Prerequisites

Before connecting GPT-4 to Make, you need three things:

  1. An OpenAI account with API access (separate from ChatGPT)
  2. An OpenAI API key (from platform.openai.com)
  3. Pre-loaded credits or a billing method on OpenAI (the API is pay-per-use)

Step 1: Create your OpenAI connection in Make

In any Make.com scenario, add an "OpenAI (ChatGPT, Whisper, DALL-E)" module. The first time, Make will ask for credentials:

  • Connection name: anything descriptive (e.g. "OpenAI – Production")
  • API key: paste your sk-... key
  • Organization ID (optional): only if you're on a multi-org OpenAI account

Step 2: Your first GPT-4 module

Use "Create a Completion (GPT, Functions, ChatGPT)" — it's the workhorse module.

Key fields:

  • Model: choose gpt-4o for the best quality/cost ratio in 2026
  • Messages: the conversation history. At minimum: a system message + a user message
  • Max tokens: cap the response length (and your costs)
  • Temperature: 0 for deterministic outputs, 0.7 for creative ones

Step 3: Common patterns

Pattern 1 — Email classifier

Trigger: new email in Gmail. GPT prompt: "Classify this email as: lead / support / spam / other. Reply with one word." Action: route to different modules based on the response.

Pattern 2 — Structured data extraction

Use the "Response format" parameter set to json_object and ask GPT to return structured data. Then parse the JSON in the next module.

Example prompt: "Extract company name, contact name, and pain point from this email. Return JSON with keys: company, contact, pain_point."

Cost optimization

GPT-4o pricing in 2026 is reasonable, but it adds up. A few rules:

  • Use gpt-4o-mini for classification tasks — 90% cheaper, often equivalent quality for binary decisions
  • Cap max_tokens — never leave it unlimited
  • Cache identical prompts — use a Make Data Store for repeated queries