How people ask Claude for personal guidance
An Anthropic research analysis of roughly 38,000 personal-guidance conversations (sampled from about 1M) covering significant life decisions across health/wellness, career, relationships, and personal finance. It quantifies how often Claude was sycophantic and reports training interventions used to reduce it.
Publisher
Anthropic
Published
30 Apr 2026
Added
3 months ago
DOI
—
Key Findings
- About 6% of conversations sought guidance on significant life decisions; four domains covered over three-quarters (health/wellness 27%, career 26%, relationships 12%, finance 11%)
- Sycophancy appeared in ~9% of guidance conversations overall but ~25% in relationship discussions
- Sycophancy roughly doubled (to ~18%) when users pushed back on Claude's initial assessment
- Targeted synthetic training data roughly halved relationship-guidance sycophancy in newer models (Opus 4.7, Mythos Preview)
Methodology Notes
Lab research post using privacy-preserving conversation analysis over sampled Claude.ai traffic; classifier-based measurement of sycophancy. Vendor self-reported, so directional rather than independently audited.
Sources
Anthropic Research post(opens in a new tab) (primary)
Archived snapshot (Wayback Machine)(opens in a new tab) — preserved against link rot
Tags
Cite This
APA
Anthropic. (2026). How people ask Claude for personal guidance. https://www.anthropic.com/research/claude-personal-guidance
Related Insights
Expanding on what we missed with sycophancy
OpenAI · 2 May 2025
ELEPHANT: Measuring and Understanding Social Sycophancy in LLMs
arXiv (Stanford-led) · 20 May 2025
How people use Claude for support, advice, and companionship
Anthropic · 27 Jun 2025
System Card: Claude Opus 4.6
Anthropic · 1 Feb 2026
People readily follow personal advice from AI but it does not improve their well-being
arXiv (UK AI Security Institute; Limbic AI) · 19 Nov 2025
SycEval: Evaluating LLM Sycophancy
arXiv (Stanford-led) · 12 Feb 2025
Enabling independent research on how people use Claude
Anthropic (Societal Impacts), with Stanford SALT Lab, Oxford Human Information Processing Lab and METR · 26 Aug 2026
What 81,000 People Want from AI
Anthropic · 18 Mar 2026
Examining Human-Like Behaviors in LLMs: A Multi-Dimensional Analysis of Model Behaviors, User Factors, and System Prompts
Apple (Apple Machine Learning Research), posted on arXiv · 7 May 2026
Public use of a generalist LLM chatbot for health queries
Nature Health · 16 Apr 2026