Email, Slack & hard conversations
Prompts for the writing you do every day — replies, follow-ups, saying no, chasing people, and the message you are dreading.
Give the model the thread and your rough intent in blunt words. Polishing your blunt intent is the job; inventing your intent is not.
The pattern that covers most email
Nearly every useful email prompt is the same shape: here is the thread, here is what I actually want to say in unvarnished words, make it land well.
The unvarnished part is the trick. Write your real feeling — “this deadline is not happening and I am annoyed they moved it twice” — and let the model handle diplomacy. If you pre-soften your intent, you get a reply that is vague about a thing you needed to be clear about.
Draft my reply to the thread below. What I actually want to say, bluntly: [your real, unfiltered intent] Relationship: [e.g. external client, six months in, generally good] Constraints: keep it under 120 words, no bullet points, warm but not apologetic. Do not thank them for their patience. Thread: --- [paste] ---
I sent the email below 8 days ago and got no reply. Write a follow-up that gives them an easy way to respond without embarrassment, restates the one thing I need, and includes a specific date. Under 60 words. No 'just bumping this' and no 'circling back'. Original email: [paste]
Saying no, pushing back, delivering bad news
This is where prompting pays for itself, because the hard part is not the words — it is being clear and kind at the same time while you are stressed.
The key input is the thing you are protecting. A no that explains what you are protecting reads as a priority; a no that explains nothing reads as a brush-off. Tell the model what it is.
Help me say no to the request below. Why I am saying no, honestly: [real reason] What I am protecting: [the commitment this would displace] What I can offer instead, if anything: [alternative, or 'nothing'] Relationship I want to keep: [context] Write it short, decisive, and without three paragraphs of apology. One clear no, one honest reason, one alternative if I have offered one. Request: [paste]
Below is a message I am about to send. Do not rewrite it. Tell me: 1. How the recipient will most likely read the tone. 2. The single sentence most likely to be misread, and how. 3. What I have left ambiguous that they will have to guess at. 4. If I am wrong or unfair anywhere, say so plainly. Message: [paste]
Inbox triage
Reading is most of email. A model that reads a thread you were CC'd into halfway through will tell you what you missed faster than scrolling will.
I was added to this thread late. In under 100 words: what is being decided, who disagrees with whom, and what — if anything — is being asked of me. If nothing is being asked of me, say so in the first line. Thread: [paste]
A note on voice
Drafts that sound nothing like you are worse than no draft, because you spend longer un-writing them than you would have spent writing. The fix is a saved voice sample: two or three of your own emails you liked, pasted in with “match this register.”
Keep them in a note. Paste them into the tools that support persistent instructions once, and you stop needing to paste them at all.