Inbox & communication

64 Emails in 10 Days, 15 Never Touched: An n8n Inbox Assistant That Only Drafts Replies, and Why Polling Missed the Rest

  • Level Intermediate
  • Setup1–3 hrs (estimated)
  • Evidence Some evidence
  • CostNot stated in source
  • No-codeNo
n8nSource-linked

What did they do with n8n?

A builder removed the test-sender filter on an n8n AI email assistant and let it run on a real business inbox for 10 days. The workflow is a Gmail Trigger that polls every minute, then a Gemini text classifier, a label per branch, a Gemini draft writer and Gmail's create-draft action. Nothing is sent automatically. Result: 64 emails arrived, 49 were sorted, and 15 were never touched. Every execution showed green, so nothing flagged the misses. He traced the 15 to five causes: the trigger only reads unread mail, so anything he opened on his phone first was skipped; polling is forward-only, so mail that arrived while it was down was gone; Max Emails per Poll was set to 5; the Gemini node had no retry, so one 503 lost an email; and his Google OAuth app was still in Testing, so the Gmail credential expired after 7 days and the trigger stopped. His fixes: Retry On Fail, 3 tries, 5 seconds apart, on the classifier and the draft writer; a separate daily workflow that searches in:inbox -label:ai-processed newer_than:1d and emails him anything the bot skipped, kept separate so it still runs when the assistant is down; and moving the OAuth app to In production. He published both workflows as free JSON.

How it went

n8n community Built with n8n post by @Huzaifa_Aamish (Nov 6, 2026): 10-day live inbox run; 64 arrived, 49 sorted, 15 never touched; five stated causes and three stated fixes; free workflow JSON on GitHub (buildthenbreak-hq/n8n-templates). Follow-up to an earlier post about drafts breaking in four ways. A YouTube video is linked; the details above are from the written post only

Worth knowing

  • n8n community Built with n8n post by @Huzaifa_Aamish (Nov 6, 2026): 10-day live inbox run; 64 arrived, 49 sorted, 15 never touched; five stated causes and three stated fixes; free workflow JSON on GitHub (buildthenbreak-hq/n8n-templates). Follow-up to an earlier post about drafts breaking in four ways. A YouTube video is linked; the details above are from the written post only.

Prompt

Cleaned version rewritten from the source — not a direct quote.

Watch my [Gmail inbox] and never send anything. For each new email, classify it as [categories], apply the matching label, and create a draft reply in my voice. If you are not sure, draft a short note to me instead. Do not mark an email handled unless you actually created a draft. Separately, once a day, search for inbox mail from the last day that still has no [ai-processed] label and email me that list, even if the main assistant is down. Retry failed AI steps [3] times, [5] seconds apart. Tell me if the inbox connection looks expired.

Recommended

Try it in Grok Bot

  1. Copy the prompt (button below).
  2. Open Grok Bot chat and paste it. Replace any [bracketed] placeholders.
  3. Connect the listed tools (Gmail, Email) as connectors when Grok Bot asks.
  4. If the task repeats, ask: "Make this a daily routine."

Alternatives: the same prompt works in n8n, Claude, ChatGPT Agent, OpenClaw.

Set it up

Tools: Gmail, Email · Stack: n8n

  1. Read the original source for full context.
  2. Connect the tools your agent needs.
  3. Paste the prompt and replace [bracketed] placeholders.
  4. Run once manually and check the output before scheduling it.

Generic suggested steps, not from the source.

Evidence

n8n community Built with n8n post by @Huzaifa_Aamish (Nov 6, 2026): 10-day live inbox run; 64 arrived, 49 sorted, 15 never touched; five stated causes and three stated fixes; free workflow JSON on GitHub (buildthenbreak-hq/n8n-templates). Follow-up to an earlier post about drafts breaking in four ways. A YouTube video is linked; the details above are from the written post only (as of 2026-11-06)

Source: Web · 2026-11-06 · Verified by us: not yet · Report broken or outdated

Related use cases

Inbox & communicationX · 2026-10

Ask your bot for a security review of an email that passed every check

An email from noreply@x.ai landed in a Tesla engineer's inbox. It passed SPF, DKIM and DMARC (the checks that confirm who sent an email) and showed the blue check, but something felt off, so he asked his Grok Bot to do a security review. It found a hidden phishing link, explained why every safety check had passed, and wrote a report to xAI with a proposed fix. He then had it add a security routine that checks his email for anything phishy going forward.

  • Level Beginner
  • Setup~30 min*
  • Evidence Strong evidence
Grok BotSource-linkedBeginner-friendly
Inbox & communicationWeb · 2026-10

12 of 20 Prompt Injections Worked on His n8n Email Bot, Then 0 of 60: Let Code Make the Decisions, Not the Model

He built an n8n workflow where a small local model reads customer emails and decides priority, refunds and replies, then attacked it with 20 injection emails, 3 runs each. 12 of 20 worked on the plain version (35 of 60 runs), and the usual 'don't follow instructions in the email' prompt made it slightly worse (13 of 20). Taking every decision away from the model fixed it: the model only fills in a form, and a code step strips hidden HTML, keeps secrets out of the prompt and decides what happens. That got 0 of 60. The cost: only 19 of 36 normal emails were answered automatically, down from 31.

  • Level Advanced
  • Setup1–3 hrs*
  • Evidence Strong evidence
n8nSource-linked
MarketingWeb · 2026-10

118 of 120 AI Social Posts Were Bad and 93 Forgot the Link: An n8n Test, and Why Code Checks Still Missed False Numbers

He built a blog-post-to-5-platforms social workflow in n8n two ways and ran both on 8 posts, 3 times each. The one-prompt version produced 118 bad posts out of 120, and 93 left out the blog link. The version with code-checked rules marked 77 posts ready and sent 43 to a human, and none of the 77 failed an independent checker. But when he read them, 10 of 21 in the first round got the blog post wrong, including saying churn rose from 6.1% to 4.3%. His conclusion: code can check format, not truth.

  • Level Intermediate
  • Setup1–3 hrs*
  • Evidence Strong evidence
n8nSource-linked