AI automation tools let you connect AI models to the rest of your software stack — without code. This guide covers how to build useful automations with Zapier AI, Make, and n8n.

Which Platform to Use

Zapier: Easiest to learn, largest app library (6,000+ integrations), best for simple linear automations. Limited flexibility for complex logic.

Make (formerly Integromat): More powerful routing and data manipulation, visual workflow builder, better value for complex automations. Steeper learning curve.

n8n: Open-source, self-hostable, most developer-friendly. Best for teams with technical users who want full control.

For AI-specific workflows: All three support OpenAI, Anthropic, and other AI APIs natively.

Foundation: Connecting AI to Zapier

Basic Setup: ChatGPT in Zapier

  1. Create a new Zap
  2. Choose your trigger (Gmail, Slack, form submission, etc.)
  3. Add action: ChatGPT by OpenAI
  4. Select: Send Prompt
  5. Configure the model (GPT-4o recommended)
  6. Write your prompt using dynamic fields from the trigger

Example: Auto-summarize incoming emails

Trigger: Gmail — New Email matching [newsletter OR report] Action: ChatGPT — Summarize this email in 3 bullet points Action: Gmail — Send summary to yourself

Practical Automation Recipes

1. AI-Powered Lead Qualification

What it does: When a form submission comes in, AI scores the lead and routes it.

Trigger: Typeform / HubSpot form submission
Step 1: ChatGPT prompt:
  "Analyze this lead:
  Company: {company}
  Role: {title}
  Message: {message}
  
  Score 1-10 for sales-readiness and explain why in one sentence.
  Format: SCORE: X | REASON: [explanation]"

Step 2: Filter — if score ≥ 7:
  → Create CRM contact (Salesforce/HubSpot)
  → Notify sales team in Slack
  
Step 3: Filter — if score < 7:
  → Add to email nurture sequence

2. Content Repurposing Pipeline

What it does: Turn a blog post into multiple social media formats automatically.

Trigger: RSS feed (new blog post published)
Step 1: Claude — Extract key points (paste post content)
Step 2: Claude — Write LinkedIn post (professional tone, 150 words)
Step 3: Claude — Write 3 tweets (under 280 chars each)
Step 4: Claude — Write Instagram caption + 10 hashtags
Step 5: Buffer — Schedule all posts across channels

3. Customer Support Triage

What it does: Classify and draft responses to support emails.

Trigger: New email in support inbox
Step 1: GPT-4o — Classify ticket:
  "Classify this support email into one category:
  BILLING | TECHNICAL | FEATURE_REQUEST | COMPLAINT | OTHER
  Also estimate urgency: LOW | MEDIUM | HIGH
  Reply with: CATEGORY: X | URGENCY: Y"

Step 2: Router based on category:
  BILLING → Draft billing response template
  TECHNICAL → Create Jira ticket
  HIGH urgency → Slack alert to on-call

Step 3: Draft response (for TECHNICAL):
  "Draft a helpful first response to this support ticket.
  Acknowledge the issue, ask one clarifying question.
  Keep it under 100 words, friendly tone."

4. Meeting Notes Processing

Trigger: Otter.ai / Fireflies.ai — New transcript ready
Step 1: Claude — Process transcript:
  "From this meeting transcript:
  1. Write a 3-sentence summary
  2. List action items in format: - [Owner]: [Task] by [Date if mentioned]
  3. List decisions made
  4. List open questions
  Transcript: {transcript}"
  
Step 2: Create Notion page with structured notes
Step 3: Email attendees the summary
Step 4: Create Asana tasks from action items

Make (Integromat) Workflows

Make is better for complex routing and data transformation.

Multi-Step Content Generation

HTTP Trigger (webhook) → receives product data
↓
OpenAI Module: Generate description
↓
OpenAI Module: Translate to 3 languages  
↓
Router:
  Route 1 (English) → Update Shopify product
  Route 2 (Spanish) → Update ES Shopify store
  Route 3 (French) → Update FR Shopify store
↓
Slack notification: "Products updated"

Data Enrichment Pipeline

Google Sheets: New row (company name, domain)
↓
HTTP Module: Hunter.io API → get email contacts
↓
OpenAI: Research company and write personalized outreach
↓
Filter: Only if email found
↓
Instantly/Apollo: Add to email sequence
↓
Sheets: Update status column

n8n Self-Hosted AI Workflows

For teams with technical resources, n8n offers the most flexibility.

Setting Up n8n with Claude

  1. Install n8n: npx n8n or via Docker
  2. Add HTTP Request node
  3. Configure Anthropic API:
    • URL: https://api.anthropic.com/v1/messages
    • Authentication: Header Auth (x-api-key: your-key)
    • Body: Claude messages format

n8n Code Node for Custom Logic

// n8n Code node — process AI response
const aiResponse = items[0].json.content[0].text;

// Parse structured output
const lines = aiResponse.split('\n');
const score = parseInt(lines.find(l => l.startsWith('SCORE:'))?.split(':')[1]?.trim());
const category = lines.find(l => l.startsWith('CATEGORY:'))?.split(':')[1]?.trim();

return [{
  json: {
    score,
    category,
    isHighPriority: score >= 8,
    originalResponse: aiResponse
  }
}];

Prompt Engineering for Automations

Automation prompts need to be more rigid than conversational prompts.

Always Specify Output Format

Bad prompt:
"Summarize this email"

Good prompt:
"Summarize this email in exactly 3 bullet points.
Format each bullet as: • [Action/Key point]
Total length: under 100 words.
Do not include greetings or sign-offs."

Build in Error Handling

"If the content is not in English, respond only with: SKIP
If the content is spam or irrelevant, respond only with: SKIP
Otherwise, [your actual instructions]"

JSON Output for Downstream Processing

"Analyze this customer message and respond ONLY with valid JSON:
{
  \"sentiment\": \"positive|negative|neutral\",
  \"topic\": \"billing|technical|general\",
  \"urgency\": 1-5,
  \"suggested_action\": \"reply|escalate|close\"
}"

Cost Management

AI automations can generate significant API costs at scale.

Cost-saving strategies:

  • Cache responses for identical inputs
  • Use cheaper models for simple classification (GPT-4o-mini, Claude Haiku)
  • Reserve powerful models for generation tasks
  • Add filters before AI steps to skip irrelevant triggers
  • Set token limits appropriate to the task

Approximate costs per 1000 automations:

  • GPT-4o-mini classification: ~$0.10-0.30
  • GPT-4o generation: ~$2-10
  • Claude 3.5 Sonnet generation: ~$3-15

Common Mistakes

No output validation: AI can format responses differently than expected. Always test edge cases and build fallback logic.

Triggering on everything: Add filters before AI steps — only process what you actually need to.

No error handling: API timeouts and rate limits will happen. Add retry logic and error notifications.

Not monitoring costs: Set billing alerts on OpenAI/Anthropic before your automations run at scale.