Earlier this year, when deep research models from OpenAI and other sources started appearing, these received a lot of attention from AI writers. OpenAI’s deep research tool, which integrates with ChatGPT and provides users with five free research prompts a month, was often hailed as the best or one of the best deep research tools.
Over the last few months, I’ve used ChatGPT’s deep research feature, and while I don’t agree with those who claim it’s as good at research or writing as most graduate students, it’s certainly a useful tool for me. I believe many educators would find it helpful when looking for research on various educational strategies or for the latest findings in their field of study.
Like many, I’ve increasingly begun to use AI as an alternative to a traditional search engine. Deep Research works as a beefed-up version of that. It can be quicker and more efficient than searching a site such as Google Scholar, and can help you find different results than you might find searching more traditional academic archives.
Here’s a closer look at how OpenAI’s Deep Research works, and how it performed on various academic prompts with which I tested it.
How ChatGPT’s Deep Research Works
Deep research is designed to work independently by finding, analyzing, and summarizing hundreds of online searches. The tool is powered by OpenAI’s o3 model and uses reasoning—meaning it can break down problems into smaller components, make predictions, infer information, and apply logic.
Those who use the free version of ChatGPT, get five free deep research prompts per month. Plus, Team, Enterprise, and Edu users get 25 per month, while Pro subscribers get 250 prompts.
To access Deep Research, you need to click on the tools icon at the bottom of the chatbot used to chat with ChatGPT. Then select “run deep research.”
After you put in a prompt, the tool will ask several clarifying questions about the type of research you want. Once you’ve answered these, it will take a few minutes to generate a response.
Here are some of the prompts I used and what I thought about the results Deep Research generated.
Flipped Learning Prompt
The Prompt: Can you write a summary of the research on flipped learning? Pay particular attention to randomized control trials looking at the educational strategy in college and k-12 settings.
The Result: ChatGPT quickly responded with a detailed summary of the research into flipped learning with effects divided between college and K-12 research. The summary concluded the intervention had a moderate but fairly consistent effect observed in most studies. ChatGPT provided links to pertinent research with which I was already familiar, but which had taken me quite a while to track down. Additionally, ChatGPT provided several interesting studies I hadn’t come across.
Impressively, the overview also described some of the limitations around flipped learning, addressing how it can increase the workload outside of class and how teachers who attempt to implement it can find significant pushback from students. ChatGPT also cited a 2019 paper describing a randomized control trial looking at flipped learning conducted at West Point that found short-term gains in Math and no effect in Economics. The study also suggested that the intervention could widen the achievement gap: as positive effects were driven by white male, and higher-achieving students.
The study is an outlier, so it’s hard to say what its significance is, but it was interesting to see and an important part of the flipped learning conversation.
Learning Styles Prompt
The Prompt: Is there any evidence that learning styles are real that you can find in academic journals?
The Result: ChatGPT did a nice job pointing out the problems with this approach to education, writing: “Major meta-analyses and reviews – especially in the last 15 years – largely refute the learning styles hypothesis, or at best find only weak effects. The current consensus in the academic community is that educators should be skeptical of the learning styles approach.”
The overview it generated also included many studies that supported this conclusion. Because of the way my prompt was phrased, I think it encouraged ChatGPT to look for evidence in support of this theory. It found some older studies that suggested a link, but was quick to point out the problems with this earlier research.
I was impressed with ChatGPT’s response. I’ve written about the persistent myth of learning styles before, and worried ChatGPT’s research mode would be confused by all the online chatter from educators in support of this debunked theory. It wasn’t.
Writing and Cognition Prompt
The Prompt: Can you research the potential positive link between writing and thinking/cognition?
The Result: This is a topic I’m curious about as a writer and writing professor, but I’ve only come across a few studies and haven’t yet done a deep research dive myself. ChatGPT may have saved me the effort.
The response here provided a wide-ranging and well-cited overview of different ways in which writing can help our thinking processes, from increasing memory to reducing stress and improving critical thinking and metacognition.
One example of the type of research included in the summary is a fascinating study that found college freshmen who wrote about the stress of starting college saw larger working memory improvements than their classmates who did not have this writing assignment.
Conclusion
Deep Research is one of the more helpful AI tools I’ve come across for academics, and I believe it can be helpful to educators in various fields and levels. By essentially generating high-quality Wikipedia articles on demand, it’s a great search tool that can help make research more efficient. I’ve found it helpful in the examples above and in other uses.
However, much like Wikipedia itself, it’s a good place to start your research and a bad place to end it. Deep Research can help you get a fast overview of a topic and suggest research to review, but actually assessing that research and deciding what implications it has is still best performed by a human.