You prompt Claude. You get a great response. You copy it. You paste it somewhere. You move on.
That's the workflow most people are running. And it works, until you're doing it fifty times a week.
The gap between "AI gave me good output" and "AI did something for me" is automation. Specifically, it's knowing how to wire the output into a system that acts on it. That's where n8n comes in.
I'm not going to sell you on n8n over Make.com or Zapier. Use whatever you want. But n8n is what we run internally at Vaylo, and the reason is control. You can host it yourself, build complex branching logic, and call APIs without hitting a wall every time you need something slightly custom.
The Problem With Stopping at the Response
When AI is just a chat interface, you're still the connector. You read the output, you decide what to do with it, you go do it manually. That's fine for one-off work. It's a bottleneck when the task repeats.
Think about the things you prompt AI for regularly:
- Drafting emails or follow-ups
- Summarizing documents or transcripts
- Generating first drafts of content
- Categorizing or tagging incoming data
- Writing code snippets for a specific pattern
Every one of those has a destination. The email goes somewhere. The summary goes into a doc or a CRM. The content goes into a queue. The categorized data updates a sheet or triggers a next step.
n8n lets you remove yourself from the middle of that loop.
How the Basic Pattern Works
The simplest n8n + AI workflow looks like this:
- A trigger fires (form submission, webhook, schedule, email arrives)
- That data gets passed into an AI node (Claude via API, OpenAI, whatever)
- The response comes back as structured output
- n8n routes that output to the right destination
That's it. Four steps. The power is in what you put at step one and what you connect at step four.
Here's a real example. We have a workflow that fires every time a new lead fills out a form on a client site. The form data goes into an HTTP request to the Claude API. The prompt asks Claude to write a personalized intro email based on the lead's industry, what they said in the form, and a few standard offer points. The output gets routed to Gmail and sends automatically. No human touches it.
The lead gets a real, relevant email in under two minutes. The client doesn't have to write anything.
Getting the Output Into a Shape n8n Can Work With
This is where most people hit friction. AI returns a blob of text. n8n wants structured data it can route cleanly.
The fix is in the prompt. You have to tell the model exactly what format to return, and then parse it in n8n before it goes anywhere.
For simple stuff, ask for JSON back. Something like:
Return your response as a JSON object with these keys: subject, body, tone. No extra text outside the JSON.
Then in n8n, you use a JSON Parse node or just reference the fields directly in expressions. {{ $json.body }} goes into the email body. {{ $json.subject }} goes into the subject line. Clean.
For longer content like summaries or drafts, you might keep it as a plain string and just pipe it into a Google Doc or a Notion page via their APIs. n8n has native nodes for both. You don't need to structure it, you just need to know where it lands.
Branching Based on What AI Returns
This is where it gets interesting. You don't have to send every AI output to the same place. You can route based on what the AI actually said.
Let's say you're using AI to categorize support tickets. The model returns a category: billing, technical, general. n8n's Switch node reads that value and sends each ticket to the right Slack channel, assigns it to the right person, or fires a different email template depending on the type.
That's conditional logic running on AI output. It sounds advanced. It's four nodes.
The pattern: AI node outputs a value, Switch node reads it, each branch goes somewhere different. You build it once. It runs forever.
Chaining Models for Better Results
One thing solo builders don't use enough: running multiple AI calls in sequence.
First call generates a draft. Second call reviews it and returns a score or flags any issues. If the score is above a threshold, it ships. If not, it goes to a Notion page for manual review.
You're not just automating the output. You're automating quality control on the output. That's the difference between a workflow that produces garbage at scale and one you can actually trust.
We use this pattern for client-facing deliverables. The first model writes. The second model checks against a rubric. Only the stuff that passes goes out automatically. Everything else gets flagged.
What to Automate First
If you're new to this, don't start with something complex. Start with one task you do more than three times a week that follows a consistent pattern.
Good first candidates:
- Summarizing meeting transcripts into action items, then posting them to Notion or Slack
- Taking form submissions and generating a custom proposal draft
- Pulling new RSS items and generating a weekly digest email
- Categorizing inbound emails and routing them to the right folder or team
Pick one. Build it ugly. Make it work. Then clean it up.
The goal isn't a perfect workflow. The goal is removing yourself from one more repetitive loop so you can focus on the stuff that actually needs you.
The prompts and node structure above are what separate workflows that actually ship from ones that break after two runs. If you want the complete internal build, it's in Inner Circle.
Keep building
- Turn Customer Interviews Into Jobs-To-Be-Done from the prompt library
- AI Output Critic from the prompt library
- How I Stopped Managing My Business and Started Talking to It
- The full Vaylo prompt library, new drops weekly
