Draft answers to common customer questions from your own FAQ
- Who it's for
- Shops and service businesses answering repeat questions
- Outcome
- Accurate draft answers built only from your own policies.
- Why Handoff 2
- Drafts for you to send. It does not reply to customers directly.
Steps
- Collect your FAQ, returns policy and delivery info in one document.
- Tell the agent to answer only from that document.
- Ask it to say 'not in the FAQ' rather than guess.
- Review drafts and send them yourself.
Copy-paste prompt
Works with any agent. Replace the [square brackets] with your details.
Answer the customer messages below using ONLY the information in my FAQ and policies: [paste or attach].
If the answer is not in the documents, write 'NOT IN FAQ: [what is needed]' instead of guessing. Do not offer refunds, discounts or exceptions. Keep each answer friendly and under 100 words. Show the drafts here; do not send anything.
Messages: [paste]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
Ask Claude to save drafts, not send them. In Cowork's Manual mode it pauses for approval before using connector tools set to 'Needs approval'. Avoid 'Skip' mode for anything sent in your name.
From official docs Official source
ChatGPT Suggested
Before an app action such as sending an email, ChatGPT may show an approval card. Keep your email app on 'Always ask' or 'Allow read actions' so sending always needs your OK.
From official docs Official source
Gemini Suggested
Gemini can draft Gmail messages in chat (Google Workspace connected app required). The draft has a Send button, so read it first, or use 'Edit in Gmail' on the web to finish it in Gmail.
From official docs Official source
Microsoft Copilot Suggested
In Outlook, 'Draft with Copilot' writes a draft you edit and send yourself, and 'Coaching by Copilot' can suggest tone and clarity fixes.
From official docs Official source
Open agents & automations Suggested
In n8n, connect the send tool through a human review step: the workflow pauses and shows the exact message for Approve or Deny.
From official docs Official source
Common failure, and how to catch it
- What goes wrong
- It fills gaps with plausible-sounding policy that you do not actually offer.
- How to catch it
- The 'NOT IN FAQ' rule, plus a quick read of any answer that mentions money.
Time and cost
- Time to set up
- 20 minutes
- Cost
- Works in a standard chat on most plans; no extra tools needed.
Sources
- https://support.claude.com/en/articles/13345190-get-started-with-claude-cowork
- https://help.openai.com/en/articles/20001495-managing-app-permissions-in-chatgpt
- https://support.google.com/gemini/answer/18741629
- https://support.microsoft.com/en-gb/outlook/frequently-asked-questions-about-copilot-in-outlook
- 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.
More customer support playbooks
- Tag and prioritise incoming support messages (Handoff 3)
- Draft replies to online reviews (Handoff 2)
- Check refund requests against your policy and prepare a decision (Handoff 4)