"Deep search" - what is it?

Imagine you need to write a summary about the history of pets in Sweden, and you want help finding relevant sources online and summarising them. This is where ChatGPT’s new “deep research” feature enters the picture - a digital research assistant that helps you do the heavy lifting.

If you’d prefer to have ‘deep research’ explained in ten minutes by one of the most prominent AI researchers and co-founder of ChatGPT, I highly recommend this segment from Andrej Karpathy’s video “How I use LLMs” - watch between 42:01 and 51:03. And do watch the rest as well - it’s an incredibly good overview of different ways to use AI-based tools!

OpenAI recently made the “deep research” feature available to paying ChatGPT users (through the “Plus” and the more expensive “Pro” plans). Deep research is a feature that starts from a broad, open-ended question from the user (which could be “describe which families have historically lived in our neighbourhood in Mariefred”, or “please help me summarise work being done in ethical application of AI as of 2025”, or perhaps “please make a report on the current state of research of potential efficiency gains of domestic farming in Denmark”).

The feature will then ask the user a few follow-up questions to make sure the query is clear and well-defined.

After that, it performs multiple internet searches, analyses the results it receives, and depending on what it finds, runs additional searches.

In the end, a report is generated for the user. The report includes links to the sources the information came from, and is more comprehensive than what a regular ChatGPT response tends to be.

The quality of these reports varies (the feature is not infallible - it can miss information or make things up (“hallucinate”)). That said, there are users who already see significant value in the feature - particularly when they compare the quality and cost of manually producing similar reports.

There are several products that offer similar functionality, and most of them have almost the same name (but not quite):

  • Elon Musk’s x.ai offers the product Grok, where the feature is called “DeepSearch”.

  • Google has the product Gemini, and the feature “deep research” (that is - the same name as the feature in ChatGPT).

  • The Chinese company DeepSeek calls its version “DeepThink”.

  • In search-focused Perplexity, it’s called “Deep Research”.

  • (Anthropic’s product Claude does not currently have the ability to search the internet and therefore lacks an equivalent feature).

So which of all these is the best? Which one should I use?

Right now, it seems like “deep research” in ChatGPT delivers the best results. It is, however, limited to 10 searches per month for a “Plus” user (something that will certainly change over time).

In upcoming blog posts, I’ll write about different ways to do “deep searches” with various tools, and how I perceived the results (and the risks) of using them!