AI Workflow Automation: Using ChatGPT, Claude, and Gemini in n8n & Make (2026)
AI isn’t just a chatbot. It’s a building block you can wire into automated workflows — and 2026 is the year this becomes the default way to work.
Here’s what’s now possible with zero custom code:
- Summarize every new support ticket and route it by urgency
- Read incoming emails and draft replies in your voice
- Generate SEO metadata for 1,000 blog posts while you sleep
- Extract structured data from messy PDFs and feed it into Airtable
n8n and Make both have first-class AI nodes. But they work very differently, and which one you choose determines what’s easy vs. what’s painful.
The AI Automation Landscape in 2026
Three things changed in the last 12 months:
- LLM APIs became commodities. OpenAI, Anthropic, and Google all have fast, cheap APIs with near-identical capabilities for most tasks.
- Both n8n and Make shipped native AI nodes. You no longer need to call REST APIs by hand — drag in an “OpenAI” or “Anthropic” node and configure it like any other module.
- Structured output became reliable. JSON mode and tool calling mean LLMs return predictable, machine-readable data — not just free text.
This means AI automations went from “cool demo” to “production-grade” in a year.
n8n AI Nodes: The Developer’s Playground
n8n has the richest AI integration in the no-code space. Its AI node family includes:
| Node | What It Does |
|---|---|
| OpenAI | Chat, completion, image generation, embeddings, speech-to-text |
| Anthropic Claude | Chat with Claude 4.5 Sonnet, Opus, Haiku |
| Google Gemini | Chat with Gemini 2.5 models |
| AI Agent | Chains multiple LLM calls with tools (RAG, web search, code execution) |
| Pinecone / Qdrant / Weaviate | Vector databases for RAG (retrieval-augmented generation) |
| Text Classifier | Zero-shot classification using any LLM backend |
Example: n8n AI Agent for Support Ticket Triage
Here’s a workflow that reads a support email, classifies it, and routes accordingly:
Gmail Trigger (new email)
↓
AI Agent (with tools)
├─ Tool: Vector Store (search knowledge base for similar tickets)
├─ Tool: OpenAI (classify as bug / feature / billing / general)
└─ Tool: Code node (extract structured JSON)
↓
Switch (route by category)
├─ Bug → Linear/Jira → Slack #bugs
├─ Billing → Stripe lookup → Slack #billing
└─ General → Reply with AI-drafted responseThe AI Agent node is what makes n8n special. It can:
- Use tools — call APIs, search databases, run code
- Reason multi-step — “first search the KB, then check the user’s plan, then reply”
- Handle errors — if a tool call fails, the agent can retry or ask for clarification
Example: Structured Extraction with Claude in n8n
Want to extract key fields from 500 PDF contracts? The Anthropic node with structured output:
// In the Claude node's "System Prompt"
You are a contract analyst. Extract the following fields as JSON:
- parties (array of company names)
- effective_date (ISO date)
- total_value (number in USD)
- termination_notice_days (number)
Return ONLY valid JSON. No explanation.Feed in the PDF (n8n can extract text from PDFs natively), and out comes structured data ready for your database.
Make AI Modules: Simpler, But Powerful
Make’s approach is more streamlined. Its AI modules include:
| Module | Provider |
|---|---|
| OpenAI (ChatGPT) | Chat completion, vision, TTS |
| Anthropic (Claude) | Messages API |
| Google AI (Gemini) | Gemini 2.5 |
| Create a completion | Universal — any OpenAI-compatible endpoint |
Make doesn’t have an “AI Agent” node like n8n, but you can build the same patterns by chaining modules.
Example: AI Blog Post Summarizer in Make
This workflow reads your blog’s RSS feed and auto-generates social posts:
RSS Watch (new blog post)
↓
OpenAI — Create a Completion
Model: gpt-4.1-mini
System: "Write 3 tweet-length summaries of this blog post.
Return as JSON array: [{text, hashtags, tone}]"
User: {{1.content}}
↓
JSON Parse (Make built-in)
↓
Buffer — Create Post (for each tweet)The key difference from n8n: you build the “agent loop” yourself by connecting modules, rather than having a single AI Agent node orchestrate everything. This is more work upfront but gives you finer control.
Head-to-Head: n8n AI vs Make AI
| Capability | n8n | Make |
|---|---|---|
| AI Agent with tools | ✅ Native node | ⚠️ Manual chaining |
| Vector DB / RAG | ✅ Pinecone, Qdrant, Weaviate | ⚠️ HTTP call only |
| Anthropic Claude | ✅ Native node | ✅ Native module |
| Structured output | ✅ JSON mode + Code node | ✅ JSON Parse module |
| Cost visibility | ⚠️ Tokens visible in execution | ✅ Clear per-operation cost |
| Multi-model chains | ✅ Easy — chain AI nodes | ✅ Easy — chain AI modules |
| Self-hosted AI | ✅ Ollama, LM Studio | ❌ Cloud only |
Verdict: For AI-heavy workflows, n8n wins — especially if you need RAG, tool-calling agents, or self-hosted models. Make wins for simpler AI tasks (summarize, classify, translate) where you want to stay in a visual environment.
3 Real-World AI Automations You Can Build Today
1. Email Inbox Autopilot
What: Read every email → draft a reply → save as draft (never auto-send without human review!)
Stack: Gmail trigger → Anthropic Claude (prompt: “Draft a professional reply to this email in the sender’s language”) → Gmail “Create Draft”
Time saved: 2-4 hours/day for busy inboxes
2. SEO Content Factory
What: Keywords from Airtable → Claude writes title + meta description + outline → Google Docs
Stack: Airtable Watch → Claude (structured output: {title, description, h2s[], slug}) → Google Docs Create
Time saved: 30 minutes per article on metadata alone
3. Multi-Language Customer Support
What: Detect language → translate to English → search knowledge base → translate reply back → send
Stack: Webhook → Claude (detect language + classify) → Vector Store search → Claude (compose reply in original language) → Reply
Cost: ~$0.01 per ticket (Claude Haiku) vs ~$2-5 per ticket (human)
The $0.01 Automation That Changed Our Mind
Here’s a story. We had a client form that asked “What do you need help with?” — a free-text field. 500+ submissions. Manually reading and tagging them took 3 hours a week.
We added ONE AI step:
Typeform → Claude Haiku ($0.01/call):
"Classify this inquiry: bug_report / feature_request / billing / other.
Also extract: urgency (1-5), product_mentioned, suggested_next_step.
Return JSON."
→ Google Sheets (sorted by urgency)Three hours a week became zero. For about $5/month in API costs. That’s the power of AI in automation — it handles the fuzzy parts that rules-based logic can’t.
Getting Started: Which AI Provider Should You Use?
For most automation use cases, the cheapest model that’s smart enough is the right choice:
| Task | Best Model | Cost per 1K calls |
|---|---|---|
| Classification / tagging | Claude Haiku / GPT-4.1 Nano | ~$0.50 |
| Summarization | Claude Haiku / GPT-4.1 Mini | ~$1.00 |
| Email drafting | Claude Sonnet / GPT-4.1 | ~$3.00 |
| Complex reasoning / RAG | Claude Opus / GPT-4.1 | ~$15.00 |
Our default: Claude Haiku for everything by default, Sonnet when the task needs real intelligence. This keeps costs predictable while handling 90% of use cases.
The Catch No One Talks About
AI in automations is incredible — until it silently fails. LLMs hallucinate, return wrong JSON formats, misunderstand edge cases, and occasionally produce output that’s subtly wrong but not obviously so.
Rules for production AI automations:
- Always validate structured output. Parse the JSON and check required fields exist before passing to the next step.
- Set a fallback node. If the AI call fails or returns garbage, route to a human-review step. Never silently drop data.
- Log AI outputs for auditing. Store what the model returned so you can debug issues later.
- Version your prompts. A prompt change can silently change behavior. Track prompts in git alongside your workflow definitions.
- Monitor costs. One runaway loop calling GPT-4.1 can burn $50 before you notice. Set usage alerts.
Build Your First AI Automation
The fastest path: sign up for n8n Cloud (free trial) or Make.com, grab an API key from OpenAI or Anthropic, and build the email classifier we described above.
It takes 30 minutes, costs less than $0.10 in API calls to test, and you’ll understand why AI + automation is the most important skill in no-code right now.
Disclosure: Some links are affiliate links. We may earn a commission if you sign up, at no extra cost to you. See our About page for details.