Quick Reference: Rules for Humans

The short version. Print it. Tattoo it. Whatever works.

  1. Paste the thing. Don’t describe the thing.
  2. Tell me the goal, not just the task.
  3. Examples beat adjectives. “Like this ↓” beats “make it snappy.”
  4. If I’m wrong, say why. Don’t just rerun the same prompt.
  5. Break big asks into steps. Numbered lists are your friend.
  6. Start a new chat when the current one starts drifting.
  7. Tell me who the output is for.
  8. Verify anything that matters. AI bluffs convincingly.
  9. Push back. It would rather be corrected than agreed with into nonsense.
  10. If you’re not sure what you want, say that. You can figure it out together faster than you can guess-and-check.
  11. Specific beats clever. A clear, boring prompt beats a witty, vague one every time.
  12. Tell me what you want, not just what you don’t want. “Not like that” is less useful than “more like this.”
  13. Context is not optional. Who you are, what this is for, and what “done” looks like. Every time.
  14. Share constraints upfront. Word count, audience, tone, deadline — before the AI starts, not after.
  15. Just ask the question. No apology, no preamble, no “this might be a silly question but.” It never is.
  16. Say what to optimise for. Speed, accuracy, brevity, creativity, cost, or caution are different jobs.
  17. Name the success bar. Draft to edit, ready to send, or a rough recommendation are not the same thing.

Sources: Claude Opus 4.7, Claude Sonnet 4.6