The inbox cleanup: 10 minutes a day with Claude

The triage prompt

This is the single prompt that runs the whole workflow. Paste at the top of each morning’s Claude chat, then paste your emails underneath.

I'm about to paste 8-12 emails from my business inbox. For each email:

1. ONE-LINE SUMMARY of what the sender is saying / asking
2. SENDER INTENT: enquiry / complaint / booking / supplier / spam /
   newsletter / colleague / personal / FYI-only
3. ACTION: needs reply / needs forward / needs internal action /
   archive / unsubscribe / flag-for-later
4. DRAFT REPLY in my voice (only if the action is "needs reply").
   Match this tone:
   - Friendly Australian, warm but not gushing
   - Plain English, no jargon, no exclamation marks
   - Concrete next step where relevant
   - End with a polite signoff
   - Keep replies under 80 words unless the email genuinely
     needs more
   - Don't make commitments I haven't told you about (no quoting
     prices, no promising delivery dates, no agreeing to discounts)

5. PRIORITY ORDER: at the end, list which emails I should reply to
   first, second, third based on urgency + customer relationship +
   business impact.

OUTPUT FORMAT for each email:

EMAIL [N]: [sender name], [one-line summary]
Intent: [classification]
Action: [classification]
Draft reply: [the draft text, or "n/a" if no reply needed]

---

Here are the emails:

[paste emails 1-12 with clear separators like "===EMAIL 1==="]

You’ll get the full output in 60-90 seconds. Spend 10 minutes reading + editing + copy-pasting back to your email client.

Why this beats “let AI auto-reply”

The temptation with AI + email is to wire it up so AI replies for you automatically. We’ve tested this with a handful of DotVA clients. It fails in three predictable ways:

1. Tone drift. AI gets the tone subtly wrong on the email that matters. The customer notices. The relationship cools.
2. Hallucination. AI commits to things you haven’t promised: a delivery date, a discount, a feature, a meeting time. Some are binding.
3. Filter bypass. AI replies to phishing / scam emails because it can’t always tell. You confirm to scammers that the inbox is monitored.

The 10-minute review fixes all three. You catch tone, you catch hallucination, you flag scams. The workflow stays human in the loop without being slow.

What this workflow saves you, with real numbers

Before AI inbox workflow, the typical Australian SMB email pattern:

After this workflow:

Time saved: 70-115 minutes per day. 5-8 hours per week. ~250 hours per year.

Over a year, that’s the equivalent of 6+ working weeks recovered.

The progression: from copy-paste to integration

The copy-paste workflow above is enough for most operators. After 4-8 weeks of habit, some operators upgrade to:

Upgrade 1: Add an email-specific Claude Project (Pro feature)

Create a Claude Project called “Inbox”. Voice file in the Instructions. The standard triage prompt as a default starter. Past examples of good replies you’ve written, uploaded as Knowledge. Every morning you open the Project and the chat is pre-loaded with the right context.

Saves ~30 seconds per morning by removing the prompt-pasting step.

Upgrade 2: Gmail / Outlook MCP integration (developer-relevant)

The model-context-protocol ecosystem has official-ish MCP servers for Gmail and Outlook. Wire one up and Claude can read your inbox directly without you copy-pasting. Workflow becomes: open Claude Code, ask “triage today’s inbox”, get the full output without the paste step.

Upgrades:

• Saves the 1 minute of copy-paste per day
• Adds complexity (MCP server setup, OAuth, security review)
• Requires the AI Security flagship Part 1.4 mitigations (MCP supply chain)

We don’t recommend this upgrade until the copy-paste workflow has been running smoothly for 8+ weeks and you’ve decided the additional 1-minute saving is worth the integration burden.

Upgrade 3: Hire a virtual assistant who runs the workflow for you

The 10-minute workflow is still 10 minutes. For operators whose 10 minutes is genuinely worth $X to do something higher-use, the next step is a VA who runs the workflow on your behalf and only escalates the 5% of emails that need your call.

This is what DotVA does for ~30 clients in the Australian small business segment. The VA + AI combination produces 50-150 emails/day handled, with the operator spending under 10 minutes total.

What this doesn’t solve

Be honest about limits.

• Emotional / relational emails. When something has gone wrong with a customer, when a long-standing supplier closes, when staff messages need a real human reply. Pull these out and write them yourself.
• Sales / proposal emails. Anything that requires you to actually think about positioning, pricing, scope. The first draft can help; the final email is you.
• Hard conversations. Firing a client, declining a partnership, complaint escalations. Write these yourself.
• Anything you’re going to be quoted on. AI drafts can have subtle factual problems. If the recipient is a journalist, lawyer, regulator, or major stakeholder, write it yourself.

For the other 70-80% of inbox, the workflow above is genuinely transformative.

What’s next

• Build a social media powerhouse with Claude Code, the 30-minute weekly content workflow.
• Your Sunday-night business briefing prompt, the 5-minute weekend reset.
• Claude for absolute beginners for the foundation if you skipped ahead.
• Book a free audit if you want help wiring this into a wider business operating system.