Why this matters
Every firm has an inbox or a web form that someone has to read all day: what is this, how urgent is it, what do we say back? At a small bookkeeping firm with about 40 inquiries a month that is roughly 3.5 hours of reading and first drafts, and the one payroll emergency that arrives at 4pm still waits until tomorrow.
AI workflow automation fixes exactly this kind of job. A trigger catches each message, an AI step reads it, sorts it and drafts the reply, and a rule decides what happens next. You build all three in n8n in about 12 minutes, using n8n's free OpenAI credits, so you need no API key.
Practice
Do every step yourself in a practice copy of the real tool. It checks your work, and “Do it for me” finishes any step you get stuck on.
What you’ll build
An n8n workflow with three nodes: an n8n Form that catches each client inquiry, an OpenAI Message a model step (GPT-5-MINI) that returns category, urgent, summary and a draft reply, and an If rule that sends urgent inquiries down the true branch.
In our test, a made-up payroll emergency came back as Payroll, urgent true, with a reply promising a call today, and landed on the true branch.
What you need
- Nothing to start the practiceEvery screen is a real n8n capture. Nothing touches your account.
- An n8n accountn8n Cloud free trial. New accounts showed n8n free OpenAI API credits on the OpenAI node in October 2026.
- The promptDownload it below and change the categories to your own services.
Concepts
Trigger, AI step, rule
A trigger starts the workflow: here, a form submission. The AI step does the part that needs judgement: reading free text and writing a reply. The rule (an If node) decides what happens next, and it is yours, not the AI's.
This is not an AI agent. A workflow always runs the same path, which makes it easier to test and cheaper to run. Start here.
Ask the AI for a fixed shape
The prompt asks for JSON with four named keys: category, urgent, summary and reply. That is what lets the If node test "urgent": true. A paragraph of prose cannot be tested reliably.
Drag the form field into the prompt instead of typing the message: n8n writes {{ $json['What do you need help with?'] }}, so every new inquiry is read automatically.
Keep a person in the loop
The AI drafts; a person still sends. Read the first few weeks of drafts before you trust them, and never send card numbers or passwords through the AI step.
Do it for real
Now repeat it in your own account. Tick each step as you go.
Downloads
- The inquiry-sorter prompt and If rule (.txt)The exact prompt, the If condition and the test inquiry we used.
Quick check
3 questions to check you’ve got it.
1. What are the three parts of this AI workflow?
- A trigger, an AI step and a rule
- A chatbot, a database and a website
- Three AI agents
Show the answer
A trigger, an AI step and a rule. The form triggers it, the AI step reads and drafts, and the If rule decides what happens next.
2. Why does the prompt ask for JSON with named keys?
- It is cheaper
- So the next node can test a value such as "urgent": true
- n8n only accepts JSON
Show the answer
So the next node can test a value such as "urgent": true. A fixed shape lets a rule check the answer. Free text cannot be tested reliably.
3. Who decides what happens to an urgent inquiry?
- The AI
- Your If rule and the person who sends the reply
- n8n picks at random
Show the answer
Your If rule and the person who sends the reply. The AI only labels and drafts. The rule and the person stay in charge.