What is Perplexity Deep Research, and how do you use it?

Besides being better than Google for search, Perplexity, the artificial intelligence (AI) company, wants to be an expert on any subject with its new Deep Research feature. 

This cutting-edge tool, launched by Perplexity AI in February 2025, combines autonomous reasoning with rapid processing to deliver exhaustive reports on specialized topics. 

Also: Perplexity is the AI tool Gemini wishes it could be

According to Perplexity, “When you ask a Deep Research question, Perplexity performs dozens of searches, reads hundreds of sources, and reasons through the material to autonomously deliver a comprehensive report.”

How Perplexity works

The company claims at its core that Perplexity Deep Research employs a proprietary framework called test time compute (TTC) expansion, which enables the systematic exploration of complex topics. 

Unlike conventional search engines that retrieve static results, the TTC architecture mimics human cognitive processes by iteratively refining its understanding through analysis cycles. The system begins by dissecting the query into subcomponents, then autonomously performs dozens of web searches, evaluates hundreds of sources, and synthesizes findings through probabilistic reasoning models. 

Also: What is DeepSeek AI? Is it safe? Here’s everything you need to know

This layered approach allows the AI to reconcile contradictory information, identify emerging patterns, and prioritize authoritative sources — a capability proven by its 21.1% score on the rigorous “Humanity’s Last Exam” AI benchmark. That may sound lousy but, by comparison, GPT-4o scored 3.1%, and DeepSeek-R1 came in with 8.5%. 

Perplexity describes this reasoning as “refining its research plan as it learns more about the subject areas. This is similar to how a human might research a new topic, refining one’s understanding throughout the process.”  

Alternatively, Ken Huang, CEO of DistributedApps.ai and VP of research at CSA GCR, describes TTC as a “model [that] takes input data and applies its learned parameters to produce an output. For neural networks, this involves forward propagation through the network layers using matrix multiplications and activation functions.” 

I don’t know about you, but that doesn’t sound much like how I reason out the answer to a question. 

Also: US sets AI safety aside in favor of ‘AI dominance’

However, Perplexity Deep Research uses parallelized data ingestion and hierarchical summarization techniques to deliver expert-level reports in two to four minutes. A human researcher might spend hours on the same request. 

Early tests suggest that Perplexity’s latest AI tool is faster than Google’s Deep Research for Gemini and OpenAI’s Deep Research. I do wonder, what with all this talk about how revolutionary AI is for business, why all three companies ended up describing their serious research functionality as “Deep Research”. True creativity doesn’t appear to be any AI’s strong suit.

Putting Perplexity to work  

So, Perplexity is very fast, but can it deliver the goods? 

The answer is “sort of”. A TechRadar review suggested Perplexity hallucinated quite a lot. In my tests, I asked the program to dig deeply into three subjects I’m an expert in and that few others are. Those subjects in full: the history and influence of the Commercial Internet Exchange (CIX); the role of Sir Édouard Percy Cranwill Girouard in the East Africa Protectorate; and the history of x86 Unix desktop distributions.

Also: Perplexity lets you try DeepSeek R1 without the security risk

Perplexity delivered a useful abstract for all three topics but in no way an expert-level report. For that depth, you need to hire me. As before, I prefer Perplexity’s output over other AI chatbots because its inline citations make it easy to double-check its answers.

In addition, while Perplexity didn’t make major blunders, it made enough minor ones, so there’s no question in my mind that you can’t just turn in a Perplexity report and expect it to pass muster. No, you still need to check its answers. Welcome to the State of AI in 2025.

Of course, you’re unlikely to need answers to any of those questions. Perplexity claims it’s good for finance, marketing, and product research, so I gave it a question near and dear to my heart: “Tell me how to make a commercially successful Linux desktop.”

The answers the AI gave me sounded good, but all too often, when it came to the fine details, they were wrong. 

For example, the AI said companies and users want long-term stability from their desktop operating systems. That’s true. After all, there’s a reason why, according to Statcounter, 69% of desktop computers worldwide are still running Windows 10. 

However, in the same paragraph that mentions this truism, the report states Ubuntu Linux only has a six-month lifecycle. Ah, wrong. You can now run Ubuntu Linux with support for up to a dozen years. I think that’s enough long-term stability for anyone.

Also: I tested DeepSeek’s R1 and V3 coding skills – and we’re not all doomed (yet)

In short, you still need real experts to double-check Perplexity’s homework even for its target subjects.

Still, Perplexity Deep Research disrupts the premium pricing trend in advanced AI research tools by offering free access to Deep Research, albeit with daily query limits. 

Non-subscribers receive five free daily queries — sufficient for casual research needs — while Pro subscribers, $20/month, get 500 daily queries. That price is much cheaper than OpenAI’s Deep Research, available to ChatGPT Pro subscribers at $200/month. On the other hand, Google Gemini Advanced, which comes with its Deep Research, has numerous other features and costs $20 a month. 

How to use Perplexity Deep Research

Anyone can try Perplexity Deep Research. The tool requires minimal technical expertise:

1. Navigate to perplexity.ai
   

2. Select “Deep Research” from the mode dropdown adjacent to the search bar

3. Input research questions using natural language (e.g. “Comparative analysis of mRNA vaccine platforms”)

4. Await report generation (typically 2-4 minutes)

5. Export results via:

Advanced users can enhance output quality through the following options:

• Query Phrasing: Frame questions with explicit scope parameters (“Limit analysis to 2022-2024 clinical trials”)

• Source Weighting: Prioritize domains via hints (“Focus on NIH-funded studies”)

• Format Directives: Specify structural needs (“Include methodology section with sample size criteria”

Pro subscribers gain additional features, such as custom template creation and application programming interface access for batch processing.

Native Android and iOS apps are scheduled for a second-quarter 2025 rollout, but, for now, you must use the web interface. 

Also: I bought an iPhone 16 for its AI features, but I haven’t used them even once – here’s why

So, is Perplexity Deep Research worth trying? Check it out for yourself and see. You can do enough with the AI for free to get a good idea if it’s helpful. 

Perplexity is still the most useful tool for me when I use it as a replacement for Google, which is why I pay for a subscription. Used with caution, I think many people will find it useful as a starting place for serious research projects.