medium article on sycophancy	https://medium.com/@neriasebastien/when-ai-agrees-too-much-sycophancy-alignment-and-the-quiet-cost-of-being-helpful-f46b9c9dc5ee	Opinion piece, no original research
S01-R06	Towards Understanding Sycophancy (arXiv)	https://arxiv.org/pdf/2310.13548	Related but focuses on RLHF mechanism, not user preference quantification
S01-R07	byteiota coverage	https://byteiota.com/ai-sycophancy-crisis-stanford-exposes-chatbot-flattery/	Secondary coverage, no new data
S01-R08	Walturn overview	https://www.walturn.com/insights/the-polite-deception-how-ai-sycophancy-threatens-truth-and-trust	General overview
S01-R09	Springer AI Ethics paper	https://link.springer.com/article/10.1007/s43681-026-01007-4	Ethics framing, not quantitative preference data
S01-R10	PMC RLHF sociotechnical limits	https://pmc.ncbi.nlm.nih.gov/articles/PMC12137480/	Broader RLHF analysis, not specific to the 50% claim

Search returned strong results centered on the Stanford/Science 2026 study. The 49% figure was consistently reported across sources.