ChatGPT Deep Research API: Access and Research Workflow

Deep Research in ChatGPT and research-capable models in the OpenAI API are two different things. Deep Research is a feature of the ChatGPT app that runs extended, multi-step web research on your behalf and produces a structured report. The OpenAI API gives developers access to models — including reasoning-focused ones with web search tools — that can be used to build research workflows programmatically. Both are useful; they are not interchangeable.

A research workflow produces a report, but it also creates a set of follow-up questions, verified sources, and decisions about which findings to act on. Keeping that record outside the chat — in a tool like fenbs — means it is available for the next session, accessible to teammates, and linked to the tasks the research informs.

The Deep Research feature in ChatGPT

Deep Research is a ChatGPT app feature that conducts multi-step web research automatically. You give it a research goal, it browses and synthesizes information from multiple sources, and it produces a structured report with citations. The process typically takes several minutes.

Deep Research is a product feature, not an API. You access it through the ChatGPT web or mobile app. It is available on certain ChatGPT subscription plans — for current availability and plan details, check OpenAI's pricing page↗ (opens in a new tab). The feature uses a reasoning model with web browsing capabilities; the exact model may change.

The output is a long-form report. The citations in that report are the key deliverable for source verification. Before acting on a Deep Research report, open the cited sources and confirm the key claims are supported. Synthesis models can misrepresent a source's meaning or weight, even when they cite it correctly.

Deep Research is useful for covering a topic quickly with a cited overview. It is less suited to tasks that require accessing non-public sources, internal documents, or real-time data that the model cannot browse to. For those cases, a hybrid approach works better: use Deep Research for the public-source background, then add your internal context separately.

API access vs the ChatGPT app feature

The OpenAI API does not expose "Deep Research" as a named endpoint. What the API offers are models, some of which have web search tools attached. Developers use these to build research workflows: a model that can search the web, retrieve pages, and reason over the results.

Build a research workflow with the API

The OpenAI API provides models with built-in web search capabilities. Call them with the responses endpoint and include the web search tool:

from openai import OpenAI

client = OpenAI()

response = client.responses.create(
    model="o3",  # or another model with web search; check OpenAI docs for current options
    tools=[{"type": "web_search_preview"}],
    input="What are the current FTC guidelines on endorsement disclosures for social media?",
)

print(response.output_text)

The model can call the web search tool multiple times to refine its findings. The output_text field in the response contains the synthesized answer. The OpenAI API documentation↗ (opens in a new tab) is the authoritative source for the current list of models that support built-in tools and how tool use is billed.

Handle research results responsibly

Research outputs — whether from the ChatGPT app or the API — should be treated as a starting point, not a final answer. Key practices:

• Open the cited sources and verify the key claims before sharing or acting on the report.
• Note which claims are supported by a single source and which are corroborated by multiple independent sources.
• Record the research brief — the exact question you asked — alongside the output, so you can reproduce or extend the research later.
• Track which findings created follow-up tasks, and keep those tasks linked to the research that prompted them.

How to do this with fenbs

1. Create a free account and add a project for the research area. Write the research question and any constraints — the scope, the sources to prefer, the sources to avoid — as project rules.
2. Run the research in ChatGPT Deep Research or via the API. Save the research brief and the output to the project.
3. Add the key sources as Links in fenbs with a status note: authoritative, unverified, contested, or outdated.
4. For each finding that leads to action, create a task in fenbs linked to the source that supports it.
5. Connect a supported AI app to fenbs from Integrations. Ask it to read the project context before extending or updating the research.

The claude vs chatgpt comparison covers how the two models differ for research-style tasks. The api documentation guide explains how to read and use OpenAI's API reference. The agent harness topic hub has more on building research workflows.