You're using Claude the hard way

You're using Claude the hard way, and writing longer prompts won't fix it.

Three habits will, and none of them needs any technical skill.

First, when an answer comes back wrong, stop replying to it. Most people fire back corrections: make it shorter, no, warmer, try again. Every reply bloats the conversation, and a bloated conversation makes every answer after it sloppier.

Go back and edit your original prompt instead. You get a clean answer from a clean slate, usually on the second go.

Second, show Claude what you want instead of describing it. You could spend a paragraph explaining your tone of voice, or you could paste in two emails you actually sent and say "write it like these." AI copies far better than it follows instructions, so give it something to copy.

Third, stop briefing the same job from scratch every week. The next time you finish a job you know you'll repeat, tell Claude to turn it into a skill. It saves the whole process and runs it the same way every time.

In Cowork you can then put that skill on a schedule, every Monday morning for example, and it runs even with the app closed.

Getting good work out of AI takes the same habits as training a new hire. You show them examples, and you write the process down once so they can follow it without you.

Start with the first one: the next time Claude gets an answer wrong, edit the prompt instead of replying.