Integrating ChatGPT and Claude in Make

Section 10.6.3 : Integrating ChatGPT and Claude in Make

🎯 Learning objective

Master AI modules in Make — OpenAI, Anthropic (Claude), and custom API calls — to create advanced intelligent automations. You'll know how to choose the right model for each task, configure key parameters, and build reusable automation patterns.

The OpenAI module in Make: complete guide

Make has a native OpenAI module with several actions. Here's a breakdown of each:

Most commonly used: "Create a Chat Completion" — this is what you'll use in 90% of cases.

Mastering OpenAI's advanced parameters

Each parameter directly influences the quality, consistency, and cost of your automations:

Temperature (0 to 2) — controls "creativity":

• →0: deterministic and reproducible. The same input ALWAYS produces the same output. → Categorization, data extraction, parsing, classification
• →0.3: slightly varied but consistent → Summaries, professional writing, rephrasing
• →0.7: moderate creativity → Marketing copy, structured brainstorming, storytelling
• →1.0-1.5: highly creative, surprising results → Creative writing, original idea generation
• →> 1.5: near-random, often incoherent → ❌ Avoid in production

Max Tokens — limits response length:

• →100 tokens ≈ 75 words → sufficient for JSON categorization
• →500 tokens ≈ 375 words → sufficient for a response paragraph
• →2000 tokens ≈ 1500 words → sufficient for a short article
• →Cost impact: you pay for tokens actually generated, not the max. But a max that's too high can produce unnecessarily long responses.

Response Format — forces the output format:

{ "type": "json_object" }

Enables "JSON mode": the model ALWAYS returns valid JSON. Essential for automation, because malformed JSON would crash the entire scenario. Without this option, the model may add text before/after the JSON ("Here's the result: ...json...").

Top P (0 to 1) — alternative to temperature:

• →1.0: considers the full vocabulary (default)
• →0.1: considers only the top 10% most probable tokens → more deterministic
• →Rule: adjust either temperature or top_p, not both at the same time.

Model selection guide for Make

Integrating Claude via HTTP API

Claude (Anthropic) doesn't have a native Make module as of March 2026. Use the HTTP > Make a Request module. Here's the step-by-step configuration:

HTTP module configuration:

1. →URL: https://api.anthropic.com/v1/messages
2. →Method: POST
3. →Headers (add them one by one):

4. →Body type: Raw → JSON
5. →Request content:

{
  "model": "claude-sonnet-4-20250514",
  "max_tokens": 1024,
  "system": "You are an analysis assistant. Always return valid JSON.",
  "messages": [
    {
      "role": "user",
      "content": "Analyze this customer feedback and return {sentiment, themes[], satisfaction_score (1-10), recommended_action}:\n\n{{1.feedback_text}}"
    }
  ]
}

6. →Parse response: Yes
7. →The response will be in the data.content[0].text field of the HTTP module (referenced by its number N in Make)

When to use Claude rather than OpenAI in Make?

• →Long document analysis: Claude has a 200K token context window (vs 128K for GPT-5) — ideal for analyzing contracts, reports, or entire documents
• →Complex instructions: Claude excels at following long and nuanced instructions
• →Tasks requiring caution: Claude is naturally more conservative (useful for content moderation, compliance)
• →Vendor diversity considerations: don't depend on a single provider (resilience)

Integrating other AI APIs via HTTP

Make's HTTP module lets you call ANY AI API. Here are 3 useful examples:

Mistral AI (European model, GDPR-friendly):

POST https://api.mistral.ai/v1/chat/completions
Headers: Authorization: Bearer {API_KEY}
Body: {
  "model": "mistral-large-latest",
  "messages": [{"role": "user", "content": "..."}]
}

Perplexity (enriched web search):

POST https://api.perplexity.ai/chat/completions
Headers: Authorization: Bearer {API_KEY}
Body: {
  "model": "sonar",
  "messages": [{"role": "user", "content": "What are the latest news on [topic]?"}]
}

Replicate (open-source models: Llama 4, Stable Diffusion, etc.):

POST https://api.replicate.com/v1/predictions
Headers: Authorization: Token {API_KEY}
Body: {
  "version": "{model_version_id}",
  "input": {"prompt": "..."}
}

Advanced AI automation patterns

Here are 4 reusable patterns you can adapt to your needs:

Pattern 1: Data enrichment (intelligent ETL)

Google Sheets (new row) 
→ OpenAI (enrich: summary + category + sentiment + tags)
→ Google Sheets (update the row with AI data)

Concrete use case: a prospect list with just the company name → AI enriches with industry sector, estimated size, interest score, and a personalized sales pitch suggestion.

System prompt for enrichment:

You receive a company name and its website.
Return a JSON with:
- sector: the main industry sector
- size: "Micro" | "Small" | "Medium" | "Large enterprise"  
- interest_score: 1-10 (based on relevance for our [your domain] services)
- suggested_approach: 2 sentences for the first sales email

Pattern 2: Automated content pipeline

RSS Feed (new competitor article) or Schedule (daily 8am)
→ HTTP (Perplexity API: "[your industry] news from the last 24h")
→ OpenAI (summarize + identify the 5 most interesting angles)
→ OpenAI (write a LinkedIn post from the best angle)
→ OpenAI (write a Twitter thread of 5 tweets)
→ Google Docs (create a document with the drafts)
→ Slack (notify the content team with links)

Pattern 3: Intelligent multi-level customer support

Incoming email (support@) or Zendesk (new ticket)
→ OpenAI (categorize + urgency + suggested resolution + FAQ match)
→ Router:
  Route 1 (simple FAQ + low urgency): 
    → Automatic response with FAQ link + solution
    → Status "Resolved automatically"
  Route 2 (high urgency OR complaint):
    → Slack #support-urgencies → immediate notification
    → Empathetic email acknowledging receipt ("an agent will contact you within 1h")
    → Zendesk → priority assignment
  Route 3 (complex technical):
    → Zendesk → assign to technical team
    → Email "our technical team is reviewing your request"
→ All routes → Google Sheets (logging)
→ Weekly: Aggregator → Report (volume, categories, resolution time, satisfaction)

Pattern 4: Automated competitive intelligence

Schedule (daily 8am)
→ HTTP (Google News API / Perplexity: "[competitor 1] OR [competitor 2] OR [your industry]")
→ Iterator (for each article/result)
→ OpenAI (summary + relevance 1-10 + impact on our business + suggested action)
→ Filter (relevance > 7)
→ Aggregator (compile into structured newsletter)
→ OpenAI (write an executive summary of the week's trends)
→ Gmail (send digest to leadership team)
→ Notion (archive for reference)

Optimizing API costs in Make

Batch example: instead of 10 separate calls to categorize 10 emails, send all 10 in a single call:

Categorize the following 10 emails. Return a JSON array:
1. {{email_1}}
2. {{email_2}}
...
10. {{email_10}}