The Claude AI Community Discord is open.Join us
Guides/Claude Prompting Masterclass 101/Mistakes That Are Easy to Make and Easy to Fix

Mistakes That Are Easy to Make and Easy to Fix

10 min readMistakes That Are Easy to Make and Easy to Fix

Vague ask

Help me improve my landing page copy. — improve how? for what metric? for whom?

Fix: define what improvement means in this specific case. Make this more direct and remove anything that sounds like marketing speak. Keep it under 80 words.

Asking multiple things at once

One prompt, one question. Multi-question prompts get shallow answers to all of them.

No format specified

You'll get whatever format Claude chooses. Sometimes that's fine. Often it's not. Just say what you want.

Correcting without specifics

That's not right, try again doesn't tell Claude what was wrong. Be surgical: The second paragraph is too technical — rewrite just that part for someone non-technical.

Not telling Claude it's okay to say "I don't know"

Claude will fill gaps confidently if you don't give it permission not to. Add: If you're not sure, say so. I'd rather know the limit than get a confident wrong answer.

Accepting the first output

This is a habit worth breaking. One round of what's weak about this? fix those parts usually makes the output noticeably better. The first pass is a draft, not a deliverable.

Not saving prompts that work

When you find a prompt structure that works for something you do repeatedly — save it. A Notion doc, a text file, anything. It compounds.


Quick Reference

When this happensTry this
Output format is wrongAdd explicit format instructions
Too genericAdd role + specific context
Reasoning is offAdd "think step by step"
Style isn't matchingPaste an example of what you want
Building an automationXML tags + constrain JSON output format
Getting close but not rightUse few-shot examples
Not sure if output is rightAsk Claude to critique it
Same bad output keeps happeningUpdate your system prompt

Updated February 2026. Claude's capabilities evolve, so treat everything here as empirical — try it and see what happens.