If your inbox routinely runs three digits, you do not need a new email client — you need a triage discipline. This workflow combines a four-bucket classifier with AI-generated draft replies and a voice-edit pass, run in two daily 15-minute sessions. Built for execs, consultants, founders, and sales reps who get 50-200 emails a day; the goal is inbox zero by 6pm with replies that still sound like you, not like ChatGPT.
What this covers
A daily, repeatable 15-minute AI-assisted triage workflow: batch ten emails, classify each into reply-now / reply-this-week / FYI / delete, generate draft replies for the first two buckets, run a voice-edit pass, and ship. The output is a clean inbox and a saved “voice file” the AI uses to keep drafts sounding like you across sessions.
Who this is for
Heavy email recipients — executives, consultants, founders, sales reps, founders of small companies who still run their own inbox. Less useful for people receiving fewer than 30 emails a day (the overhead exceeds the benefit) and for legal, HR, or therapist roles where every email is privileged and must be drafted by hand.
When to reach for it
Daily, in two windows: 15 minutes mid-morning (after deep work but before noon meetings) and 15 minutes end-of-day (before you close the laptop). Avoid first-thing email — opening the inbox at 8am is the fastest way to lose the morning. Also run a longer triage on Monday morning to clear weekend buildup.
Before you start
- Build a “voice file” once: paste 5-10 of your real past emails into a doc with a header like “examples of how I write.” This is the single most important input for keeping drafts sounding like you.
- Decide your four buckets: most use reply-now / reply-this-week / FYI / delete, but customize. Sales reps might add “send to CRM”; founders might add “forward to assistant.”
- Configure an AI sidebar or shortcut: Gemini in Gmail side panel, ChatGPT keyboard shortcut, or Claude desktop. Switching tabs kills the 15-minute pace.
- Decide your “skip AI” rules up front: HR matters, legal, personal relationships, anything involving sensitive personnel data — always written by hand.
Step by step
- Open inbox, sort by date. Process newest first to handle time-sensitive items before they age.
- Process in batches of 10 emails. Larger batches break attention; smaller wastes the AI’s context warmth.
- For each email, paste subject + body into AI. Prompt: “Classify (reply now / reply this week / FYI / delete) and draft a reply if the bucket is reply-now or reply-this-week. Use my voice file as style reference.” Attach your voice file once per session.
- Read each draft aloud. AI drafts skew formal and overlong; cut to 50-70% of the AI’s word count. Adjust to sound like you. If a reply needs a fact you do not have, mark and skip.
- Send, schedule, or delete based on classification. Schedule replies for tomorrow morning if “reply this week” — gets them off your mind tonight.
- For FYI emails, archive or labelize, do not file in elaborate folders. Search is faster than folders.
- Repeat 2-5 cycles. Stop when the timer rings, not when the inbox is empty. Tomorrow’s session handles the rest.
- For sensitive emails, skip the AI entirely. HR, legal, personal — write by hand from a clean draft.
The voice file (one-time setup)
Voice notes for AI drafting:
- Default to 2-4 sentences; only longer if the topic demands it.
- Open with the answer, not pleasantries.
- Avoid "I hope this finds you well" and "circling back."
- Use first names; sign off with first name only.
- For asks: state the request, the why, and the deadline in
the first paragraph.
Examples of how I write:
[paste 5-10 real emails verbatim, with greetings/signatures]First-run exercise
- Process your next 10 emails using only the workflow.
- Time yourself. If you took more than 15 minutes for 10 emails, AI drafts are likely too long — tighten the prompt with “max 3 sentences.”
- Read the day’s sent items the next morning. Anything that doesn’t sound like you means the voice file needs more examples.
- Adjust one variable for tomorrow’s session — usually the voice file or the bucket definitions.
Quality check
- Do your sent replies sound like you, or like ChatGPT (em-dashes, “I’d be happy to,” three-clause sentences)? Cut those phrases.
- Are you sending facts you did not personally verify? AI sometimes adds confident filler. Cut anything the recipient might quote back to you.
- Did you skip the AI on truly sensitive emails? If you ran HR or legal through the AI, undo and rewrite.
- Are the buckets accurate? If most emails land in “reply this week” the bucket has become a holding tank — split into “reply-Mon” and “reply-Fri.”
How to reuse this workflow
- Keep your voice file as a Google Doc or Notion page; update it monthly with new exemplars.
- Save the classification prompt as an AI shortcut or text expander so you do not retype it.
- After two weeks, audit your sent folder. Common rewrites become new lines in the voice file.
- For recurring email types (intro requests, pricing questions, scheduling), build per-type templates the AI fills in.
Recommended workflow
Batch 10 emails -> AI classify + draft using voice file -> voice edit (read aloud, cut 30%) -> send / schedule / delete -> next batch. If you live inside Gmail, the Gemini in Gmail deep workflow shows how to wire this batching directly into the Gmail side panel using a “voice file” Doc.
Common mistakes
- Sending AI drafts without a voice-edit pass. Recipients notice; relationships notice; you sound like a bot.
- Using AI on sensitive emails (HR, legal, personal, layoffs, performance issues). Hand-write these every time.
- Trying to process all emails in one session. Fatigue makes triage worse; cap at 15 minutes and walk away.
- Letting the AI invent facts in replies. If you do not know a number, do not let the AI guess one.
- Building elaborate folder taxonomies. Search beats hierarchies; one “Archive” bucket plus key labels is enough.
- Running triage as your first task of the day. Email is reactive; protect the first 90 minutes for deep work.
FAQ
- Does AI see my email content?: Yes, when you paste it. For sensitive emails skip the AI entirely. Some sidebar tools (Gemini in Workspace, ChatGPT Business) offer data-handling guarantees; read them before piping all email through.
- How long should a draft reply be?: 2-4 sentences for most. Longer only if the topic explicitly needs it. AI defaults way too long; aggressive trimming is the discipline.
- Should I let the AI auto-reply?: No. Even with a perfect voice file, the human judgment about when to send, when to ignore, and what to actually say is the part you must keep.
- What about email triage for a shared team inbox?: Same workflow, but classification buckets become assignment buckets: “to-me,” “to-Sarah,” “to-CS-queue.” AI handles routing; humans handle replying.
- Will this work with Outlook?: Yes, with Copilot or a clipboard-based AI workflow. Gmail has tighter integrations but Outlook works with a 10-second copy-paste step per email.
- What if my voice file is too thin?: Pad it with annotated examples (“here I’m declining but warmly,” “here I’m pushing back firmly”). Three categories of 5 examples each beats 15 random emails.