Make AI-written email not sound like AI: keep a banned-phrases list and have Claude check every draft against it
"Sound more human" is not something AI can check; a list is. The author pairs 9 common AI-email tells with a fix each, saves them as email-banned.md, and has Claude check every reply against it, rewrite, and say which rules it hit.
Start step 1 ↓The problem
Everyone now spots the "Hope this email finds you well" AI opener. Models learned to drop that line, but the AI tone moved to the sentences next to it: "Just circling back", "Let me know if you have any questions", "Great question!".
AI tools used
Tap a tool to see what it is and which other cases use it
What to prepare
- email-banned.md
How to do it
- 1
Start from 20 emails you sent
Mark every sentence that makes you cringe. The author found 9: warm-up pleasantries, "circling back", thanks for nothing, a vague "hop on a call", restating their email, headers and bullets in a two-line reply, "let me know if you have questions", "Great question!" cheerleading, and a pile of sign-offs.
- 2
Write one example and one fix for each
For example, replace "circling back" with what you actually need: "Did the contract arrive? Want me to resend it?". End with the next step instead of "let me know", and save it all as email-banned.md.
- 3
Have Claude check before you send
Put the list in your Claude project or chat and add this line to every draft request:
Prompt - 4
When you rewrite, say the real thing
The author warns that swapping in a nicer pleasantry is just new AI tone. Add each new tell you catch to the list, and reread as the recipient: whatever you would skip, they will too.