Deep research does extended, multi-source research for you and returns a written report with citations. This article explains when to use it in University of Maine System (UMS) ChatGPT Edu, how to run it, and how to check what it finds.
You get 5 deep research tasks free each day. After that, each task uses 50 credits. The daily allowance runs on a rolling 24 hours from your first use. See ChatGPT - Plans & Usage Limits.
A regular ChatGPT answer takes seconds. Deep research takes several minutes, because it:
- Works out what information your question needs.
- Searches the web across many sources.
- Reads and compares what it finds, noting where sources agree, disagree, or are uncertain.
- Writes a structured report with citations linked to the original sources.
Think of the difference between asking a colleague a quick question and asking them to research a topic and come back with a written briefing. You can watch its progress while it works.
Good uses
- Literature reviews and research landscape summaries
- Policy comparisons across institutions or states
- Market or technology scans
- Benchmarking against peer institutions
- Background research on complex topics
- Overviews of emerging fields or trends
- Background for decisions
When something else works better
- Quick facts: Use Chat on Instant or web search. Both are free and faster.
- Your own or internal files: Deep research searches the public web. To work with files you have, upload them to a chat or Project.
- Live data: Such as stock prices or event results.
- Writing, brainstorming, or code: A regular chat is usually faster.
- Sources you already have: Upload them to a chat or Project and ask ChatGPT to work across them.
- Open a new chat, or a chat in a Project if you want the report kept with related work.
- Select Deep research from the tools menu.
- Type your research question.
- Answer any clarifying questions ChatGPT asks before it starts.
- Let it run. It usually takes several minutes.
When the report is ready, you can copy or export it, or keep going in the same chat: ask follow-up questions, dig into one finding, ask for a shorter summary, reformat it for a different audience, or turn it into a table or presentation outline.
A vague prompt gets a broad report. A specific, well-scoped prompt gets something you can use.
Be specific about the question. Instead of "Tell me about AI in education," try:
What are the most common approaches universities are using to integrate generative AI tools into undergraduate STEM courses, and what early outcomes have been reported?
Set the scope. Include limits such as geography, timeframe, institution type, or discipline:
Focus on U.S. public universities and research published since 2023.
Describe the output. Ask for the structure, comparison format, or level of detail you want:
Organize the findings by approach type and include a summary table comparing implementation models.
Explain your purpose, so it can prioritize what matters:
I am preparing a proposal for our provost to expand AI tool access on campus. Focus on evidence of impact, cost models, and implementation challenges.
Example prompts
Literature review: A faculty member starting a new project maps recent work in the field.
Summarize the key developments in federated learning applied to healthcare data over the past two years. Identify major research groups, commonly used datasets, and open challenges noted in the literature.
Curriculum research: A department compares peer institutions before changing a course sequence.
How have AAU member universities restructured their introductory computer science sequences in the last three years? Focus on approaches to integrating AI literacy and changes in programming language choices.
Policy benchmarking: An administrator compares published AI policies.
What AI acceptable use policies have been published by public R1 universities in the United States? Summarize common elements, notable differences, and policies that specifically address generative AI in coursework.
Technology scan: A department starts evaluating software.
Compare leading identity governance platforms, focusing on integration with Microsoft Entra ID, SCIM support, and suitability for higher education environments. Include pricing models where publicly available.
Thesis background: A graduate student maps the theories and studies on a topic.
What are the primary theoretical frameworks used in research on student retention in online STEM programs? Summarize the key models, their proponents, and recent empirical studies that have tested them.
Deep research reports include citations. Always check them before you rely on the report.
- Open the key sources. Make sure each one actually supports the claim it's cited for.
- Judge source quality. Citations can range from peer-reviewed journals and government reports to news articles, company websites, and blogs. Consider each source's authority, methods, and possible bias.
- Look for gaps. Deep research can miss paywalled papers, very recent publications, specialized sources, and anything that doesn't rank well in web search. For academic work, add library databases and other approved sources.
- Cross-check important claims with other sources or your own expertise.
- Refresh time-sensitive topics. A report reflects what was available when it ran.
Deep research supports your judgment. It doesn't replace it. It can misread a source, draw a wrong conclusion, or leave out context. Verify anything that will inform an important decision.
Questions about deep research?
Environment
- University of Maine System (UMS) ChatGPT Edu workspace
- Deep research
- Eligible UMS students, faculty, and staff
- Web browser at chatgpt.com, plus the ChatGPT desktop and mobile apps