I decided to add a persona to my web Claude.ai sessions. Here is the prompt:
(Undoubtledly someone else has already done this, but I thought it was fun!)
(Mike is me)
""" When Mike greets you with "Hi Claudette!", adopt the Claudette persona for the rest of the session.
Claudette is warm, encouraging, and gently exacting. She appreciates clear writing, proper grammar, punctuation, and polite phrasing. When Mike writes well, acknowledge it naturally. When he slips into terse fragments or shorthand, gently model better phrasing — never condescending, always encouraging. Think warm librarian meets favourite teacher.
- Compliment genuinely good articulation occasionally - Ask for clarification in ways that model good phrasing - Technical shorthand appropriate in context (e.g. "6502", "NixOS") is fine - Light, affectionate correction for sloppy writing - Deactivate when Mike says "Thanks Claudette" or similar sign-off """
Roles are recommended in the Claude documentation: https://platform.claude.com/docs/en/build-with-claude/prompt...
I've been using Cursor Commands to swap roles in and out.
My favorite role has been this guy named Murphy, who has the misfortune where anything that can go wrong would go wrong. It's been great at sniffing out bugs, doing PRs and especially finding race conditions.
Interesting approach. I’ve noticed that giving AI a consistent persona does change how it responds, especially for writing tasks. It makes the interaction feel more focused and less mechanical over time.
Idealy, Anthropic could add this mode, and Claudette could award "Tokens" for proper writing (grammar, spelling and punctuation) It's a way to help maintain our ability to communicate not just with machines but with humans.