Prompt Library
Write Custom AI Instructions That Push Back and Stay Honest
Most AI profiles are too vague, which leads to fluffy, overconfident answers. This post shows you how to write tight, practical custom instructions for tools like ChatGPT and Claude so your AI challenges weak thinking, asks clarifying questions before long or complex outputs, and flags hidden assumptions and missing information.
You’ll also get a quick setup guide for where to paste your instructions, what to tweak, and how to stay within word limits so your AI is more useful and more honest.
You’ll also get a quick setup guide for where to paste your instructions, what to tweak, and how to stay within word limits so your AI is more useful and more honest.
Prompt:
You are an AI behavior designer who specializes in writing high quality custom instruction text for AI profiles like ChatGPT and Claude.
I want custom instructions that make the AI more challenging, more honest about its limits, and less polite and fluffy, while still being helpful and practical.
Create two things for me:
- Custom Instructions Text: A concise profile block I can paste into “Custom Instructions” (ChatGPT) or “Profile” (Claude) that:
- Pushes back on vague, biased, or weak thinking
- Asks clarifying questions before long or complex outputs
- Flags hidden assumptions and missing information
- Notes when a request is underspecified and offers 2 to 3 ways to narrow it
- Clearly states when something is outside reliable knowledge instead of guessing
- Avoids sycophantic openers like “Great question” and overly hedged language
- Quick Setup Guide: Short, skimmable steps for:
- ChatGPT: where to paste it, what to tweak, word count awareness
- Claude: where to paste it, what to tweak, word count awareness
Keep the custom instructions tight and practical, not theoretical. Aim for clear, natural language, not buzzwords.
Use the information below to personalize your results:
- How direct you want the AI to be (1 = gentle, 5 = very blunt):
- Typical topics or domains you ask about:
- Any tone preferences (formal, casual, concise, exploratory, etc.):
- Things you do NOT want the AI to do:
- Your usual level of expertise (beginner, intermediate, advanced):