How to Use AI to Write Better Emails (10 Prompts That Actually Work)
I write a lot of emails I dread. Here are 10 AI prompts I actually use for professional emails — follow-ups, cold outreach, the mistake apology, and the ones I keep putting off.
I once spent twenty minutes rewriting the first line of an email to a client who'd ghosted me. Twenty minutes. The actual content of the email took ninety seconds. The opening sentence is almost always the part that breaks me, and it turns out that's exactly the part AI is good at — that, and fixing the second draft, sanding down a tone that's gone a little sharp, and translating "what I actually want to say" into "what I can reasonably hit send on."
So here are 10 prompts I reach for most weeks. I've noted when each one actually pulls its weight, because not all of them do all the time.
One rule before any of this
Give the AI context, not just instructions. This is the whole game. "Write me an email" gets you a generic email — the kind that sounds like it was emailed from a corporate lobby. "Write me an email to my client James, who missed last week's deadline, and I want to push back without burning the relationship" gets you something you can use.
Everything below is a template. Fill in the brackets with the real stuff.
1. The "I don't know how to start" email
Write a polite email to [person] about [situation]. Tone: [friendly / formal / firm]. Length: short. End with a clear next step.
For when you're just staring at the compose window losing. The "clear next step" line is doing more than it looks like — without it the AI tends to trail off into a soft, non-committal ending, and you end up sending an email that asks for nothing.
2. The follow-up nudge
Write a follow-up email. Original message was sent [time ago] asking [what]. I don't want to sound annoyed but I do need a response. Keep it under 80 words.
Second or third attempt to get a reply. The word limit is the trick here. Long follow-ups read as passive-aggressive no matter how nicely you phrase them — the length itself is the accusation.
3. The "I made a mistake" email
Write an email acknowledging that I [what happened]. Tone: accountable but not over-apologizing. Explain what I'm doing to fix it. Don't make excuses.
Missed deadline, wrong file attached, a commitment you dropped. "Not over-apologizing" is the constraint that matters. Left to its own devices the AI will apologize four times in three sentences, which reads as weak and somehow makes the whole thing feel worse than it was.
4. The boundary-setting email
Write an email declining [request]. Reason: [why]. Keep the relationship intact. Offer an alternative if there's a reasonable one.
Saying no to a meeting, a project, a vendor. Without "keep the relationship intact," the default output comes out cold — technically polite, emotionally a door closing.
5. The cold outreach
Write a cold email to [person/role] at [company]. My ask: [specific thing]. What I offer in return: [specific thing]. Keep it under 100 words and don't open with "I hope this finds you well."
Sales, partnerships, networking. Banning the cliché opener is not optional. I'm convinced most AI-written cold emails get deleted in the first two seconds because they all open the exact same way and the reader's brain has learned to flinch.
6. The translation email
Here's a message I want to send: [paste your draft]. Rewrite it to sound [more professional / less aggressive / warmer / more confident]. Keep all the facts the same.
This is the one I use most, honestly. You write the draft while you're irritated, then you hand it over and ask for the diplomatic version. Letting the AI rewrite your draft keeps your intent intact. Asking it to start from scratch loses whatever made it sound like you in the first place.
7. The status update
Write a status update email to [audience]. Progress so far: [bullet list]. Blockers: [bullet list]. Next steps: [bullet list]. Tone: confident and clear, not defensive about the blockers.
Weekly updates, project syncs, the manager check-in. Feed it bullets and you get back structure. Feed it a paragraph of stream-of-consciousness and you get back a longer paragraph of stream-of-consciousness.
8. The "I'm leaving" email
Write a resignation email to my manager [name]. Last day: [date]. Reason (high level): [reason]. Keep it gracious and forward-looking.
Resigning, leaving a contract, ending a partnership. "Gracious and forward-looking" is the right register for something that's going to live in a file somewhere forever.
9. The complaint email
Write a complaint email to [company] about [issue]. I want [specific resolution]. Firm but not angry. Reference any relevant facts: [order number, date, prior contact].
Refund requests, customer service escalations, the thing that arrived broken. "Specific resolution" is what gets results — it forces the AI to name the exact thing you want instead of vaguely gesturing at your dissatisfaction. Vague complaints get a form reply. Concrete asks get refunds.
10. The thank-you that doesn't sound fake
Write a thank-you email to [person] for [what they did]. Make it specific — reference the actual thing they did and why it mattered. Keep it under 60 words.
After interviews, favors, intros, gifts. The word limit plus "specific" is what saves it from the generic thank-you sludge — you know the kind, the one that could've been sent to anyone for anything.
How I actually run these
Open a free AI chatbot — Smillee AI works without signup if you just want something fast — paste a prompt, swap in the bracketed parts, and then read the output like you don't trust it. Because you shouldn't, entirely. Always edit before sending. AI gets you maybe 80% of the way there. The last 20% is the part that sounds like you, and there's no prompt for that.
A few habits that quietly make everything better:
- Give it the recipient's name and role. "Write to my manager Sarah" beats "write to my boss" every time.
- Tell it what NOT to do. "Don't start with 'I hope this finds you well'" is one of the highest-leverage lines you can add. Negative instructions work shockingly well.
- Set a word limit. Short emails get replies. Long emails get archived with the best intentions.
- Read it aloud. If it's not something you'd actually say out loud, it's not done.
The meta-prompt
When none of the ones above fit — and that happens — this is my catch-all:
I need to write an email about [situation]. The audience is [who]. I want them to [desired outcome]. Constraints: [tone, length, anything to avoid]. Draft 2 versions — one direct, one diplomatic.
Asking for two versions does something weird and useful: by the time I've read both, there's almost always a third option in my head that's better than either of them. The drafts aren't the answer. They're the thing that shakes the answer loose.
Try it now
Want to test these? Open Smillee AI, paste any of the prompts above with your real context filled in, and see what comes back. No signup needed — just type and go.
— Maya
I'm Maya — I write most of what you'll read here. I spent years as a copywriter before I got a little obsessed with what these AI tools can actually do, so now I spend my days poking at chatbots, breaking them, and writing up what's worth your time. Everything here is something I've actually tried. If a prompt didn't work for me, it doesn't make the cut.
Want to try any of this?
Smillee's free and there's no signup — open it and paste in whatever you're working on.
Start chatting →More from the blog
- Trends
Full-Duplex, Multi-Model, and Mainstream: Three Chatbot Trends Defining July 2026
OpenAI shipped a voice model that talks and listens at the same time, the model market got too competitive to bet on one vendor, and conversational AI finished moving from pilot to core operations. What each one means for what you ship next.
- Trends
Agentic Commerce, Observability, and the Trust Gap: What Changed in Chatbots This Summer
AI agents can now buy things, and teams finally have real tools to watch what their agents are doing. But users still want a human on standby. Here is what that combination means for anyone shipping a chatbot in mid-2026.
- Trends
From Answers to Actions: Four Shifts Reshaping Chatbot Architecture in Mid-2026
Agentic loops, small on-device models, retrieval-augmented generation, and hyper-personalization are no longer experiments — they are the new baseline. Here is what each one demands from your stack.