How to Use AI for Email
The exact prompts to draft replies, sound like yourself, and follow up without nagging, plus which tool to use for which message and a two-minute inbox workflow that ties them together.
Eddie Ochieng
September 28, 2026· Updated Oct 12, 2026

Most advice about "AI for email" stops at "use ChatGPT." That is useless, because the whole game is in how you ask. This is the practical version: the exact prompts I use, in the order I use them, plus which tool to reach for and how to wire them into a two-minute pass through your inbox. Copy the prompts, they are written to be pasted straight into ChatGPT, Claude or Gemini.
How to use this
Everything below works on the free tiers of ChatGPT, Claude or Gemini. Prices for the paid email tools were checked in September 2026. The prompts are the point, so take them.
Turn a Thread into a Reply You Would Actually Send
The highest-value move is turning a long thread into a finished reply. The trick is telling the model exactly what the reply must do, in one line, then constraining length and tone. Paste this into any assistant along with the thread.
Below is an email thread. Draft my reply.
What the reply needs to do: [e.g. politely decline the meeting but offer async
notes / confirm Tuesday and ask for the agenda / chase the missing invoice].
Length: under 120 words.
Tone: warm but direct. No corporate filler, and do not open with
"I hope this email finds you well."
--- THREAD ---
[paste the whole thread here]The line that does the work is "what the reply needs to do." A vague "reply to this" gets you vague, beige, forgettable email. One clear sentence of intent gets you something you can send after a glance.
Make It Sound Like You, Not a Template
The single biggest reason AI email reads like a robot is that you never showed it how you write. Adjectives like "professional but friendly" mean nothing to a model. Your actual last three emails mean everything. Paste a real sample.
Here is a sample of how I actually write, keep this voice:
[paste ~100 words from an email you really sent]
Now rewrite the draft below in that voice. Match my rhythm and word choices.
Do not make it more formal or more polished than the sample.
--- DRAFT ---
[paste the AI draft from the previous step]Run this once and you will feel the difference immediately. It is the same reason our writing tests found Claude best at mirroring a given voice, a sample beats any instruction.
The Follow-Up That Earns a Reply
Most follow-ups fail because they add nothing, "just checking in" tells the reader you had no reason to write but poked them anyway. A follow-up that works leads with a new reason to reply. Give the model that reason.
I emailed [name] on [date] about [topic] and have not heard back.
Since then: [what is new — a result you can share, a link, a deadline,
a genuine question].
Write a short follow-up (under 80 words) that opens with that new reason,
not "just checking in". End with one specific, easy-to-answer question.Concretely, the difference looks like this. The nag: "Hi Sam, just following up on my last email, let me know your thoughts." The version that earns a reply: "Hi Sam, since I wrote we shipped the feature you asked about, here is a two-minute demo. Worth a quick call this week?" AI drafts both in seconds. Your job is to feed it the second one’s reason to write.
Fix the Tone in One Line
Once you have a draft, you almost never need to start over. Reply to the assistant with a single instruction and it adjusts in place. Keep these to hand.
Make it 30% shorter.
Warmer, without being gushy.
Firmer, I need a decision by Friday.
Less salesy, more like a note to a colleague.
Cut the last sentence.Which Tool for Which Job, and How to Actually Use It
You do not need all of these. Match the tool to the message.
- Gmail with Gemini, for everyday replies. In the compose window, click "Help me write", give it a one-line intent, and edit. It is included in paid Google Workspace tiers, so there is nothing extra to buy. Best for routine mail where speed matters more than nuance.
- ChatGPT or Claude (free, or about $20/mo), for the hard, sensitive or high-stakes messages. Open a chat, paste the thread, and use the reply prompt above. A generalist beats a dedicated email AI here because you can hand it the full context and iterate until it is right.
- Superhuman ($30/mo Starter, $40/mo Business), for people who genuinely live in their inbox. It builds speed around AI, fast triage, Auto Drafts and Ask AI on the Business plan. A premium tool that is overkill for most.
- Grammarly (from about $12/mo), as a final tone-and-clarity pass across everything you send, wherever you send it.
Putting It Together, a Two-Minute Inbox Pass
Here is how the tools and prompts fit into one workflow rather than five disconnected tricks.
- Triage: let Gmail or Superhuman summarise any thread longer than a scroll, so you reply to the gist, not the whole history.
- Draft: for routine mail, Gmail’s "Help me write". For anything that needs judgement, paste the thread into ChatGPT or Claude with the reply prompt.
- Voice: if the message matters, run the voice prompt with your sample so it sounds like you.
- Tone: one-line tweak if needed, then a quick Grammarly pass for polish.
- Read it back and send. Always the read-back. It is the five seconds that stops you automating being ignored.
| Tool | Price | Best for |
|---|---|---|
| Gmail + Gemini | included in paid Workspace tiers | Everyday drafting and thread summaries in place |
| ChatGPT / Claude | free, or about $20/mo | Hard messages that need context and iteration |
| Superhuman | $30/mo Starter, $40/mo Business | Inbox power users who want speed and Auto Drafts |
| Grammarly | from about $12/mo | A tone and clarity pass on everything you send |
| lemlist / Instantly | about $30–59/mo | AI-personalised cold sequences (with care) |
+ Pros
- + A finished, on-voice reply from a long thread in seconds
- + Follow-ups that add value instead of nagging
- + One-line tone fixes instead of rewriting
– Cons
- – The default output is characterless until you feed it a voice sample
- – Autopilot cold sequences wreck deliverability fast
The Two Traps to Avoid
First, the robot voice. AI email is grammatically perfect and completely forgettable by default, so the voice prompt is not optional if the message matters. Second, autopilot. For cold outreach at volume, lemlist and Instantly run AI-personalised sequences, and they work, but full autopilot is where it collapses. A flood of near-identical AI emails from one domain tanks your deliverability, and a single broken merge field, the infamous "Hi {FirstName}," can burn a whole campaign. Draft with AI, review before every send, never let it run unattended into real relationships.
Common Mistakes
- Typing "reply to this" instead of one line of what the reply must do.
- Never giving it a voice sample, then wondering why it sounds generic.
- "Just checking in" follow-ups that give the reader no reason to answer.
- Running cold sequences on full autopilot until deliverability collapses.
- Sending the first draft without the five-second read-back.
These prompts are a specialised version of the habits in how to write better AI prompts. For the assistants themselves, see ChatGPT vs Gemini vs Claude, and for the wider team-overhead problem, best AI tools for running a team.
Email is one job among many. For the rest, from meetings to spreadsheets, see the best AI productivity tools.
FAQ
What is the best prompt to reply to an email with AI?+
Give it the thread plus one line of exactly what the reply must do, a length limit, and a no-cliché instruction. The full prompt is in the "turn a thread into a reply" section above, written to paste straight into ChatGPT, Claude or Gemini.
How do I stop AI emails sounding robotic?+
Paste about 100 words of an email you actually wrote and tell the model to keep that voice. A real sample mirrors you far better than any instruction like "professional but friendly." The voice prompt above does exactly this.
Which AI is best for email?+
For everyday mail, the Gemini AI built into Gmail, nothing extra to buy. For hard or sensitive messages, a generalist like ChatGPT or Claude, because you can give it context. Superhuman suits inbox power users who want speed.
How do I write a follow-up email with AI without nagging?+
Lead with a new reason to reply, a result, a link, a deadline, a real question, not "just checking in." Feed the model that reason. The follow-up prompt above is built around it.



