When we talk about “deep research” in Make.com using OpenAI, there are a few important steps and two major things to keep in mind. Let’s break it down clearly.

Step 1: Create a Model Response

The first thing we need to do is open the OpenAI node and select the action Create a model response.

Here, you will configure:

• Model: Choose the model you want (for example, o4-mini-deep-research).
• Prompt: Define the exact prompt you want to pass. This is where you explain the research task.
• Tools: Add the tools that OpenAI can use during research. Examples:
  • File Search
  • Web Search Preview
  • MCP
  • Function
  • Code Interpreter
  • Custom

These tools allow OpenAI to go and fetch, search, or process the information needed for deep research instead of just relying on the input prompt.

Two additional configurations are critical here:

• Parallel Tool Calls: Keep this to Yes so OpenAI can call different tools at the same time.
• Run in Background: Keep this to Yes so the research can continue even if it takes longer.

Temperature and other parameters can also be adjusted depending on the depth or creativity required.

Step 2: Get a Model Response

Once the model response is created, the next step is to use the Get a model response action.

This basically helps in retrieving the Response ID of the deep research task that was triggered earlier.

This is important because deep research runs asynchronously. Instead of just returning one big final answer, OpenAI breaks down the response into different sections or chunks.

Step 3: Iterating through the Output

Now comes the part where we process what OpenAI gives us. The response is not always perfectly structured.

It comes in bits and sections.

To handle this, we use Iterator in Make.com to go through each part.

After iterating, we use Text Aggregator to merge all the pieces into a single response.

This ensures that you don’t lose any information and the output becomes one consolidated research answer.

Step 4: Structuring the Final Answer

Because the output might still not be in a clean JSON or consistent format, we can add another OpenAI node with Create a completion (prompt).

This final node can take the aggregated response and structure it properly — for example, into JSON format, or into sections like:

• Introduction
• Findings
• Comparisons
• Sources
• Final Summary

This way, the deep research doesn’t just stay raw, but becomes neatly formatted for use in Google Docs, Sheets, or any other app you connect in Make.com.

Summary

So in short, the process looks like this:

• Create a model response → define model, prompt, tools, enable parallel + background.
• Get a model response → capture the response ID and results.
• Iterate + Aggregate → loop through sections of output and combine them.
• Refine (optional) → use Create competion to structure the final research output.

This is how deep research can be automated inside Make.com using OpenAI.