Tag and prioritise incoming support messages
- Who it's for
- Businesses with a shared support inbox or helpdesk
- Outcome
- Every new message labelled by topic and urgency, so the right ones get answered first.
- Why Handoff 3
- It applies labels. Labels are easy to change, but a wrong 'low priority' can delay an angry customer.
Steps
- Write down your labels (for example: delivery, refund, product question, complaint) and what 'urgent' means.
- Ask the agent to suggest labels for 10 recent messages without applying them.
- Correct its suggestions and tighten the rules.
- Let it apply labels only (no replies, no closing, no deleting).
- Spot-check a handful each week.
Copy-paste prompt
Works with any agent. Replace the [square brackets] with your details.
Here are my support labels and rules: [labels + what each means]. Urgent means: [e.g. complaint, payment problem, safety issue].
Step 1: For the 10 most recent messages, suggest a label and urgency for each, with a one-line reason. Do not apply anything.
Step 2 (only after I approve the rules): apply labels and urgency to new messages. Never reply, close, archive or delete a message. If unsure, mark it urgent.Notes for each agent
Only features we found in each provider's official documentation on 14 Oct 2026 are stated as product facts, with the source linked. Product features change; check before relying on them.
Claude Suggested
Claude (Cowork) can open sites, click and fill forms. Use Manual mode: Anthropic advises staying close or using 'Manually approve' for work involving money, messages sent as you, or important files.
From official docs Official source
ChatGPT Suggested
ChatGPT agent mode has been retired; OpenAI points to ChatGPT Work. Its cloud browser is designed to ask for confirmation before hard-to-reverse actions or financial, legal or account commitments, and pauses for you to sign in.
From official docs Official source 1 Official source 2
Gemini
We have not verified a Gemini feature that completes this kind of website action for you. Use Gemini to prepare the details and do the final step yourself.
General guidance, not a verified product claim
Microsoft Copilot
We have not verified a Copilot feature that completes this kind of action for a small business outside Microsoft 365 apps. Use Copilot to prepare it and do the final step yourself.
General guidance, not a verified product claim
Open agents & automations Suggested
OpenClaw's docs say host commands run without prompts by default on gateway/node hosts; switch to the 'cautious' preset (allowlist, ask on anything else) first. In n8n, put the action tool behind a human review step.
From official docs Official source 1 Official source 2
Common failure, and how to catch it
- What goes wrong
- Sarcastic or polite complaints labelled as low priority.
- How to catch it
- 'If unsure, mark urgent' plus a weekly spot-check of low-priority items.
Time and cost
- Time to set up
- 30 minutes
- Cost
- Uses the AI plan you already pay for. Agent features (browsing, connectors, scheduling) may need a paid tier; check your plan before relying on them.
Sources
- https://support.claude.com/en/articles/13345190-get-started-with-claude-cowork
- https://help.openai.com/en/articles/20001280-using-cloud-browser-in-chatgpt
- https://help.openai.com/en/articles/11752874-chatgpt-agent
- https://docs.openclaw.ai/tools/exec-approvals
- https://docs.n8n.io/build/integrate-ai/ai-examples/human-in-the-loop-for-tools
Reviewed 14 Oct 2026 against the sources above. This playbook has not been tested on every agent listed; it is written to be agent-agnostic.