A different kind of tool.
If you came to Claude from ChatGPT, you brought habits with you. Short prompts. Wall-of-text dumps. Pasting and praying. That worked, sort of. It works less well here.
Claude is a different kind of tool. Once you see the difference, you stop fighting it and start using it. This guide is that mental shift in five minutes.
Claude reads what you paste. Not skims. Reads.
Drop in a 40-page PDF and ask a question about page 31. It will answer. Drop in a 50-message email thread and ask who's actually blocking the deal. It will tell you. ChatGPT can do this too, but most people use ChatGPT in tiny chat-bar bursts and never give it enough material to be useful.
Claude was built for the bigger version of that conversation. Use it that way.
What separates great answers from generic ones.
Give it the raw material
Don't summarize. Don't trim. Don't curate. Paste the whole brief, the whole thread, the whole transcript. Claude is good at finding the signal. You don't need to find it for it.
Tell it who reads it
A prompt that says "rewrite this email" gets generic output. A prompt that says "rewrite this for a first-time customer who hasn't heard of us" gets useful output. The reader is the most important variable.
End with a shape
Tell it what you want back. Three bullet points. Five versions. A 200-word summary. Two options ranked by risk. Without it, Claude guesses. It usually guesses wrong.
Two extra lines change everything
Reader plus shape, every time. Add style and avoid when they matter. Skip them when they don't.
The structure that works on almost everything.
You are reviewing this for [reader]. Raw material: [paste the thing] Deliverable: [shape, length, format]. Style: [tone, voice]. Avoid: [the thing you don't want].
Paste this into any new Claude conversation. Replace the bracketed parts.
This template looks simple because it is. The whole point is that you stop typing one-line prompts and start typing four-line prompts. That's the whole game.
Start to finish.
You have a long Slack thread you want to send to your boss as a clean update.
The good version takes 30 extra seconds to write. The output is something you can actually send.
Common traps.
- Asking Claude to "do the thing" without telling it who reads the thing.
- Pasting half the context. If you wouldn't summarize a brief for a coworker, don't summarize one for Claude.
- Accepting the first answer. Claude will revise. Tell it what's off and ask again.
- Writing one-word prompts. The longer your prompt, the shorter your work.