A good research prompt is not a search query with extra words. It is a carefully considered set of instructions that tells an AI tool what you are trying to investigate, what context matters, and what kind of result would be useful.
That distinction matters for doctoral students and anyone doing substantial research. If you type a quick question into Gemini and accept the first answer, you may get something helpful. But if you spend time defining the task, supplying relevant material, and refining your instructions, you can ask much more of the tool.
I have spent half an hour or an hour developing a prompt for a project that might otherwise take several hours to write. In that situation, prompt development can account for 10 to 20 percent of the total work. That is not wasted time. It is the research equivalent of thinking through an outline before writing.
What You’ll Learn
- Write complex research prompts like plans, with clear context, scope, and instructions.
- Improve through frequent practice, a reusable prompt log, and examples from peers.
- Use AI to critique your work while retaining responsibility for the thinking and writing.
- Scope broad research first, then investigate promising organizations and verify important findings.
Table of Contents
- Think of a Prompt as a Research Plan
- Get Better by Doing a Lot of Research
- Learn from Other People’s Prompts
- Use Documents to Give Your Research Context
- Practice With Creative Research Formats
- Start Broad, Then Narrow Your Research
- Break Large Research Tasks Into Stages
- Build a Research Habit You Can Reuse
Think of a Prompt as a Research Plan
When I write a complex prompt, I am often doing two things at once: outlining the work and explaining how I want it carried out. A short search-style question rarely communicates the scope of a project. A stronger prompt can establish a role, identify the subject, describe the intended sections, and explain what each section should accomplish.
For example, some of my prompts ask Gemini to approach a topic as a consultant for the gospel rescue mission movement. They then give instructions for several parts of a report. These are not the kinds of things I would type into a web search. They are closer to a paper outline with directions attached.
Before submitting a research prompt, ask yourself:
- What question am I actually trying to answer?
- What context would someone need to understand the question?
- Which documents, organizations, or examples should inform the work?
- What sections or comparisons would make the result useful?
- What should I examine or verify myself afterward?
You do not need every prompt to be long. A quick question can be answered with a quick prompt. But when the research task is complicated, the instructions should reflect that complexity.
Get Better by Doing a Lot of Research
Prompt engineering is like riding a bike or learning a language. You can learn something by observing others, but you improve by practicing. During a doctoral program, I encourage students to produce hundreds of Deep Research reports over time. The point is not to accumulate reports for their own sake. It is to develop judgment about which instructions produce useful research and which ones need work.
Make AI part of your ordinary question-answering routine, too. I keep a Gemini tab open and regularly consider whether a question is better suited to Gemini or a web search. The more often you make that choice deliberately, the more familiar you become with what each approach can do.
This applies to technical questions as well as academic ones. If you run into a technology problem that you would normally take to a tech-savvy colleague, try asking Gemini how to do it. It can be especially useful for step-by-step help. You still need to determine whether the instructions fit your situation, but asking is often a productive first move.
Keep a Prompt Log
Save your substantial prompts in a Google Doc or another document you can search later. Complex prompts take time to develop, and parts of them may be useful again. A good instruction for comparing organizations, structuring a report, or analyzing a document does not have to be recreated from scratch every time.
Keep more than the final wording if it helps you learn. Note what you changed and whether the revision improved the result. Over time, your prompt log becomes a record of your own research methods, not just a collection of text to paste into Gemini.
Learn from Other People’s Prompts
Practice is the biggest teacher, but observation helps. When a Gemini Deep Research article is provided in a course, look at the prompt behind it. Ask why it identifies a particular audience, requests certain sections, or supplies specific context. Those decisions often explain why the resulting article fits the course rather than reading like a generic overview.
Peers are another source of ideas. When sharing a Deep Research report, include the prompt that produced it. A finished report shows what came out; the prompt shows how the researcher framed the problem. Comparing both gives everyone a better opportunity to learn.
Templates and prompt suggestions are useful starting points as well. Treat them as material to build on, not as formulas that must remain unchanged. The goal is to develop a prompt suited to your particular research question.
Use Documents to Give Your Research Context
One of the most useful changes you can make to a prompt is to provide the material you want analyzed. Uploading a document lets you ask questions about its arguments, organization, or contents instead of relying on a general description you type from memory.
Ask for Critique, Not a Substitute Author
For students, a particularly valuable use is asking Gemini to critique a paper you have written. You might ask where an argument has gaps, what objections you have not addressed, or what additional points could strengthen your case. You can then discuss the feedback and decide what, if anything, to change.
There is an important boundary here: you must remain the person doing the thinking and writing. Treat the tool more like a friend who reads a draft and offers comments. If that friend effectively writes the paper for you, the arrangement is no longer just feedback. The same principle applies to Gemini.
Its suggestions are not instructions you are obliged to follow. Evaluate them against your argument, your sources, and the requirements of your course. Critique can help you sharpen your reasoning; it should not replace that reasoning.
Choose the Right Source Material
For a focused task, uploading the relevant documents directly may be enough. If you have a larger body of related material, you can also consider using Google Drive or NotebookLM as a source. That can help when the research you need is spread across more documents than you want to attach individually.
I do not use that approach for every task. First consider which documents are actually relevant. Giving an AI tool access to a broad collection is not automatically better than supplying a small, well-chosen set of sources.
Use Templates for Document Work
Documents can serve as examples of an output format, not only as sources of facts. In course design, I have worked with a Canvas course export of about 100 pages that was difficult to read in its exported form. By providing a template, I could ask Gemini to reorganize that material into a format closer to how the course appears in Canvas.
I have also used audited financial statements to help identify what belonged on individual lines of a Form 990. Gemini did not fill in the form itself; it produced a line-by-line list for me to review. That distinction is crucial. A tax form is a legal document, and an AI tool can make up a number or put a real number in the wrong place. Check the figures and the source documents yourself before entering anything.
The same caution applies to research generally. A polished answer is not proof that every detail is correct.
Practice With Creative Research Formats
Research does not always need to end as a conventional written report. NotebookLM offers formats such as audio, video, mind maps, infographics, slides, and data tables. Try them and think about which format suits the material. I have been especially impressed by its audio and infographics, and its slides can also be useful.
There is a practical difference between slide options. NotebookLM slides are PDFs, while Gemini can be prompted to create Google Slides that you can edit yourself. If you expect to revise the presentation, that difference matters.
Gemini also has creative possibilities involving images, music, video, and coding. These may not be central to every doctoral research project, but familiarity with them gives you more ways to communicate ideas when a written report is not the best fit.
Start Broad, Then Narrow Your Research
One of the most effective research strategies is to begin with the big picture and progressively focus your questions. This is especially helpful when studying organizations and looking for best practices. Rather than picking a handful of familiar organizations immediately, you can first build a larger peer list and ask which organizations merit closer attention.
Here is the basic sequence:
- Build a peer organization list. Identify organizations relevant to the field you are studying.
- Run a broad scoping search. Use a criterion related to your research question to find promising candidates.
- Investigate a smaller group. Provide organization URLs and request a more detailed review of their work.
- Identify people to contact. Use the findings to guide further inquiry with experts.
For one rescue mission project, I began with a list of about 200 organizations. I wanted to find missions with strong social enterprises, so I asked which organizations had the highest revenue from earned income. That broad question helped narrow the field.
I then focused on 15 organizations and supplied their URLs. The next prompt asked for a review of their social enterprises, a summary of what they offered, and an identification of best practices. From there, the research could help identify experts worth contacting.
The method is more important than the particular numbers. A broad search helps you find candidates; a narrow search lets you ask better questions about them. It also makes the final report more purposeful than a general request to “find best practices.”
Make Your Criteria Explicit
When you build an organization list, state what makes an organization relevant. For the rescue mission example, earned income was a way to find organizations that might have developed substantial social enterprises. A different project would need a different criterion.
Supplying URLs also helps define the group you want examined. If you already have a peer list, use it rather than asking the tool to guess which organizations belong. The more clearly you define the pool and the question, the more useful the subsequent comparison can be.
Break Large Research Tasks Into Stages
Sometimes you want a very large document: a 40-page report, a 50-page report, or even something approaching a book. A single enormous prompt may not work well. The task can become too complex for the tool to execute, including when it runs into limits on how much text it can process or produce at once.
Instead, start with a general prompt that establishes the project, then add sections in follow-up requests. If your plan has six sections, you might begin with the first three. After that, request section four with its own details, then section five, then section six.
This staged approach gives each part enough attention. It also creates natural opportunities to inspect the work before continuing. If an early section heads in the wrong direction, you can correct it before extending the document.
For a student paper, remember the distinction between using this method to explore or organize research and having AI write the assignment for you. The staged technique can help you investigate a complicated topic, but responsibility for your own writing remains yours.
Build a Research Habit You Can Reuse
Improving at AI prompt engineering for research does not depend on finding one perfect phrase. It depends on repeatedly framing meaningful questions, supplying appropriate context, examining the results, and revising what you ask next.
Save prompts that work. Study the prompts behind useful reports. Ask for criticism of your own arguments without handing over authorship. Provide documents when the task calls for them, verify important claims and numbers, and narrow broad research into focused investigations.
Above all, keep practicing. Your prompts will become more precise as you gain experience, and the tools will continue to change. Developing the habit of thoughtful, hands-on research puts you in a better position to make use of both.