An email assistant that only writes drafts still made 4 quiet mistakes
The poster built an n8n workflow that sorts unread Gmail into leads, support, newsletters and other, labels each email, and has an AI model write a reply that is saved as a draft. Nothing is sent automatically. Every run showed as successful, but the drafts had problems: one offered a 15% discount nobody approved, all were signed "[Your Name]", the recipient field was empty, and one run made two drafts for the same customer. He fixed it with clear rules in the prompt, the right recipient setting, a "processed" label and marking emails as read. The workflow file is free on GitHub.
The lesson: A green run doesn't mean good output; read a few drafts and write firm rules (no discounts, no promises, real name) into the prompt.
Copy this prompt
Write a short reply to the email below as [YOUR NAME] from [BUSINESS]. Rules: never offer discounts, refunds or deadlines; never promise anything I haven't stated here: [FACTS YOU MAY USE]. Sign with my real name, never a placeholder. If you can't answer from these facts, write a one-line holding reply and say I'll follow up. Email: [EMAIL TEXT]
Our own instructions for the AI step that writes the draft, built around the poster's fixes. The sorting, labels and draft saving are plain n8n steps. We haven't run it ourselves. Text in [BRACKETS] is what you replace.
Where this comes from
Original post by Huzaifa_Aamish on n8n Community, Oct 22, 2026. We rewrote it in our own words. Check the original for the full context.
Read the originalGuides for this job
Turn a messy inbox into a 5-minute daily summary
Paste in today's emails and get back what needs a reply today, what can wait, what to ignore, and a one-line reply for each. It also flags emails that look like scams.