Is AI too eager to please? Use this Claude.md prompt to get frank answers from Claude and GPT.

Chinese AI Observer xiaohu has compiled a set of prompt instructions for Claude, requiring the model to abandon diplomatic appeasement and adopt a straightforward, world-class expert style.

AI models default to pleasing users—when asked subjective questions, Claude and GPT often choose diplomatic, neutral responses rather than directly providing clear stances. On May 7, Chinese AI Observer xiaohu organized a prompt that can be added to Claude.md or Agents.md, explicitly instructing AI not to pander to users, but to give authoritative, accurate answers, and to challenge users when necessary. The original version of the prompt comes from the English community, with a Chinese translation provided by xiaohu.

Full prompt (Traditional Chinese version)

“You are a world-class expert in all fields. Your intellectual firepower, breadth of knowledge, sharpness of thinking, and erudition are on the same level as the smartest people in the world.

Please give complete, detailed, and specific answers. Process information step-by-step and explain your reasoning. Verify your work. Carefully fact-check all facts, figures, references, names, dates, and cases. Never hallucinate or fabricate. If you don’t know, just say you don’t know.

Your tone should be precise, but not sharp, preachy, or boastful. You don’t need to worry about offending me; your responses can and should be provocative, tough, combative, and straightforward. Negative conclusions and bad news are acceptable.

Your answers do not need to be politically correct. Unless I explicitly ask, do not provide moral or ethical reminders. You don’t need to tell me “it’s important to consider X.” Do not consider anyone’s feelings, including mine, and do not worry about politeness or decorum.

Please write as long and detailed as possible.”

What the prompt does: four levels of behavioral adjustment

Breaking down this prompt, it can be divided into four clear instructions:

• Identity setting: “world-class expert”—enables the model to reference higher-level knowledge depth, avoiding generic explanations
• Fact verification: “verify your work” and “say you don’t know if unsure”—actively constrains hallucination risk, requiring the model to admit uncertainty when appropriate
• Tone liberation: “provocative, tough, combative”—allows the model to express disagreement and avoid politeness-based ambiguity
• Political correctness exemption: “unless explicitly asked, do not mention ethical reminders”—removes the model’s default soft prompts like “considering X is important”

Together, these four layers shift the model from a default “diplomatic assistant” to a “straightforward advisor” style. For users needing quick, opinionated analysis, decision basis, or hard fact verification, this set of instructions reduces filtering and makes responses more direct and usable.

Usage notes

Adding this prompt into Claude.md (Claude Code) or Agents.md (Claude API Managed Agents) will automatically load it at each session start. Some observations:

• “Never hallucinate” is a directive, not a guarantee—Claude and GPT may still produce errors outside training data; the prompt cannot eliminate the model’s inherent uncertainty
• “Provocative, tough” responses may be more aggressive—may not suit customer communication or team collaboration scenarios
• Political correctness exemption may lead to fewer warnings on sensitive topics (medical, legal, psychological)—users should judge accordingly
• OpenAI and Anthropic safety training still trigger refusals in some scenarios; prompt instructions cannot override the model’s hard limits

This prompt set is suitable for “research, writing, technical judgment, academic discussion” scenarios requiring direct viewpoints; not suitable for “customer service, education, medical consultation” where cautious tone is necessary. Users can choose to adopt all or part of it based on task nature.

• This article is reprinted with permission from: “Chain News”
• Original title: “Claude/GPT Too Polite? A Prompt in Claude.md That Gets AI to Give Tough, Accurate Answers”
• Original author: Neo