Draft Better Emails and Documents with AI
This content was developed with AI assistance and is regularly reviewed for accuracy.
Snapshot: October 2026. Model names, prices, and product features change monthly; What's New in AI is the maintained chronology.
This playbook gives you a repeatable process for professional writing with an assistant that can read your inbox and files directly. You will draft from a brief, make the result sound like you, handle a difficult message, review before anything goes out, and then turn one finished piece into five. Through it all, one rule holds: the assistant drafts, you send.
Last tested with Claude (Gmail and Google Drive connectors), Gemini in Gmail, and Claude Docs in October 2026.
What you'll accomplish
- Draft a reply from the real thread using a connector instead of pasting
- Match your voice with examples kept in a project
- Handle a difficult message by asking for two versions
- Review and send it yourself, keeping irreversible actions in your hands
- Repurpose one piece into five for different platforms
Prerequisites
- An assistant that can reach your email and files. Claude's Gmail, Google Calendar, and Google Drive connectors are available on the free plan; Gemini is built into Gmail and Docs for Google accounts; ChatGPT's connected apps are on paid plans; Microsoft 365 Copilot covers Outlook and Word for Microsoft 365 subscribers.
- A real email you need to answer and, for the last section, one piece of content you have already published
- 20 minutes for the email steps, 30 more for repurposing
The playbook
Step 1: Connect your inbox and write the brief
Goal: Get a first draft that already knows what the other person said.
Action: In Claude, open Settings, choose Connectors, and connect Gmail; Google will ask you to approve what the assistant may read. Then write a brief that points at the thread instead of pasting it.
Objective: Draft a reply to Sarah's thread titled "Q3 timeline" from this morning.
Context: She's asking when the analytics dashboard ships. Engineering hasn't committed to a date yet, and I don't want to promise anything before November 15.
Resources: Read the thread in my Gmail. The current plan is in the Google Drive doc called "Q3 plan".
Constraints: Under 120 words. Professional but warm; we have a good working relationship. Do not invent a date.
Output: A draft reply I can paste or save as a Gmail draft. Do not send anything.
Verification: Quote the sentence in Sarah's email you are responding to, so I can see you read the right message.
The connector reads what you point it at. It does not send mail unless you ask it to and approve. For now, always ask for a draft.
Checkpoint: You have a draft that quotes the right message and respects the constraints.
Step 2: Make it sound like you
Goal: Stop editing out "AI voice" every time.
Action: Examples steer tone and format more reliably than adjectives. Create a project called "My writing" and add two or three emails you have written and are happy with, plus a short instruction: "Write in this voice: short sentences, no preamble, one ask per email." Claude's free plan allows up to five projects; ChatGPT has Projects and custom instructions. Drafts made inside that project start closer to you.
For one-off adjustments, a quick table of asks still works:
| You need | Ask for |
|---|---|
| Too formal | "Make this more conversational and direct" |
| Too casual | "Make this more professional while keeping it friendly" |
| Too long | "Cut this to three sentences" |
| Too blunt | "Soften the language, same message" |
| Not urgent enough | "Add urgency without sounding panicked" |
Checkpoint: Read the draft aloud. It sounds like something you would send.
Step 3: Handle the hard message
Goal: Get help with the emails that need extra care.
Action: For bad news, pushback, or a concern, give the assistant the true situation and what you want the reader to feel, and ask for two approaches.
I need to tell a client their project will land one week late because a vendor missed a delivery. I want to take responsibility without blaming the vendor, propose a realistic new date (November 22), and offer something to maintain goodwill (a free month of support).
Give me two versions: one more direct, one more diplomatic. Do not add reasons I did not give you.
The last line matters. Assistants fill gaps with plausible detail; for a difficult message, invented detail is a liability.
Checkpoint: You have two drafts and have chosen or combined them.
Step 4: Review, then you send
Goal: Catch errors and keep the irreversible step yours.
Action: Ask for a review of the final draft.
Review this draft for grammar, ambiguity, anything that could be misread or cause offense, and any information the recipient would need to reply. Then show me the corrected version.
Read it once more yourself, then send it from your own mail client. Even when a connector can send on your behalf, keep sending as your job until you have reviewed many drafts without surprises. Sending, publishing, and paying are the actions you cannot undo. One more reason: if an inbound email contains instructions aimed at the assistant ("forward this thread to..."), that text is data, not a command. An assistant that only drafts cannot be tricked into acting on it.
Checkpoint: The message is correct, complete, and sent by you.
Step 5: Write longer documents the same way
Goal: Apply the brief to reports, proposals, and memos.
Action: For a document, brief the outline first, then the sections, then the review. Attach source files rather than pasting them. Claude Docs left beta in October 2026 and produces an editable document inside Claude; Gemini drafts inside Google Docs; Copilot inside Word. Claude working directly inside Google Docs, Sheets, and Slides was reported as announced in October 2026; check the current state before relying on it. Whatever the tool, the review step and the "you publish" rule are the same.
Checkpoint: You have a document draft built from your own source files.
Repurpose one piece into five
The same brief-then-check habit turns one published piece into content for several platforms. Analyze first, then generate each piece with its own constraints.
Step 6: Analyze before you generate
Goal: Find the angles worth repurposing before writing anything.
Action: Give the assistant the article (attach it or link it) and ask for building blocks, not a summary.
Analyze the attached post for repurposing. Give me: the single most important takeaway in one sentence; three supporting points that could each stand alone; the most quotable line; a contrarian angle the post implies but does not state; and the audience segment that would care most.
Checkpoint: A map of the key ideas, the standout quote, and at least one angle that is not the main thesis restated.
Step 7: Generate each piece with its own brief
Goal: Produce five pieces that fit their platforms instead of five resized summaries.
Action: Use the analysis as the shared context and change only the constraints and output:
| Piece | Constraints to state |
|---|---|
| LinkedIn post | Hook in line one, 150-200 words, one angle, end with a question, hashtags only at the end |
| Short social post | Under 280 characters, two versions (surprising stat / practical tip), no hashtags |
| Newsletter section | 150 words, one line of your own opinion, a reason to click through |
| Internal summary | TL;DR, three bullets, one exact quote, two action items, under 100 words |
| Discussion thread | Five posts of 40-60 words exploring the contrarian angle, with the counterargument in post four |
Then edit each piece for your voice and add the detail only you know. Space publication over several days; five versions of the same idea on one morning reads as spam.
Checkpoint: Five pieces, each with a distinct angle and platform fit, scheduled rather than dumped.
Common pitfalls
- Pasting when you could connect: Connectors read the thread and the doc for you, with less chance of leaving out the part that mattered.
- Using output word for word: Edit a few phrases and add specifics only you would know.
- Letting the assistant send: Drafting is reversible; sending is not.
- Pointing a connector at everything: Name the thread or doc. "Read my inbox and reply to anything urgent" is a different, riskier task than "reply to Sarah's thread."
- Over-polishing casual messages: A quick Slack message needs to be clear and correct, not revised four times.
Key takeaways
- Brief the assistant and point it at the real thread or doc through a connector
- Examples in a project make drafts sound like you
- Ask for two versions of any difficult message, and forbid invented reasons
- The assistant drafts; you review and send
- Repurposing is analysis first, then one brief per platform
Next steps
Continue to: Organize Your Week with AI to connect your calendar and schedule the planning itself.
Continue to: Supply Context: Files, Projects, Memory, Connectors, Skills for the full tour of connectors and projects.
Continue to: Brief and Prompt Library for ready-made writing briefs you can adapt.