Connecting ChatGPT and Claude APIs

Now that your accounts are set up, let's understand how AI APIs work inside automation platforms. Knowing how to configure AI steps properly is the foundation for every automation you will build.

How AI APIs Work in Automations

When you add an AI step to your workflow, here is what happens behind the scenes:

1. Your automation sends a request to the AI provider (OpenAI or Anthropic)
2. The request includes a system prompt (instructions) and a user message (the data to process)
3. The AI generates a response based on your instructions
4. Your automation receives the response and passes it to the next step

You do not see any of this. Make and Zapier handle the API calls for you. But understanding this flow helps you configure your AI steps effectively.

The System Prompt: Your AI's Instructions

The system prompt is the most important configuration in any AI automation. It tells the AI:

• Who it is (role)
• What to do (task)
• How to respond (format and tone)
• What to avoid (constraints)

Example: Customer Email Classifier

You are a customer support classifier.

Given an incoming customer email, classify it into exactly one category:
- BILLING (payment issues, refunds, invoices)
- TECHNICAL (bugs, errors, how-to questions)
- GENERAL (feedback, feature requests, other)

Respond with ONLY the category name in uppercase. No explanation.

This prompt is specific, constrained, and produces predictable output, which is exactly what you need for automation.

System Prompt Best Practices for Automation

Practice	Why It Matters
Be explicit about output format	The next step in your automation needs predictable input
Keep responses short	Shorter responses cost less and process faster
Include examples	Reduces inconsistent or unexpected outputs
Set constraints	"Respond in 50 words or fewer" prevents rambling
Define edge cases	"If unsure, respond with UNKNOWN" prevents errors

Configuring AI in Zapier

The OpenAI "Conversation" Action

This is the most versatile AI action in Zapier. Here is how to configure it:

Model: Choose the AI model.

• gpt-4o-mini - Fast, cheap, good for simple tasks (recommended for most automations)
• gpt-4o - More capable, better for complex reasoning

System Prompt: Your instructions (the example above goes here)

User Message: The data from previous steps. Use Zapier's data mapping to insert dynamic content.

For example, if your trigger is "New Email in Gmail," you would map the email body field into the User Message.

Temperature: Controls randomness (0 = deterministic, 1 = creative).

• Use 0.0-0.3 for classification, data extraction, and structured output
• Use 0.5-0.7 for writing tasks like drafting emails or social posts

Max Tokens: Limits response length. Set this to prevent unexpectedly long (and expensive) responses.

• 100 tokens for short classifications
• 500 tokens for paragraphs
• 1000 tokens for longer content

Mapping Data Between Steps

Every step in a Zap produces output fields. You can insert these fields into later steps using Zapier's data mapper.

Example flow:

1. Trigger (Gmail): Produces fields like From Email, Subject, Body Plain
2. Action (OpenAI): You map Body Plain into the User Message field
3. Action (Slack): You map the OpenAI Response into the Slack message field

This data mapping is how information flows through your automation.

Configuring AI in Make

The OpenAI "Create a Completion" Module

Make's OpenAI module is similar but with a visual configuration panel.

Connection: Select your saved OpenAI connection

Model: Same options as Zapier (gpt-4o-mini recommended)

Messages:

• Add a System message with your instructions
• Add a User message with the data to process

Temperature and Max Tokens: Same guidelines as Zapier

Mapping Data Between Modules

In Make, you map data by clicking into a field and selecting from the available output of previous modules. Make shows a panel with all available fields from upstream modules.

Example flow:

1. Module 1 (Gmail - Watch Emails): Produces fields like From, Subject, Text content
2. Module 2 (OpenAI - Create a Completion): You map Text content into the User message
3. Module 3 (Slack - Create a Message): You map the OpenAI Result into the message text

Using Claude (Anthropic) Instead of ChatGPT

In Zapier

Zapier does not have a native Anthropic integration at the time of writing. To use Claude:

1. Add a "Webhooks by Zapier" action
2. Set the method to POST
3. URL: https://api.anthropic.com/v1/messages
4. Add headers:
   • x-api-key: your Anthropic API key
   • anthropic-version: 2023-06-01
   • content-type: application/json
5. Set the body to the API request format

This is more advanced, so we recommend starting with OpenAI in Zapier.

In Make

Make has a built-in Anthropic (Claude) module:

1. Search for "Anthropic" when adding a module
2. Select "Create a Message"
3. Add your Anthropic API connection
4. Configure the system prompt and user message just like OpenAI

Handling AI Responses

AI responses are text strings. To use them effectively in your automation, you may need to:

Parse Structured Output

If you ask the AI to respond in a specific format (like JSON), use a parsing step:

• Zapier: Use the "Formatter" action to extract values
• Make: Use the "Parse JSON" module to convert text into usable fields

Handle Errors

Sometimes AI calls fail (rate limits, network issues, invalid input). Both platforms offer error handling:

• Zapier: Add a "Filter" step after the AI action to check if the response exists
• Make: Right-click a module and select "Add error handler" to define fallback behavior

Cost Management Tips

AI API calls cost money. Here is how to keep costs low:

1. Use gpt-4o-mini for simple tasks (10-20x cheaper than gpt-4o)
2. Set max tokens to the minimum needed for your use case
3. Use filters to only call the AI when necessary (do not process spam emails, for example)
4. Monitor usage in your OpenAI or Anthropic dashboard
5. Set spending alerts to get notified before hitting your budget

Key Takeaways

• The system prompt is the most important part of your AI automation configuration
• Use low temperature (0.0-0.3) for structured tasks and higher (0.5-0.7) for creative tasks
• Map data between steps using each platform's visual data mapper
• Start with gpt-4o-mini for cost-effective automations
• Always set max tokens and spending limits to control costs