Very few former students can claim they never wrote a last-minute research paper the night before it was due. AI tools are already giving students all new ways to fake their papers. Now, OpenAI’s new “Deep Research” tool seems perfectly designed to help students fake their way through a term paper unless asked to cite sources that don’t include Wikipedia.
OpenAI’s new feature, built on top of its GPT-4o model and released on Sunday, resembles one Google introduced late last year with Gemini 2.0. Google’s “Deep Research” is supposed to generate long-form reports over the course of 30 minutes or more, depending on the depth of the requested topic. Boiled down, Google’s and OpenAI’s tools are AI agents capable of performing multiple internet searches while reasoning about the next step to generate a report.
The Deep Research feature is available for ChatGPT Pro subscribers (at $200 a month) in a new tab underneath the prompt box. Simply put, it’s doing the Googling for you. Instead of deciding on your own sourcing, the chatbot will do all that work for you.
For example, OpenAI asked ChatGPT to look at iOS and Android adoption rates and the percentage of those who want to learn another language, then find “markets ChatGPT could better expand into.” The chatbot interface then asked for clarification on several points, such as whether engaging with a language app constituted interest. It sounds like the kinds of questions your college professor in a 400-level course would ask when presenting your planned research paper topic.
Unlike Google’s tool, OpenAI’s Deep Research works in a side panel. The chatbot will update you about what it’s working on. In one of the examples OpenAI shared in its video demo, the AI searches for surveys on language learning by country and reports, “I’m looking into recent surveys and reports on smartphone penetration and OS adoption to understand digital trends and language interest globally.”
Occasionally, the bot will cite what sources it’s looking at, namely Wikipedia. In that market research example, the bot repeatedly looked at Wikipedia, just like every lazy undergrad working on a last-minute paper the night before it’s due. Remember, this is just a bot scraping the internet, so it won’t be accessing any non-digitized books or—ostensibly—any content locked behind a paywall.
Once the chatbot generates the report, you can find the full source list in a separate panel. Because it’s essentially an auto-Googling machine, the AI likely won’t have access to the most up-to-date and large-scale surveys from major analysis firms. In the final paper on smartphone adoption rates, the bot cited data from some trusted sources like Pew Research. Still, it also had various sources, including Wikipedia, Statista, and the UAE-based industry magazine Gulf Business. That’s not to say the information was inaccurate, but anybody who generates a report is at the mercy of suspect data and the AI’s interpretation of that data.
In one of OpenAI’s examples on “Civilian Supersonic Air Travel – 10-year Market Analysis” (with the bolded H2 title and all), the AI cites Wikipedia, the webpage for Boom Supersonic, the developer of the XB-1 jet plane, and the front page of a few industry publications. In another example, OpenAI said its model could find multiple papers on advanced biology topics. However, another research paper breaking down an episode of Counterpart ripped parts of its info from a fan wiki.
The process does take much longer than any typical ChatGPT request. Depending on the depth of a request, it could range from under 10 minutes to up to 30 minutes or more. OpenAI claims it boils down to what would normally take humans “many hours.” The Sam Altman-led company claimed its Deep Research tool scored higher than Gemini Thinking, GPT-4o, and on Humanity’s Last Exam, an AI benchmark test from the Center for AI Safety. If the name of that benchmark sounds too close to OpenAI’s and Altman’s “end of humanity” doomerism about the future of AI, that’s because the group that made the test has close ties to the AI industry and tech billionaires.
OpenAI admitted that Deep Research is “very compute intensive.” However, it was working on creating a version that would let Plus and Team subscribers eventually use it, likely using a less powerful AI model. Most college students won’t pay $200 monthly for a cheating tool. Eventually, those who can’t spend as much may also be able to pay for their Wikipedia-citing machine.
English (US) ·