The Most Important Product of Research Is You

What does it mean to own your research?

In recent guidance on the use of AI in college applications, the University of Pennsylvania offered applicants a simple test: if you cannot explain in a conversation how you arrived at what you wrote, the work isn’t ready to submit. This same test applies to student research. A polished paper shows us your results, but ownership becomes clear in conversation. Can you reconstruct the reasoning behind your research methodology? Can you explain how your research question changed, why you chose one source or method over another, what went wrong, and how feedback affected your thinking?

Ownership does not mean completing a project without help. Research is rarely solitary. Researchers learn from teachers, mentors, librarians, peers, editors, and increasingly, AI tools. The important question isn’t whether you receive assistance on your work. It is whether that assistance supported your thinking or replaced it. A student who owns a project remains its intellectual decision-maker. They can evaluate suggestions rather than simply accept them. They can identify where an idea came from, determine whether evidence supports a claim, and take responsibility for what appears in the final paper. If an AI tool disappeared tomorrow, they might work more slowly, but they would still understand what they were doing.

What a defense reveals

The thesis defense is an important example of project ownership in the age of generative AI, when it is possible to outsource the thinking required to produce written work. At Harvard, I have participated in around a dozen senior thesis defenses, in which a student discusses their year-long research project with an examination committee. The process goes something like this: after a student submits their final paper, each member of the committee reads it independently and sends the student written comments. The student reviews these comments and prepares for the defense. They open the defense by briefly summarizing their thesis while responding to our comments, and then we, as the committee, ask questions.

Because none of us advised the student, they cannot rely on our knowing what they intended to do or how the project developed, only our read of their final product. Despite best intentions, that submission might not always be polished, but our aim in the oral exam is not to critique their editing choices but to have an engaging conversation about their work. Because committee members bring different interests and expertise, our questions may address the literature, theory, research design, analysis, or conclusions, and may do so at different levels of depth. I believe the most revealing questions are often quite simple: Why did you choose this method? How did you arrive at these categories? What else could explain this finding? What can your evidence not tell us? What would you change if you conducted the study again? Questions like these move beyond “What did you find?” They ask students to consider how the design of a study shapes its findings, whether another explanation could fit the evidence, and where the boundaries of an argument lie.

Importantly, defending your research does not mean defending every choice as correct. Some of the strongest students recognize a valid criticism, explain the constraint or reasoning that led to their original decision, and describe what they would now change. They distinguish between what their evidence demonstrates and what they merely suspect. When they do not know an answer, they reason from what they do know rather than bluffing.

A research paper can have serious methodological flaws and still reflect genuine learning. Students often encounter real-world data access issues, fixed project deadlines, and mistakes that come along with being a first-time researcher. What matters is whether they understand the consequences of these limitations and can explain what they would change. As a mentor with Polygence, I help students understand why they might make certain methodological decisions and what those decisions allow them to claim. But the student remains responsible for owning their work, just as college students are responsible for owning their thesis projects.

When AI replaces the learning

I saw the importance of academic ownership while mentoring a student through an original empirical project. The student conducted interviews and set out to transcribe the interviews, code them for themes, then connect those findings to quantitative measures.

This kind of work takes time. Each week I assigned tasks meant to move the project forward and build the research skills needed for a methodologically sound paper. For several weeks, the assignment submissions didn’t arrive, so we talked through their work instead. Throughout our work together, I asked to see the underlying materials: the codebook, examples of coded transcripts, the dataset, and the file used to calculate the results, but the student did not provide them. I asked them to update the way their quantitative measure was defined, and the reported statistical relationship remained exactly the same, which didn’t make sense. They couldn’t explain what they had done or how they had conducted the test.

I offered to review the materials myself, thinking I could help them work through the process, but instead of materials I received a polished description of a research process that explained what the codebook contained, how they had moved from codes to themes, and how they had constructed the quantitative measure. On the surface, the response sounded methodologically sophisticated, but the polished account did not make the research more defensible. They had begun describing the research rather than carrying it out. Ultimately, they did not complete the project.

This experience changed how I think about the risks of AI in student research, or for anyone when learning a new skill. The most obvious risk is that AI may produce inaccurate information. The more concerning risk is that it can generate a convincing account of work a student does not understand—or worse, may not have completed. It can name plausible codes, describe standard methodological steps, and produce statistics that look plausible to the untrained eye. But when you cannot show how the evidence was collected, reproduce the analysis, or explain why the decisions make sense, the appearance of research begins to replace the research itself.

Ownership requires more than being able to repeat a polished explanation. A researcher should be able to show their process: notes, sources, drafts, coding decisions, data, calculations, and revisions. They should be able to trace a finding backward from the final sentence to the evidence it came from. When something does not make sense, they should be able to reopen the analysis and figure out why.

This figuring-it-out part can be hard, and we should normalize that strong research doesn’t have to be perfect. The figuring-it-out part is where the real work happens. Early drafts may be rough. The methods section may need clarification. The arguments may change as learning happens. Sometimes, a plan for data collection needs to be abandoned or modified due to unanticipated issues. But, in making these changes, you develop a deeper understanding of both your topic and the research process. Your papers become stronger, and you become a better scholar. The difficulty is not an unfortunate obstacle standing between you and your final paper. Much of the time, it is where the learning occurs.

You, as the student researcher, are also a product of the process.

Research programs often emphasize tangible outcomes: a paper, presentation, poster, or publication. Those outcomes matter. But they are not the only products of a research experience. Your goal as a researcher is not only to produce something that looks like research, but to become someone who can ask a meaningful question, make informed methodological decisions, evaluate evidence, recognize uncertainty, respond to criticism, and revise an idea. These goals apply whether you write an empirical paper or use existing research as your data when you write a literature review. These capacities develop through the slow and often frustrating work of doing the work, not just producing a polished final product.

AI can help you navigate the work: it can serve as a sounding board, help clarify a confusing concept, or assist with a task whose substance you already understand. If, however, you use AI to take over the decisions and reasoning that the project was meant to teach, you will produce a final product that might look good without developing skills that allow you to produce such work without assistance in the future.

Said another way, when you use AI to produce that final product, you hand your thinking over to a tool that's just predicting the next word. Your final paper may improve while your understanding does not. A student who does the thinking themselves may not produce a polished first draft but develops the thinking skills to produce a final product that is far more nuanced than what a generative AI model will write for them -- work they will be able to defend in conversation.

Put your project through the thesis defense test

Before submitting your research, ask yourself whether you could sit down with several curious readers and discuss it without hiding behind the words on the page. You do not need a formal committee to determine whether you own your research. Ask a mentor, teacher, parent, or friend to interview you about it. Do not give them a presentation to follow. Let them ask questions, including questions you have not prepared for.

At a minimum, you should be able to explain:

1. Why did you choose this question?
   What interested you initially, and how did the question change as you learned more?

2. How did you make your most important research decisions?
   Why did you choose these sources, cases, participants, measures, or methods? What alternatives did you consider?

3. How did you get from the evidence to your conclusions?
   What patterns did you identify, and why do you interpret them as you do?

4. What else could explain what you found?
   How have you evaluated competing explanations, potential biases, or the effects of your own research design?

5. Where does your project fall short?
   What can you not conclude from this evidence? What would you change or study next?

6. What assistance did you receive?
   Where did mentors, peers, or AI tools contribute, and how did you evaluate and incorporate that assistance?

If you struggle to answer one of these questions, you have identified where you need to return to the work. Reread the source. Revisit the data. Ask why you accepted a suggestion. Revise the section until its logic is genuinely yours.