OpenAI announces Deep Research

Yesterday OpenAI followed up the Operator release with Deep Research, an agent that can conduct multi-step research for you. Like the Gemini feature it shares a name with, ChatGPT Deep Research can source information across the web for you, reason over it, and ultimately provide a more comprehensive and extensively cited answer vs. ChatGPT Search.

Tl;dr for web pros: Deep Research uses the same underlying methodology as ChatGPT Search. If you want to enhance your content's visibility in Deep Research, start with ChatGPT Search. Blocking ChatGPT-User should also block Deep Research. More details below, but if you're concerned about how your site functions with AI search, Deep Research or Operator, get in touch!

Positioning: Like Operator, OpenAI positions Deep Research as a step towards agentic AI. I see them as very different, both in terms of market dynamics and from a technical perspective.

Operator is remarkable because it's the first "robotic" browser use agent that's oriented towards non-technical users. (For devs or automation experts, there's lots of prior art for Operator.) On the other hand, Deep Research has more market precedent in the form of Google's NotebookLM and Gemini Advanced Deep Research and (to some extent) Perplexity's Pro Search product.

Technology: From a technology point of view, the emphasis in Deep Research should be on the word "search". Intuitively, you might expect Deep Research to browse around your website to find related resources – for example, following links to cited papers or browsing to the "pricing" page when it lands on a product page.

In testing, Deep Research does not do this. Instead, it runs multiple searches – for both breadth and depth – as it tries to find information relevant to the user's question. Like a human, it will try different search terms if the first one isn't helpful, and run additional searches if the content needs more context.

Takeaways for site owners:

• Like ChatGPT Search, only content that is indexed by search engines is findable. Deep Research respects robots.txt.
• Unlike Operator, Deep Research cannot interact with websites. It goes back to search when it wants more info (although it can retrieve multiple sources for the same search.)
• The Deep Research user agent is the same as the ChatGPT Search user agent.
• Like ChatGPT Search, Deep Research does not retrieve or execute JavaScript (or any related site assets.)
• Deep Research will conduct additional searches based on the content it retrieves – so if your site content references other content, concepts or brands, Deep Research may search for them. However, there's no guarantee that this follow-up search will use your content, even if it's linked from the original source page.

As always, AI platforms continue to evolve rapidly, so these are subject to change in the future.