Connect OpenAI to WhatsApp

OpenAI's chat completions endpoint is the most widely used AI API in the world, and it connects to WA.Expert through the same External API Request step as every other provider. The main differences from Gemini: auth goes in an Authorization: Bearer header (not a custom header), and the messages array lets you set a system persona alongside the customer's question.

If you have not read the foundation guide on the External API Request step, read that first. This guide assumes you know the seven fields and focuses on what OpenAI specifically needs.

Step 1: Get your OpenAI API key

Step 2: Fill in the External API Request step

Here is the complete configuration. Each field is explained below.

Method

POST. You are sending a conversation to the model.

Request URL

https://api.openai.com/v1/chat/completions

This URL does not change with the model; you specify the model in the request body. That means switching from gpt-4o-mini to gpt-4o only requires changing one word in the body, not the URL.

Auth and headers

OpenAI uses standard HTTP Bearer auth. The word Bearer is part of the value, not the header name.

Body

The messages array is the key difference from Gemini. It lets you stack a system instruction before the customer's message, which shapes how the model behaves.

{
  "model": "gpt-4o-mini",
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful support assistant for {{business_name}}. Answer briefly and politely."
    },
    {
      "role": "user",
      "content": "{{customer_message}}"
    }
  ],
  "max_tokens": 300
}

Replace {{business_name}} with a fixed value or map it from your flow. Replace {{customer_message}} with the value from the conversation. max_tokens limits the reply length, which also limits cost.

Step 3: Map the reply into WhatsApp

The model's answer sits inside the choices array. OpenAI always returns at least one choice when the call succeeds, and the text you want is inside the first one.

{
  "choices": [
    {
      "message": {
        "role": "assistant",
        "content": "...the model's reply is here..."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": { "total_tokens": 87 }
}

Map choices[0].message.content into your next WhatsApp message step. Set Response Type to Default Response and the field will be available to reference.

Worked example: a returns and refund assistant

A customer messages: 'I got the wrong size, how do I return it?' Your bot does not have a scripted answer for this phrasing. The flow routes it to the OpenAI step with a system message that explains your policy.

{
  "model": "gpt-4o-mini",
  "messages": [
    {
      "role": "system",
      "content": "You are the support assistant for an online clothing store. Our return policy: items can be returned within 30 days with the original receipt for a full refund. No returns on sale items. Answer customer questions about this policy briefly."
    },
    {
      "role": "user",
      "content": "{{customer_message}}"
    }
  ],
  "max_tokens": 200
}

The model reads the policy from the system message and replies with a coherent, accurate answer in the customer's language. You never scripted that specific question; the AI covered it from the policy context you gave it.

Troubleshooting

The response from OpenAI almost always contains an error field with a clear message. Read it before troubleshooting further.

Common questions