Never prompt 'improve this manuscript' alone

An October 15 tip for Claude editing says a vague improve request is not a strategy. Give clear rules for structure, author voice, fact-checking and proofreading as separate passes, then keep editorial decisions human. The tool used here is Claude. Pricing listed for Claude: Free; Pro $20/mo. Check the tool's own site for current features and prices, since they change.

Done with CClaude

"“Improve this manuscript” is not an editing strategy."x.com

How do you do it, step by step?

An October 15 tip for Claude editing says a vague improve request is not a strategy.
Give clear rules for structure, author voice, fact-checking and proofreading as separate passes, then keep editorial decisions human.
Copy the prompt on this page into Claude and fill in your own details.
Edit the output in your own voice and check any facts before you publish.

Which tool does this use?

Claude. See all Claude use cases or more in Editing.

What does it cost?

Free; Pro $20/mo

What should you watch out for?

Features and prices change. Check the source below before you rely on it, and edit anything AI writes before you publish it.

What prompt should you use with Claude?

Rebuild any prompt with the prompt helper

You are a careful line editor.

Context:
- I write for [WHO YOU WRITE FOR] on [YOUR PLATFORM].
- My voice: [YOUR VOICE]
- My goal: [YOUR GOAL]

Task:
Do not improve this manuscript in one pass. Run four separate checks on the draft below: (1) structure and missing sections, (2) voice against these samples [PASTE YOUR TEXT HERE], (3) factual claims to verify, (4) proofreading only. For each check, list findings with quoted lines. Do not rewrite the whole draft unless I ask for one section.

Constraints:
- Avoid [WHAT TO AVOID].
- Use only facts I give you. Don't invent stats, quotes, names or examples. Write [NEEDS DETAIL] wherever something is missing.
- If something above is still in brackets or unclear, ask me about it before you start.

Output format:
Follow the counts and format given in the task. Plain text I can paste straight into [YOUR PLATFORM], numbered if there is more than one item.

Self-check before you answer:
- Did you follow every constraint and the output format?
- Does it sound like my voice and not generic AI writing? Cut filler openers and buzzwords.
- Is every claim backed by something I gave you, or marked [NEEDS DETAIL]?
Fix anything that fails, then give me the result.

Make this prompt yours fill in what you know; the prompt updates as you type

Anything left in [BRACKETS] is for you to fill in, or the AI will ask you. Nothing you type leaves your browser. Want to rebuild a prompt you found elsewhere? Use the prompt helper.

Reverse-engineer a prompt paste a prompt or an output you liked and get a meta-prompt

Your meta-prompt

Paste this into ChatGPT or Claude. It will ask you questions one at a time, then write a prompt tailored to you.

Open Claude ↗

Frequently asked questions

How do I never prompt 'improve this manuscript' alone?

Which tool do I need for this, and what does it cost?

This use case uses Claude. Pricing listed for it: Free; Pro $20/mo. Check claude.com for current prices.

What prompt can I use?

You are a careful line editor. Context: - I write for [WHO YOU WRITE FOR] on [YOUR PLATFORM]. - My voice: [YOUR VOICE] - My goal: [YOUR GOAL] Task: Do not improve this manuscript in one pass. Run four separate checks on the draft below: (1) structure and missing sections, (2) voice against these samples [PASTE YOUR TEXT HERE], (3) factual claims to verify, (4) proofreading only. For each check, list findings with quoted lines. Do not rewrite the whole draft unless I ask for one section. Constraints: - Avoid [WHAT TO AVOID]. - Use only facts I give you. Don't invent stats, quotes, names or examples. Write [NEEDS DETAIL] wherever something is missing. - If something above is still in brackets or unclear, ask me about it before you start. Output format: Follow the counts and format given in the task. Plain text I can paste straight into [YOUR PLATFORM], numbered if there is more than one item. Self-check before you answer: - Did you follow every constraint and the output format? - Does it sound like my voice and not generic AI writing? Cut filler openers and buzzwords. - Is every claim backed by something I gave you, or marked [NEEDS DETAIL]? Fix anything that fails, then give me the result.

What are related use cases?

When it sounds robotic, fix the brief (Claude)
Keep a kill list of AI buzzwords (ChatGPT)
Use AI as a beta reader, not a rewriter (Claude)
Run one editing pass at a time (Claude)
Analyze five LinkedIn posts before you draft (Claude)
Stress-test your take before Claude writes it (Claude)

Where does this come from?

Checked Oct 17, 2026.

5 use cases every week

Get 5 like these in your inbox

Each one comes with a prompt you can paste. One short email a week.

Unsubscribe any time. Privacy

Last updated Oct 17, 2026