Deep Research: Grok 3 vs. ChatGPT vs. Perplexity vs. Gemini

Deep Research is currently a major trend among AI chatbots.

And for good reason:

The feature where an AI chatbot performs deep internet research - autonomously searching for many (!) relevant sources on the internet and evaluating them on its own - is one of the first truly well-functioning use cases for autonomous AI agents.

I've tested the Deep Research functions of many different chatbots for you. Here's what I found:

1. ChatGPT Plus & Pro

To thoroughly test ChatGPT's Deep Research, I specifically got ChatGPT Pro (the most expensive subscription at $200/month):

Unlike ChatGPT Plus, it offers not just 10 Deep Research uses per month, but unlimited access.

I was particularly interested in Deep Research with o1 pro, OpenAI's currently best reasoning model.

And the latter is really good. Much better than I expected.

It's great for getting a deeper overview of a topic, writing newsletters or blog articles. It's also excellent for academic papers, as it can cite sources very accurately (including with the appropriate citation style when asked):

What unfortunately doesn't work as well sometimes is follow-up questions or revising generated research.

The first version of my AI newsletter (see excerpt above) was great, for example. After that, I wanted to make some changes and additions, which didn't work so well:

What I also don't like is the formatting of the research. While Deep Research in some other tools contains too many "bullet points," with ChatGPT Deep Research you often get a wall of text with very long paragraphs:

2. Perplexity Deep Research

Perplexity Deep Research is much cheaper at $20 per month than ChatGPT Pro and not much worse for it.

While the "initial research" isn't as comprehensive and accurate as ChatGPT Pro's (ChatGPT typically includes more sources).

However, Perplexity is significantly faster, provides better-structured output, and is equally good at assigning sources to text excerpts:

The big advantage of Perplexity is that it's better at follow-up questions and continuing research than ChatGPT (I'm not exactly sure why, but I suspect it has something to do with the context window):

It's also super helpful that Perplexity offers possible follow-up questions after each answer:

What's unfortunate:

Unfortunately, you can't choose the model for Deep Research yourself. I suspect either Claude 3.7 Sonnet or a modified variant of DeepSeek R1 is being used (I consider the latter more likely, since Perplexity hosts DeepSeek R1 themselves, which is significantly cheaper than using Claude 3.7 Sonnet via API).

3. Grok 3 Deep Research

You can think what you want about Elon Musk. But Grok 3 and its Deep Research function are really good.

The results are comprehensive, contain many source links (directly in the text), are mostly accurate, and, unlike o1 pro Deep Research, better structured:

The only downside:

Grok 3 works best in English. In German, it tends more than other AI models/chatbots to produce "Denglish" (German-English mix).

4. Gemini Deep Research

Google Gemini's Deep Research feature (available in Gemini Advanced) lands in last place for me.

It's not bad in itself and would probably rank somewhere between second and third place.

Many sources are searched and the research reports are very well formatted and coherent (which is no surprise, as Google is the leader in search engine technology):

The problem currently is:

Unfortunately, it doesn't work well yet and is highly error-prone. It took me 4 attempts for Deep Research to start. On the attempt that worked, I needed an incredible 5 prompts.

For example, the very first prompt is always just repeated (without the research starting):

Very high frustration potential. What a shame...

Deep Research Comparison Table