Capstone and Thesis Defenses: Answering AI-Use Questions From the Panel

When a thesis or capstone panel asks about AI, treat it as an ordinary scholarly question about your process — because that’s what it usually is — and answer specifically: what you used, how you verified it, and confirmation that the ideas and analysis are yours. A defense exists to establish that you understand and own your research. Questions about tools are just part of that, no more hostile than a question about your sample size. The students who handle them worst are the ones who hear an accusation where none was meant and get vague or defensive.

Whether you used AI heavily, lightly, or not at all, a little preparation turns this from a tense moment into a routine exchange. Here’s what the panel is really after and how to answer it cleanly.

Key takeaways

• Panels ask about AI to confirm the work and thinking are yours and met program standards — usually not to trap you.
• If you used AI, be specific: what for, how you verified it, and that the ideas are yours.
• If you didn’t, say so plainly and let your command of the work prove it.
• A detector flag is a probability, not a verdict; live command and process evidence outweigh it.
• Prepare three things: your program’s policy, an honest account of any AI use, and your process trail.

Thesis defense AI-use questions: what the panel is asking

It helps to hear the question behind the question. A committee’s core job at a defense is to verify authorship and understanding — that you did this research, you understand it deeply, and it’s genuinely your contribution. AI-use questions are a modern extension of that same job. They’re checking that any tools you used were within bounds and didn’t do the intellectual work for you.

There’s often an institutional layer too. Many programs and journals now expect authors to disclose AI assistance in their methods and explicitly forbid crediting AI as an author. So a panel asking “did you use AI, and how?” may simply be following the same disclosure norms your eventual publication will require. Reframing it this way — as scholarship’s version of a routine methods question — takes the sting out and lets you answer like a professional. For the disclosure conventions specifically, see how grad students disclose AI in a thesis or dissertation.

If you used AI: be specific and calm

Legitimate, disclosed assistance is not a problem at most institutions — undisclosed or misunderstood assistance is. So the winning move is precise honesty. Name the use, the verification, and the ownership:

• *”I used an LLM to draft boilerplate data-cleaning code, which I tested against manual calculations before trusting it.”*
• *”I used it to tighten the grammar of sections I’d written; the arguments and wording are mine.”*
• *”I used it to help me brainstorm alternative framings for the literature review, then developed and wrote the review myself.”*

Each of those does three things: it states what the tool did, shows you stayed in control by verifying, and confirms the thinking is yours. That’s far more convincing than a vague “a little, here and there,” which invites follow-ups and reads as evasive. Crucially, match your account to what your program actually permits — don’t inflate your assistance out of nervous honesty, and never downplay something you’ll have to walk back. If you’re unsure whether a specific use was within policy, it’s better to have clarified that before the defense than to discover the boundary in the room.

If you didn’t use AI: let the work speak

If you didn’t use AI-generated text, say it plainly — “I wrote and analyzed everything myself” is a complete answer — and then let the defense do what it’s designed to do. A defense naturally reveals authorship, because you can explain your choices, your data, your reasoning, and your dead-ends in a depth no one who didn’t do the work could fake. Fluent, detailed command of your own thesis is the strongest possible evidence, and it emerges automatically when you know your material.

If a detector flagged part of your writing and it comes up, don’t panic or over-argue. Note calmly that detectors produce false positives — OpenAI even retired its own classifier in 2023 for low accuracy — and offer your draft history and notes. Then return to demonstrating that you understand every page. A committee weighs your live command far above any software score.

When a flag surfaces in the room

Sometimes a detector result is on the table before you walk in. Handle it the way you’d handle a challenge to any methodological choice: acknowledge, explain, evidence.

• Acknowledge the flag without conceding guilt: “I understand a detection tool flagged chapter three.”
• Explain your genuine process: how the chapter was written, over what timeline, and why the writing reads as it does.
• Evidence it: point to version history, notes, code repositories, and data that show the work developing.

Detectors output probabilities, not proof, and they’re least reliable on exactly the formal, polished prose a thesis is written in. A committee knows this. What persuades them is your demonstrated authorship, backed by a visible trail — not a debate about the software. If the situation has escalated beyond the defense itself, understanding what an academic integrity hearing and due process look like is worth doing in advance.

How to prepare so it’s a non-issue

Walk in ready and the whole topic becomes routine:

1. Know your program’s AI policy and be able to state, in a sentence, how your work complied with it.
2. Write out an honest, specific account of any AI assistance and how you verified each use. Practice saying it aloud until it sounds calm, not rehearsed-defensive.
3. Assemble your process evidence — draft history, dated notes, code, data — so you can point to a trail rather than just assert one.

It’s also worth reading your chapters with fresh eyes before you submit, especially any sections you edited heavily. You can check how a chapter reads before you submit to see whether over-editing left it sounding machine-like, and if a genuinely-yours section reads flat, reword it in your voice so the prose matches the command you’ll show in the room. Do that groundwork and the panel’s AI questions land as easy pitches. For more on defending your work with confidence, keep reading the blog.

Frequently asked questions

Why is my thesis committee asking about AI use?

Usually not to trap you, but to confirm the work and the thinking are genuinely yours and that any AI use met your program’s standards. A defense exists to verify that you understand and own your research, so questions about tools are part of establishing authorship. They may also be following institutional or journal norms that now expect authors to disclose AI use in methods. The best mindset is to treat these as ordinary scholarly questions about your process, not accusations, and answer them the same way you’d answer a question about your statistical methods.

How should I answer if I did use AI in my thesis?

Be specific and matter-of-fact: name what you used it for, how you verified the output, and confirm the ideas and analysis are yours. For example, ‘I used an LLM to draft boilerplate code for data cleaning, which I tested against manual checks,’ or ‘I used it to improve the grammar of sections I wrote.’ Honest, precise disclosure of legitimate assistance is far stronger than vagueness or denial, because it shows you understood what the tool did and stayed in control. Match your answer to what your program actually permits, and don’t overstate or understate.

What if I didn’t use AI at all — how do I handle the question?

Say so plainly and be ready to demonstrate your command of the work. ‘I didn’t use AI-generated text; I wrote and analyzed everything myself’ is a complete answer. Then let your knowledge do the rest: a defense naturally reveals authorship because you can explain your choices, your data, and your reasoning in depth. If a detector flagged part of your writing, calmly note that detectors produce false positives and offer your draft history. Deep, fluent command of your own thesis is the most convincing evidence there is.

Can an AI detector flag get my thesis rejected at the defense?

A flag alone shouldn’t decide a defense, because detectors output probabilities and produce false positives, especially on formal academic prose. What actually persuades a committee is whether you can demonstrate authorship and understanding in the room. If a flag comes up, treat it as a question to answer rather than a verdict: acknowledge it, explain your genuine process, and present draft history and notes. Committees weigh your live command of the material and your documented process far more heavily than a software score, which is exactly why preparation matters more than the detector.

How do I prepare for AI-use questions before my defense?

Do three things. First, know your program’s AI policy and be able to state how your work complied with it. Second, prepare an honest, specific account of any AI assistance you used and how you verified it. Third, assemble your process evidence — draft history, notes, code repositories, data — so you can point to a visible trail. Rehearse saying your disclosure out loud so it sounds calm and confident rather than defensive. Walking in with a clear, practiced answer turns a potentially tense moment into a routine part of the conversation.