The 20-Minute Inbox

What this is for

You've got 40 unread emails that matter and 200 that don't, a reply you've been avoiding for a week, and someone who's ignored you twice. This is the routine for handling the ones that count without writing everything from scratch or sounding like a form letter.

An honest note on the clock: triaging a 40-email backlog end to end takes longer than 20 minutes the first time, because pasting subjects and intents in has a cost. The 20 minutes is real for the work that matters: triage the pile once, then draft, decline, and chase the 10-12 emails that actually needed you. The 200 that don't count were never 20-minute work; they were archive work.

Before you start

• Do Step 1 (the voice profile) once; every other prompt here reuses it.
• Have your inbox open where you can copy subject lines and short excerpts. You don't need full email chains, just enough for the AI to sort them.
• One confidentiality check before anything gets pasted: client emails and anything sensitive follow the same rules as every other work product. Strip names to initials or roles and cut identifying details, exactly as your AI policy would require on a team. If your employer provides Copilot or Gemini in your mail client, that's the approved place for full email content; see the native-tools section below.

Step 1: Set up your voice profile, once

Do this once, then reuse it everywhere instead of re-explaining your tone each time.

Building the profile, either model, one-time setup:

ROLE: You are helping me write a voice profile that describes how I
write email, so it can be reused as a standing instruction.
CONTEXT: Below are 3-5 real emails I've sent that sound like me.
MY EMAILS: [PASTE 3-5 REAL EMAILS, VARIED IN TOPIC]
CONSTRAINTS: Describe sentence length, formality level, how I open and
close emails, and any phrases I overuse or specifically avoid. Do not
describe generic "professional" traits. Describe what's actually
distinct about these samples.
OUTPUT FORMAT: A profile of 5-8 bullet points I can paste into a
custom instructions field.

Where to save it so it's always on:

• ChatGPT: Settings, then Personalization, then Custom Instructions, and paste the profile into the field asking what traits ChatGPT should have, since this one is about voice. If you'd rather scope it to email work only, create a Project for your inbox and put the profile in that project's instructions.
• Claude: click your initials in the lower left, then Settings, then General, and paste the profile into "Instructions for Claude" for an always-on version. Claude's reusable style feature is being folded into its Skills system as this is written, so if you want the profile on a toggle rather than always on, look under Skills rather than for a style picker.

The four prompts

Prompt 1: Triage a backlog into act / delegate / archive / decline

Use either model.

ROLE: You are an executive assistant triaging my inbox backlog.
CONTEXT: Below is a list of subject lines and one-line summaries from
my unread emails.
EMAILS:
[PASTE YOUR LIST, e.g.
1. Re: Q3 budget approval. Finance needs my sign-off by Friday
2. Newsletter: 5 AI tools you need
3. Can you speak at our conference in October (from someone I don't
   know)
4. S: draft contract review, needs my comments
5. Reminder: expense report overdue]
CONSTRAINTS: Sort each into exactly one bucket: Act (only I can do
this), Delegate (someone else should own it), Archive (no action
needed), or Decline (needs a no). If an email is ambiguous, put it in
Act and say why, rather than guessing wrong on a decline.
OUTPUT FORMAT: Four headed lists (Act, Delegate, Archive, Decline)
with the email number and a one-line reason under each.

Prompt 2: Batch-draft replies in your own voice

Use either model, with your voice profile from Step 1.

ROLE: You are drafting email replies in my voice, using the voice
profile below.
CONTEXT: Voice profile: [PASTE YOUR VOICE PROFILE FROM STEP 1]. Here
are the emails needing replies, with what I want each reply to say.
EMAILS AND INTENT:
[PASTE YOUR LIST, e.g.
1. S's contract review: tell her I'll have comments back by
   Wednesday, flag that section 4 needs legal eyes.
2. Budget approval: approve as-is, no changes.
3. Vendor asking for a call: decline, we're not evaluating new
   vendors this quarter.]
CONSTRAINTS: Match the voice profile's tone and length, not a generic
"professional email" tone. Keep each reply to what the intent line
actually says. Do not add pleasantries or hedging the voice profile
doesn't use.
OUTPUT FORMAT: One drafted reply per email, ready to paste, no
placeholder brackets left in.

Prompt 3: Decline politely without burning the relationship

Use either model.

ROLE: You are helping me decline a request in a way that's warm but
clearly a no, using my voice profile.
CONTEXT: Voice profile: [PASTE VOICE PROFILE]. The request: [THE ASK,
e.g. a former colleague asked me to review their startup's pitch deck
and give detailed feedback; I don't have time this month]. My
relationship with them: [THE RELATIONSHIP, e.g. friendly, worked
together for 3 years, want to stay on good terms].
CONSTRAINTS: Say no in the first two sentences. Don't bury it. Give
one honest, brief reason, not an over-explained excuse. Offer one
small alternative only if I'd genuinely follow through on it,
otherwise leave it out.
OUTPUT FORMAT: A single reply, 3-5 sentences, no filler apology
sentences ("I'm so sorry, I feel terrible").

Prompt 4: Chase a non-responder for the third time without sounding annoyed

Use either model.

ROLE: You are helping me write a third follow-up to someone who
hasn't responded to my last two emails.
CONTEXT: Voice profile: [PASTE VOICE PROFILE]. Original ask: [THE
ASK, e.g. needed their sign-off on the vendor contract]. First
follow-up sent: [DATE]. Second follow-up sent: [DATE]. I need this
resolved by [DEADLINE] because [REASON, e.g. the vendor's rate goes
up after this week].
CONSTRAINTS: No passive-aggressive language ("just following up
again!", "per my last email"). State the deadline and consequence
plainly. Give them an easy out: a one-click way to respond (yes/no,
or "reply with just a thumbs up").
OUTPUT FORMAT: One short email, under 80 words, with a clear,
low-effort action requested.

The AI already in your inbox, and when to use it instead

If you're on Google Workspace, Gemini is in Gmail's side panel as "Ask Gemini", and on a personal Google account the AI is inline instead, in the compose box and on top of threads. On Microsoft 365 with a Copilot license, it's the Copilot button in Outlook. All of them summarize threads, draft replies, and search your mailbox without any copy-paste, and if your employer licenses them, they're covered by the company's data agreement, which makes them the right place for sensitive content.

The honest trade, and it now depends on which mail client you're in. Gmail has closed most of the gap: "Help me write" and its suggested replies match the tone and style of your previously written emails, and you can refine a draft with your own typed instruction rather than a preset button. If you're in Gmail, try the built-in first. Outlook has not shipped that. Its "sound like you" setting is a short instruction you write yourself under Draft Instructions; it does not read your sent mail to learn your voice, so the drafts still arrive in house style until you tell them otherwise. A workable split: built-in AI for summarizing long threads and searching ("what did we decide about the vendor?"), the voice-profile prompts for anything a human will read closely, and the built-ins as the only place full sensitive threads get processed. One thing the hand-built profile still wins on either way: you decide what goes into it, rather than letting a tool infer your voice from every email you have ever sent, including the ones that do not sound like you at your best.

Getting it right

• Build the voice profile first. Generic drafts read like everyone else's AI email, and recipients notice. Spend the ten minutes once; it pays back on every reply after.
• Decide the political ones yourself. Sensitive people-management or client emails deserve your own judgment, not a "decline" from the triage prompt. Use the triage output as a sort, not a decision.
• Read every batch-drafted reply before it goes. These are fast first drafts. Skim each one before sending; the AI doesn't know if a fact in the email chain changed since you copied it in.
• Keep the confidentiality check in the routine. The routine makes pasting a sensitive thread into a personal account feel normal; the check at the top is part of the routine, not a disclaimer.

The 2-minute version

Build your voice profile once and save it (ChatGPT's Custom Instructions, or Instructions for Claude). Triage your backlog into the four buckets. Batch-draft the Act pile in your voice. Send declines and chase-ups same day; don't let them age into a fifth email. The 12 Prompt Shortcuts make every one of these prompts sharper.

---

More where this came from. The resource library has 24 free guides like this one, written for people with real careers rather than developers.

Weekly, on Tuesdays. The GenXcelerate newsletter: practical AI intelligence for people who already know how to do the job. Subscribe free.

Experience Is the API™. GenXcelerate