The 20-Page Research Paper in 20 Minutes: AI Is Changing Everything

What are the implications for teaching research when students can generate a college level research paper instantly?

In December of 2022, mere weeks after ChatGPT became available to the public, High School teacher David Herman wrote a piece that went viral entitled “The End of High-School English.” Herman argued:

“The arrival of OpenAI’s ChatGPT, a program that generates sophisticated text in response to any prompt you can imagine, may signal the end of writing assignments altogether—and maybe even the end of writing as a gatekeeper, a metric for intelligence, a teachable skill.”

Herman’s dire prediction was made a little over two years ago, and the AI landscape has grown decidedly more complex since then.

I had a similar moment to Herman while test-driving OpenAI’s Deep Research model released earlier this month. Unlike with typical queries, ChatGPT in Deep Research mode will spend anywhere from 5 to 30 minutes “researching” your prompt (the more detailed and specific the better). It will comb the web for sources, evaluate its progress, and ultimately produce the equivalent of a single-spaced 10 to 20 page report, with headings, sub-headings, and citations. You can observe the research in real time with GPT narrating as it iterates the entire process. You can also toggle back and forth to see the sites and sources that serve as the basis for generating the report.¹

Everyone now has access to a high level research assistant who can do a deep dive on any topic and generate a fully formed piece of writing for your perusal in less than a half hour. In response to your initial prompt, before starting, ChatGPT will ask questions, such as whether you want an academic report, policy brief, or essay, a specified length or word count, the types of sources you prefer, whether to include maps, graphs, or charts, and what type of citation format to use.

In other words, you can tailor your research request to meet almost any specified format or analysis for your needs. As with all AI models, it’s not perfect and “may hallucinate facts or make incorrect inferences”, though at a “notably lower rate than existing ChatGPT models.”² Additional features will no doubt further enhance its capabilities and usefulness.

Reviews of OpenAI’s Deep Research model have been all over the place, but most users, even skeptics, have been impressed. The critics—many of them field experts—point to limitations in the quality of reports and the narrow range of sources used. Yet considering where we were just two years ago, these criticisms seem short-sighted.

The Current State of AI Research Tools

Before I was able to use Open AI’s model, I’d been playing with several others released in the previous weeks and months. Each of them, Google Gemini, Perplexity, and Grok, have similar research models that spend significantly more time than usual creating an output and offers insight in real time into how they are approaching the question. Their final version also contains a list of citations.³

The quality of these vary considerably and have different strengths and weaknesses (for example, Gemeni allows you to open the report directly into Google Docs while OpenAI’s model allows you to attach files and documents to add to its source material when performing its research). Grok and Perplexity, both currently free, were significantly less thorough than either Gemini or OpenAI, which is the clear frontrunner at the moment.

But I was impressed with all of them simply by their mere existence. Like seeing ChatGPT for the first time two years ago, Deep Research models demand that we reexamine our expectations about what is possible as a result of AI.⁴

Given the rapid developments in AI research capabilities⁵, what are the implications of these models on teaching research in schools?

Implications for Research Education

A key aspect of research instruction, especially at the high school level, involves understanding how to locate and evaluate quality academic sources. Students need to understand what kinds of materials to look for (tertiary, secondary, and primary sources) and where to find them (library and print resources, online databases, and other reputable websites, platforms, and publications).

To do this well requires step by step instructions, multiple examples, and practice.

The Deep Research models require us to rethink and reimagine how research might look going forward. At the most basic level, it allows for a jump start to the process that truncates much of the initial “search” skills previous generations had to learn. With a well crafted prompt, a student can have at their fingertips a basic primer on their topic, complete with a long list of potential sources to check out, in minutes.

I can envision three ways Deep Research models will impact classes that teach research.

The first, and likely most common in the short term, will be to ignore them entirely. Most teachers and professors will continue to teach research skills the way they have always taught them and pretend these tools don’t exist or dismiss them outright. And that’s assuming they are even aware of them. Maybe this is the right approach at the present moment, but I do not see how it is sustainable long term.

A second approach involves teachers who acknowledge Deep Research models but move cautiously. They will ask themselves at each stage which “research” skills we teach are worth preserving (source evaluation!) and which may ultimately succumb to off-loading to AI (citation formatting?). This approach will involve risk taking, trial and error, and administrative support, especially in K-12 schools with rigid policies regarding student AI usage.

Finally, a small number of educators may fully embrace Deep Research and set students loose, using AI throughout every aspect of their project. The downsides here are obvious. The majority of high school students are simply not equipped with the basic skills necessary to use the models effectively. Even if they could, many would only superficially engage with their topic and ultimately submit a final product that they barely had anything to do with.

Finding a Path Forward

My current thinking leans tentatively towards the middle ground. I recognize that any short cut, while it may improve the final product, may also short-circuit the valuable “friction” necessary, for example, to understand the difference between an unhelpful and unreliable source and a useful one. However, we must reconsider what we're teaching when we ask students to conduct research in a world where AI can gather and organize information instantly. The reality is clear: today's students will be using AI for virtually every research task, whether personal or academic, in the coming years.

Even with its current limitations, I can see interesting ways in which students can use a Deep Research report as a “source” itself and examine its citations carefully. Unlike a simple Wikipedia page, the Deep Research report gives a much more narrow and focused response if the prompt is well-written, making its citations more targeted and relevant to the student’s topic.

Thirty years ago, the advent of the Internet and the ability for all of us to just “Google it” did something similar. Teachers once thought Wikipedia would be the death knell of academic research, derided in much the same way as LLMs given its potential for containing unreliable information. Today, most teachers are fine with students using Wikipedia as a place to get familiar with a topic, peruse citations, and poke around related or connected issues, before locating more specific and reputable academic sources. Perhaps AI Deep Research reports will be viewed the same way.

But the ability of new AI models to generate sophisticated research papers in minutes isn't just a technological advancement—like so much else about AI, it's a fundamental disruption of how we conceptualize and think about research. We stand at a crossroads: we can either remain fixated on the "cheating" narrative, or we can boldly reimagine how we teach research skills in an AI world. The question isn't whether these tools will transform academic research—they already have. Our students' success in the rapidly changing AI environment depends on educators rising to the occasion and engaging with AI not as a threat, but as an opportunity to redefine what it means to conduct meaningful research in the first place.

1. This post is one of many that offers a more detailed description of Deep Research and its process.

2. OpenAI, Introducing Deep Research.

3. Here are sample responses from each model to the following research question: "To what extent did new military technologies used in the Mexican-American War (1846-1848) influence tactics and outcomes in the early years of the Civil War? Consider both successes and failures in your analysis." OpenAI, Google Gemeni, Perplexity, Grok.

4. When reflecting on the insanely rapid acceleration in AI advances, I'm reminded of Louis C.K.'s observation: "Everything is amazing and nobody is happy." While I may come off sounding “pollyannish” about AI potential for research, it's equally crucial to acknowledge just how extraordinary these capabilities would have seemed mere years ago. Today's "unimpressive" technology was yesterday's science fiction—when you step back to marvel at what these models are capable of, it's impossible not to wonder what breakthroughs are in store in the next few years.

5. In addition to the Deep Research models of the major tech behemoths, specialized AI research platform such as Elicit and Consensus are already offering sophisticated search capabilities with limited free access. Soon, library databases and online research archives will have little choice but to integrate AI-powered search—those that don't risk becoming obsolete. Here is Elicit’s response to the research question posed above.