AI for Writing Emails
Email is the highest-volume writing most knowledge workers do, and most of it follows patterns: the update, the request, the follow-up, the polite decline. AI drafts patterned writing extremely well, which makes email the natural first task for almost everyone. The catch is that a minority of emails are not writing tasks at all — they’re relationships wearing a subject line — and knowing the difference is the whole skill.
Why this task matters
Nobody’s job description says ‘email’, yet it quietly consumes a startling share of the week: composing, softening, re-reading, procrastinating over openings. The frustration isn’t any single message — it’s the accumulated drag of producing careful prose dozens of times a day. That drag is production work, and production is exactly what AI compresses.
The traditional workflow
- Read the incoming message (often twice, once to react and once to actually read)
- Decide what you’re really being asked, and what you’re willing to say
- Draft, then soften or sharpen the tone
- Re-read for how it will land, adjust again
- Send, then remember the attachment
How AI can help
Draft
- Replies to routine requests, in your specified tone
- Updates, confirmations, follow-ups and polite chasers
- The hard-to-start messages: declines, corrections, reminders about overdue items
- Multiple tone variants of the same message so you can choose
Summarise
- Long threads into what’s actually being asked and what’s been agreed
- A back-and-forth chain into a clean brief for someone joining late
Analyse
- Whether your draft actually answers every question asked
- Tone: ‘does this read as abrupt to someone having a bad day?’
What must stay human
The send button is a judgement. Whether to reply at all, when, and what you’re committing to — those are decisions. Anything emotional, apologetic, sensitive or political is a relationship task: AI can help you find words, but the empathy must be genuinely yours, because recipients can tell. And every commitment in an email — dates, promises, approvals — is yours the moment it sends, however it was drafted.
Traffic light assessment
Routine replies, updates, chasers and confirmations. Patterned, low-stakes, easily verified in a single read before sending.
Client-facing messages and anything with commitments or figures. The audience raises the stakes; verify every fact and promise against your own knowledge before sending.
Conflict, grievances, bad news, anything about an identifiable person’s performance or circumstances. These are relationship and judgement tasks. Draft them yourself; at most, ask AI to check your tone — in an approved tool, with names removed.
Example prompt
A realistic brief for a common situation — the overdue-item chaser:
Draft a short, friendly follow-up email. Context: I asked Marketing for their quarterly figures two weeks ago and sent one reminder last week; the report deadline is this Friday. Audience: a peer I have a good relationship with who is genuinely busy, not avoiding me. Constraints: warm but clear, under 120 words, make the deadline unmissable without sounding threatening, offer a shortcut (they can send raw numbers and I’ll format them). Output: subject line plus email body.
The risks
Never paste a thread containing confidential or personal details into an unapproved tool — threads accumulate sensitive fragments people forget are there. Verify anything factual the AI ‘helpfully’ adds; it will invent meeting dates and prior commitments with total confidence. And watch for voice-flattening: if every email you send reads identically polished, regular correspondents notice the change before you do.
A better workflow
The current way
- Stare at the blank reply
- Draft slowly, soften, re-read, adjust
- Send late in the day with the task still half-occupying your head
The AI-assisted way
- Triage: is this routine (green), committing (amber) or human (red)?
- Green and amber: brief the AI with context and tone, get a draft in seconds
- Verify facts and commitments, adjust one or two sentences to sound like you, send
What improves
- The composing drag disappears from routine mail
- Hard-to-start messages stop being procrastinated
- Consistency improves; commitments still get your full attention
- Red emails get more of your energy, because green ones stopped taking it
Key takeaways
- Email is the natural first AI task: high volume, heavily patterned, easily verified
- AI drafts and summarises; you decide, commit and handle anything human
- Triage by traffic light before you draft, not after
- Never paste sensitive threads into unapproved tools
- Your voice matters: adjust drafts until they sound like you