Email is one of the highest-value, lowest-risk places to use AI regularly. You’re always the one reviewing before it sends, so the downside of a bad draft is basically zero.
Paste in the email you’re replying to. Say what you actually want, are you declining, negotiating, following up. “Write a reply” gets you something generic. “Write a reply declining this meeting but offering next Tuesday, keep it warm since we work together closely” gets you something you can send with minor edits.
Feed it two or three emails you’ve actually written and ask it to match that specific voice, not a description of your style. A real example beats “professional but friendly” every time.
Use it most for the emails you’re avoiding, delivering bad news, pushing back, asking for something uncomfortable. It gives you a draft to react to, which is often easier than staring at a blank message. And read it before sending, the same discipline covered in our guide to fact-checking AI content. AI drafts can misread tone or include a detail you didn’t mean to share, always treat the output as a starting point, never something to send unread.
See Anthropic’s prompt engineering documentation for more on giving useful context.
