Business ops

Allow-List, Don't Block-List: An n8n Support Bot That Can't Promise a Refund Without a Person

  • Level Advanced
  • Setup1–3 hrs (estimated)
  • Evidence Anecdotal
  • CostNot stated in source
  • No-codeNo
n8nSource-linked

What did they do with n8n?

This builder made a 37-node AI support desk in n8n that reads tickets, looks up Shopify orders and drafts replies. A single Policy Guard step decides when it may send: only order status, shipping and product questions can auto-send. Angry or urgent customers, low confidence, placeholders, repeat contacts and any promise of a refund or discount go to a person with the draft and the reason attached. It ships in draft mode.

How it went

n8n Community 'Built with n8n' post by NasarHayat (Nov 11, 2026): 37-node support desk (Gemini classification, Shopify lookup, knowledge-base drafts, SLA checks); Policy Guard code shared; tested end to end with real tickets. No volume or time-saved numbers. Regex catches only the phrases you list; review held tickets weekly and add new risky phrases

Worth knowing

  • n8n Community 'Built with n8n' post by NasarHayat (Nov 11, 2026): 37-node support desk (Gemini classification, Shopify lookup, knowledge-base drafts, SLA checks); Policy Guard code shared; tested end to end with real tickets. No volume or time-saved numbers.
  • Regex catches only the phrases you list; review held tickets weekly and add new risky phrases.

Prompt

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

Add a policy check between my AI support drafts and the send step. Only allow auto-send for [order status, shipping, product questions]. Send to a person if the customer is angry or urgent, confidence is below [0.8], the draft has a placeholder, it's their 3rd contact this week, or it mentions refunds, discounts, compensation or guarantees. Start everything in draft mode, and tell the human why each ticket was held.

Recommended

Try it in Grok Bot

Grok Bot can draft support replies under the same rule: send only safe categories, hold anything that promises money.

  1. Copy the prompt (button below).
  2. Open Grok Bot chat and paste it. Replace any [bracketed] placeholders.
  3. Connect the listed tools (as needed) 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

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 NasarHayat (Nov 11, 2026): 37-node support desk (Gemini classification, Shopify lookup, knowledge-base drafts, SLA checks); Policy Guard code shared; tested end to end with real tickets. No volume or time-saved numbers. Regex catches only the phrases you list; review held tickets weekly and add new risky phrases (as of 2026-11-11)

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

Related use cases

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
Business opsReddit · 2026-10

About 2,100 Building Alarms Logged Since March: A Grok Bot Escalates Only the Clusters That Mean a Real Problem

An HVAC controls company's Alarms bot reads every building-automation alarm email and compares it with that site's history. It has logged about 2,100 alarms since March, with weekly to yearly summaries, and escalates only clusters that probably mean a real problem, like a boiler plus fan alarms at a church early on a Sunday. It also caught that a cell carrier had shut off email-to-text, which was silently breaking one site's alerts. A Monday recap goes to the service team.

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