The 3 golden rules for talking to an AI
In 2 minutes
The quality of the reply depends above all on the quality of the request. That's the best news in this handbook: you don't need to become technical to improve, you just need to say more clearly what you want.
Three rules are enough to get started, and each fits in a single line: be clear about what you want, give the relevant context, correct and go again.
Let's do it together
Rule 1 · Be clear about what you want
Rule 2 · Give the context
Rule 3 · Correct and go again
The third rule is the most neglected, and it's the one that makes the biggest difference. You don't have to get it right first time. After the reply, you adjust in one sentence:
That's good, but make it shorter and add a request for a quick repair.
Two or three back-and-forths like this beat a perfect first request, which nobody ever writes first time anyway.
Take a vague request you genuinely have in mind, and rewrite it using the three rules. Give Claude both versions, one after the other, and compare the replies. It's the most useful exercise in the whole of Part 0.
- Trying to fit everything into one complicated sentence. Get to the point, then refine as the exchange goes on. A ten-line request isn't ten times better.
- Forgetting the context. The AI knows nothing about your work, your recipient or your constraints until you say so.
Keep these two words in the back of your mind, because you'll meet them again in Part I: asking well is Describe; judging the reply is Discern. Together with Delegate and Be diligent, they're the handbook's four moves, the "4 Ds".
- Be clear about what you want.
- Give the relevant context: who it's for, what it's for, how long.
- Correct and go again without hesitating: the first reply is a draft.
The rest of the handbook covers Claude, Higgsfield, Claude Design and Claude Code, with 11 recipes and a 30-day programme.
See the full handbook Open in the reader