I tried deep research on ChatGpt, and it's like a very smart but slightly absent librarian from a children's book.

I’ve always been the type of person who gets lost in research. Some people scroll through social media before bed; I delve deep into archaeological studies of the use of color and blogs obsessively chronicle the evolution of old TV shows. When Openai developed ChatGPT’s new deep search feature, it seemed like a dream companion, capable of conducting real-time web research and compiling detailed reports on its own.

Deep Research started out as an exclusive feature of ChatGpt Pro, the $200-per-month subscription service, but is now available to those who pay $20 per month for ChatGpt Pro, though you only get 10 deep research queries per week at that tier.

At its core, deep research attempts to do what I, and probably many others, already do when we have a question that’s too big for a quick search. Researching something usually means clicking through multiple sources, separating marketing fluff from useful insights, and resisting the urge to fall down irrelevant Wikipedia rabbit holes. Deep research claims to take all of that work off your plate, giving you a curated report instead.

And ChatGPT isn’t the only AI trying to tackle this problem. Perplexity has a feature with the same name and broadly similar goals, as do Google Gemini and Deepseek. Each system has its own quirks, but ChatGPT’s deep search, at least in theory, aims for something more structured and thoughtful—a full report rather than a handful of search results.

I decided to put it to the test with three research challenges I thought might be fun. The reports were impressive, but sometimes a bit rambling. Imagine a brilliant but slightly absent-minded librarian who can find you an obscure 18th century manuscript in under five minutes and sometimes hand you a 20-volume dissertation when you ask for a little reading on the beach.

espresso

I started with a request for help choosing an espresso maker. ChatGPT asked me some follow-up questions about price and other details, leading to this as a final prompt: “Provides a beginner's guide to setting up an espresso station at home, including recommendations for budget-friendly espresso machines, grinders, and accessories, along with maintenance tips and common beginner mistakes.”

A regular chatgpt response is almost instantaneous, but a deep search can take anywhere from five to thirty minutes to return results, depending on the complexity of your request. This took about ten minutes, but the deep search returned with a very comprehensive Espresso setup guide.

It covered everything from machine recommendations (Breville Bambino, Gaggia Classic Pro, and a few others) to grinder options, the importance of fresh beans, and even a brief lesson on coffee extraction. It also covered common beginner mistakes like using pre-ground coffee, not weighing shots correctly, or ignoring the importance of good milk.

There were quibbles. Some of the product recommendations leaned toward pricey options when there were budget-friendly alternatives. But it was a helpful, almost enthusiastic guide in its accuracy, but I like it.

Star Search

For my next order, I went with something I'd been thinking of pursuing as a hobby locally, and it ended up being this prompt: “Provides an overview of beginner-friendly astronomy, including equipment needed, recommended resources for learning, and local astronomy clubs or events in the Nyack, NY area.”

Deep Research provided an introductory guide to amateur astronomy covering telescopes, binoculars, and the naked eye. It gave some good recommendations for equipment and locations to go to, and even websites and apps to plan nights and celestial events to look for. It also listed nearby astronomy groups and clubs that I might join.

Despite being neutral in tone, there was a lot of enthusiasm baked into the report that I found charming. The equipment suggestions were logical and didn’t automatically go for the most expensive options. Some of the event information it suggested was a bit outdated, but that seems more the fault of the sites not being updated.

Monster Mash

For my last test, I went with something a little less to see how the AI ​​handled a report about something based mostly on rumors: “Investigating the origins and history of the legend of the ‘Lake George Monster,’ analyzing the earliest known references to it, how it has evolved over time, and whether there is any real historical basis behind it.”

This took the shortest amount of time, just about five minutes. A fictional character confined to local legend would probably take less time to research. However, the deep research came back with a surprisingly detailed breakdown of the Lake George Monster, a local legend in New York State. He traced the first major sightings to the late 1800s, citing old newspaper clippings that described a mysterious, serpent-like creature lurking beneath the lake. He explained how the legend was fueled by hoaxes, including a 20th-century prank involving a mechanical sea creature built by a local prankster.

I also tried to analyze the plausibility of a real creature living in the lake, pointing to known aquatic wildlife and the scientific skepticism surrounding such legends.

As a report, this was definitely the most enjoyable one to read. It read like a good local historian, complete with source citations and fun anecdotes. It even mentioned other lake monsters, like the hero from Lake Champlain, drawing comparisons to regional folklore. It wasn’t without its flaws. While it did a great job of recounting past sightings and debunked hoaxes, it struggled to make clear which sources were firsthand accounts versus modern retellings. But as I read about a bit of local culture, it was a pleasure.

deep thoughts

Deep Search is one of the more ambitious AI tools, and I have to admit, I kind of like it. It feels like reporting from someone else who enjoys exploring through the weeds for hidden gems as much as I do. And I will say that, compared to regular ChatGPT responses, Deep Search felt like it made a real effort to find new and relevant information.

It's far from flawless, but when it works, it really does the trick by providing organized, easy-to-read reports that save you time and effort. Instead of clicking through endless links, fact-checking articles, and wondering if a recommendation is actually useful or just a cleverly disguised ad, you get a report that at least tries to distill everything for you.

I wouldn’t trust it with choosing a car seat for my child, but I can at least say it might give me a place to start my own research. That’s the thing with all AI research tools, of course. A librarian, a search engine, or an AI report is no substitute for the work of finding and organizing information, but it does help simplify the process sometimes.

However, understanding that deep research is not, in fact, actual deep research, can be a great way to get a start.

• I tried to dig deep into the confusion and it does not live up to the potential of ChatGPT research.
• لقد استبدلت قائمة المهام الخاصة بي بميزة مهام ChatGpt وقمت بتغيير طريقة التخطيط لحياتي تمامًا
• Google Gemini Adds Personal AI Researcher to Your iPhone — If You Have the Right Subscription
• I pitted Gemini 2.0 Flash against Deepseek R1, and you might be surprised by the winner