Is Deep Research Worth $200/mo?

In partnership with

Welcome to AutomatED: the newsletter on how to teach better with tech.

In each edition, I share what I have learned — and am learning — about AI and tech in the university classroom. What works, what doesn't, and why.

Today, I discuss the second “Deep Research” AI tool released in the past few months. This time around, the tool comes from OpenAI, competing with Google’s version, and it has many academics impressed and worried. I explain how it works and share one of its outputs below so you can see for yourself what it can do.

📣 By the way, feel free to email me if you handle your department’s faculty-to-course or course-to-room scheduling and you’re interested in the AI tools I’ve developed to help improve your processes — saving you time and producing more optimal schedules, without requiring any new inputs from your colleagues. I’ve only got 2-3 more slots available, as a range of departments and other academic units around the world are going to be using my service this semester.

🧰 An AI Tool For Your Toolbox: 
OpenAI's Deep Research

Should you get ChatGPT Pro for $200/month?

A month ago, my answer would have been "probably not."

o1-pro, the reasoning AI model that was the main benefit of Pro, truly is amazing. In fact, I’ll show off some of its powers next week when I discuss its viability as a thought partner to cutting-edge researchers.

But other tools are enough for most use cases, from the freely available Gemini models in Google AI Studio to the custom GPTs you can build with ChatGPT Plus for $20/month.

And this is all the more true now that everyone can freely use o3-mini, which goes toe to toe with o1-pro, via ChatGPT.

However, OpenAI's new Deep Research feature, launched last week, changes the equation and should tempt all of us to find the funds to get access to ChatGPT Pro.

It's the first AI tool I've seen that can immediately assist with academic research — both your own and your students' — without the user having both field-specific knowledge AND prompting skills.

OpenAI’s Deep Research is more than just ChatGPT with web access (that’s ChatGPT Search). It's a specialized research agent that spends 5-30 minutes autonomously exploring academic sources, analyzing findings, and producing detailed reports with proper citations.

And the crucial thing is this: it’s really good at playing this role, out of the box.

Unlike Google's similar tool, which aggregates a wide range of Internet sources quickly, OpenAI's version performs something closer to genuine scholarly synthesis: it identifies patterns across sources, draws meaningful connections, and structures findings based on your specific interests and parameters. (Google’s tool is limited to a narrow range of output formats.)

And you can guide OpenAI’s research process through back-and-forth dialogue, much like working with a research assistant, while Google’s tool is somewhat inflexible in my experience.

One month ago, I declared, after testing Google’s tool, that "the research report is dead and we have killed it” and that Google’s tool was to the research report what ChatGPT was to the essay. I was very impressed, despite the flaws of Google’s Deep Research.

Only one month later, I’m blown away by how much better OpenAI’s tool is, although I’m sure Google will provide good competition throughout this year and beyond.

How It Works

To access it after purchasing ChatGPT Pro, you select "Deep Research" from button below the chat field.

You can pair it with different models via the model selector in the upper left corner of your screen, but its best performance comes with smarter reasoning models like o3-mini-high or o1-pro.

In terms of prompting, no special prompt engineering skills are needed.

It is key to be detailed about what you want — the format, scope, other relevant information, and type of sources to focus on — because it performs best when you provide a ton of context and background. Don’t be afraid to dump more information in!

Deep Research then plans its approach, clarifies what you want (if necessary), systematically explores sources while you watch (if you want), and delivers a comprehensive report with clear citations/links.

You can further refine the output through follow-up questions or by providing additional context.

How We Can Use It

For professors and learning specialists, Deep Research opens up two key possibilities:

First, it's a powerful tool for your own research and course development.

Whether you're designing a new syllabus, keeping up with developments in your field, or preparing lecture materials, Deep Research can accelerate the initial stages of academic investigation. Unlike standard ChatGPT, it provides verifiable citations and draws from current academic sources, so long as they are available in some form on the Internet.

A task that might have taken hours of preliminary research — like finding alternative readings for a course unit or surveying recent developments in your field — can be completed in minutes, letting you focus on the deeper analysis and synthesis that requires your expertise.

Second, and perhaps more importantly, Deep Research changes how we need to think about teaching research skills.

Just as ChatGPT transformed writing pedagogy, this tool requires us to rethink how we teach students to conduct research. The ability to quickly compile and cite sources is no longer a differentiator — instead, we need to focus on teaching students how to:

• Frame sophisticated research questions

• Evaluate and critique AI-generated research syntheses

• Identify what Deep Research misses or misunderstands

• Use AI as one tool in a broader research workflow

But I will go more into pedagogical implications at a later date. For now, my focus is on showing you what the tool can do.

Let me demonstrate Deep Research's capabilities with a real example from my own course development...

✨ Demonstration & Discussion

To test Deep Research's capabilities in course development, I asked it to analyze Module 2 of my "AI and the Future of Humanity" syllabus. The philosophy course, which I taught several times at UNC-CH, covers the ethical, political, practical, metaphysical, and epistemological challenges and opportunities presented by AI. (The screenshots above are from this request.)

Specifically, I wanted it to find and review nine key readings — from Sven Nyholm on self-driving cars and Regina Rini on deepfakes to Natasha Singer on AI tutors and Philippe Van Parijs on universal basic income — and suggest potential improvements to my syllabus that would better serve the course objectives. I only provided it with a prompt and a copy-paste of my syllabus, with each reading identified only by the author’s last name and a topic under which it fell.

Ideally, I would have provided more information, both about my readings and about ways that I already know my course could be improved, but it still performed amazingly, all things considered.

Deep Research figured out which journal article or think piece I referred to (from context plus its searches) and proposed a really impressive set of alternative readings.

Here’s the prompt I used, followed by a link to the results and my discussion of the upshots of this new tool:

After reviewing, analyzing, and synthesizing a report on 19 sources over 7 minutes, Deep Research produced this report that I share via the “share conversation” link (so you can see the full context):

📢 Quick Hits:
AI News and Links

1. Google has outlined eleven ways educators can use its AI tools in 2025, including:

• NotebookLM for creating lesson plans and study guides with source citations

• Gemini integration into learning management systems like Canvas

• NotebookLM Plus which allows professors to create shared notebooks containing course materials.

2. ICYMI (“in case you missed it”): OpenAI has announced o3-mini, a cost-efficient version of its reasoning model that matches OpenAI o1's performance in STEM fields while being 24% faster. The model is available through both ChatGPT and API, with free users getting limited access for the first time (just select “Reason” in the message composer). Notable features include three "reasoning effort" levels, function calling capabilities, and integration with web search, though it lacks vision capabilities. Their showed users preferred o3-mini's responses to o1-mini 56% of the time, with 39% fewer major errors. I’ve found it is extremely reliable and powerful for a range of my AI tool development use cases.

3. Here’s a deep dive into how LLMs were developed, how to conceptualize them, and how to use them, from Andrej Karpathy (an OpenAI co-founder).

4. Statistician Nate Silver argues that AI will be a major political issue after being largely absent from the 2024 election. He criticizes the political left for dismissively comparing AI to crypto, arguing they risk being "dealt out of the hand" on crucial policy decisions. Silver suggests AI will rank at least "a high 7 or low 8" on his "Technological Richter Scale" of societal disruption, even without achieving superintelligence, and estimates a 5-10% chance of existential risk aligned with expert consensus.

5. Google has made Gemini 2.0 generally available, releasing three versions: Flash (their workhorse model with 1M token context), Pro (their most capable model with 2M token context, especially for coding), and Flash-Lite (a more cost-efficient option). While all versions currently support multimodal input with text output, Google says additional output modalities like image generation and text-to-speech are coming soon. (Try them in their raw forms in Google AI Studio here; Tutorials on using this tool here and here).

6. OpenAI has launched "ChatGPT for Education,” a new newsletter run by their education team. It will start with profiling 30 educators who are effectively using AI in teaching.

📝 Register for February’s Webinar

Want to learn more about on using AI to generate course content?

If so, just answer “Yes” in the below poll and you are thereby registered for my February Zoom webinar. I think it will be on February 28th from noon to 1:30pm Eastern, although this isn’t set in stone…

If you select “Yes.”, that’s enough to register. You’ll get a follow-up email in a few weeks with more information.