ChatGPT’s Deep Research Connectors

OpenAI article

ChatGPT's deep research has been around for a while. I have personally used it to conduct research on topics I find interesting, and am blown away by the level of quality and depth of knowledge it is able to consolidate.

But deep research just got a huge upgrade.

Deep Research Connectors allows you to “easily access and analyze content from third-party apps you already use, helping to provide quick, accurate answers directly from your files, code, documents, and more.”

Enabling Deep Research connectors

To connect ChatGPT to your external apps for deep research, go to Settings > Connected Apps. Here you can connect to a couple of pre-defined apps.

Connect to apps like GitHub, DropBox, SharePoint and Box

Note: as of now (23/05/2025), connectors are currently in Beta for ChatGPT Plus, Pro and Team users, with Enterprise coming soon.

How to use Deep Research connectors

The ChatGPT UI is constantly changing - it might even look different a week from now! But for now, here’s how you can use Deep Research connectors:

1. Enable Deep Research

Activate deep research

2. Choose your connectors

Each connector may have a different “mini-UI” for specifying details. For e.g. with GitHub, you can specify the repository you want ChatGPT to search through.

My takeaway

Example of using Deep Research with the Dropbox connector

After testing Deep research connectors, particularly the Dropbox connector, I have gained a solid understanding of the pros and cons of using deep research, versus building a custom RAG solution.

Deep research connectors

• Pros

  • Very easy and simple to use

  • High quality and accurate outputs

  • Able to handle questions that require data from multiple files

  • Able to thoroughly search through Word documents, Excel files and PDFs, with complex tables and graphs

• Cons

  • Can be extremely slow in response time (from 3 to 10 minutes). My tests were on a Plus account, not sure if it's any faster on Pro/Enterprise.

  • It does not seem to create an index of the file sources, and it sometimes struggles to find the particular file that you’re referencing.

Custom RAG

• Pros

  • RAG sources can be indexed, thus allowing more powerful search

  • Higher control over the UI/UX and custom features

  • Can use multiple models, not just OpenAI’s

  • Much faster response time

• Cons

  • Takes time to develop

Deep Research connectors = RAG?

One question you might have is whether ChatGPT is using RAG (Retrieval Augmented Generation) to pull information from the third-party sources.

It depends on how you define RAG. The simplest definition would be improving a model's response by grounding it in external information (info it wasn't necessarily trained on). Since it's returning data from documents (albeit not through indexing, just through text search), it's still augmenting the final response. By this way of thinking it would be considered RAG.

It's tough to say whether there is any indexing (with embeddings) going on at all - perhaps when the connection to GitHub/DropBox is first made, ChatGPT creates a small index of your files. But then it would also have to continuously update this. So it’s tough to say whether ChatGPT is indexing the files it references.

Either way, it’s able to pull accurate information from the documents it has access to.

Conclusion

In conclusion, Deep Research Connectors in ChatGPT enhance its ability to conduct thorough research by integrating with platforms like GitHub, Dropbox, SharePoint, and Box. Now we can easily access data from documents, code, and files with minimal setup.

The speed and accuracy of this system is likely to improve over time, with more connectors being added and more powerful search baked in.

Thanks for reading!