Yes, you can talk about a project you built with AI in a grad school interview, as long as you can explain every decision in it yourself. Faculty interviewers aren't checking whether you typed each line. They're checking whether you can define a problem, make trade-offs, test your own work and say honestly what didn't work. Be plain about what AI did and specific about what you decided.
Can I talk about a project I built with AI in a grad school interview?
Yes. More students now arrive at interviews with something they built using AI coding tools: a small web app, a data dashboard, a tool for a lab or a club. The project counts. What decides whether it helps you is not the tool, but whether you can stand behind every part of it when a professor starts asking questions.
A grad school interview, especially for a research-based program, is not a job screen. The faculty member across the table is asking a quiet question the whole time: would this person be good to work with on an unknown problem for several years? A project helps you answer that if it shows how you think. It hurts you if it shows you can't explain your own work.
What are faculty actually checking when you mention a project?
Most professors will spend under a minute on the product itself. Then they go for the reasoning. Expect questions like these:
- Why this problem? Who had it?
- What did you try first, and why did you change it?
- How do you know it works? What did you measure?
- What's the weakest part of it right now?
- What would you do with another three months?
- Which part did AI write, and how did you check it?
None of these are about syntax. They're about judgment, testing and honesty, which are the same things a research advisor needs from a student. That's why an AI-built project can be a strong topic: it often gives you more decisions to talk about, because you moved fast enough to try several things.
The University of Pittsburgh's career center makes the same point in its guide to talking about AI in interviews: interviews are a place to show critical thinking, not technical expertise. Its suggested structure is context, the type of tool, your judgment in checking the output, and the outcome. It also suggests practicing one sentence aloud: "I used AI to ___, but I decided ___, and the outcome was ___."
How should you describe what AI did?
Plainly, early and in proportion. Three common mistakes:
- Hiding it. If a professor asks "how did you build the backend?" and the honest answer is "an AI assistant generated most of it", saying anything else is a risk you don't need. Interviewers who build things can usually tell.
- Over-apologizing. "I didn't really do anything, the AI did it all" is rarely true and sells you short. Choosing the problem, deciding what to build, testing it with people and fixing what broke were your work.
- Listing tools instead of decisions. Naming three AI products tells the interviewer nothing. One decision you made and why tells them a lot.
A good line sounds like: "I used an AI coding assistant to write most of the front end and the database code. I designed what it should do, tested it with five classmates, and rewrote the part that handled uploads after it broke on large files."
A 2-minute answer structure
When an interviewer says "tell me about this project," use this:
| Part | Time | What to say |
|---|---|---|
| Problem | 20 s | Who had the problem, and how you know it was real |
| What you built | 15 s | One sentence; offer the link if they want to see it |
| How AI was used | 15 s | What AI generated or drafted, what you decided |
| One hard decision | 30 s | A trade-off you made and why |
| Evidence | 20 s | How many people used it, what they did, what you measured |
| What failed | 20 s | The weakest part, honestly |
Then stop. Let them pick the thread they want to pull. A two-minute answer that ends on "what failed" invites the best follow-up question you could get.
How do you connect a product to research?
If you're applying to a research program, the interviewer will want to know whether you can do research, not just build. An AI-built product isn't research by itself, so don't call it that. But you can show the research habits it used:
- A question. "I noticed people stopped using it after day two. I wanted to know why."
- A measurement. "I logged which feature people opened first and compared it with what they said in interviews."
- A careful claim. "With eight users I can't say much, but the pattern was consistent enough to change the design."
- A next study. "If I had more time, I'd test whether the reminder or the layout made the difference."
That last point often matters most. A student who can turn a small product into a sharp question is showing exactly what a lab wants.
A checklist for the week before the interview
- Open the live link on a phone and a laptop. Make sure it works.
- Read through the main parts of the code until you can explain what each one does in plain language.
- Pick the one decision you're proudest of and the one thing that failed. Practice both out loud.
- Write down the real usage numbers and where they come from. Don't round up.
- Prepare one sentence on how you used AI and how you checked its output.
- Ask a friend to play professor and push on the weakest part for five minutes.
If you can do all six, the project will help you. If you can't explain a core part, spend the time understanding it, or leave the project out. A project you can't explain is worse than no project.
How STARC approaches this
STARC is a small nonprofit, and its one-person company cohort is built around the same idea: what counts is a working product plus an honest record of how it was built. Students use AI to build a small product over eight weeks, find real users, and write a build log every week covering what shipped, the numbers, what users said, where they got stuck, and how they used AI. Numbers in the logs need evidence.
Those logs are, in effect, interview preparation: a written record of decisions, failures and fixes. Logs are private to the cohort by default and published only if the student agrees. The first cohort hasn't run yet, so there are no past results to show. The rules, including what STARC won't do, are on the How it works page, and the research track covers real research with an active researcher.
The short version
Talk about the project. Say what AI did. Spend most of your time on the problem, your decisions, your evidence and what failed. If you can answer "how do you know it works?" and "which part did AI write?" without flinching, an AI-built project can be one of the strongest things you bring into the room.
Related questions
Should I tell the interviewer I used AI to build it?
Yes, and say it early and plainly. If it comes out later under questioning, it looks like you hid it. One sentence is enough: which parts AI wrote or drafted, and which decisions and checks were yours.
What if the professor asks about code I didn't write myself?
Explain what that part does, why it's there and how you checked that it works. If you truly don't know, say so and say how you would find out. Pretending is the only answer that costs you.
Does an AI-built product count as research experience?
Not by itself. It shows you can build and test something, which is useful, but it isn't research. You can connect it to research honestly by talking about the question it raised, how you measured something, or what you would study next.
Is a small project with few users worth mentioning?
Yes, if it's real and you can explain it. Five people who actually used it, and what they did, is a better story than a polished demo nobody touched.
Want to build your own product?
The OPC cohort: 12 students, 8 weeks, one real product with real users.
Last updated: 2026-10-10. Written by the STARC team. Rules quoted here come from our transparency page; if they change, this post is updated.
More questions
Do students from less well-known universities stand a chance in a blind-reviewed program?
What a blind review removes, what it can't, and how a student from a less well-known university can build an application that wins on the material itself.
What's the difference between a student one-person company program and an AI side hustle course?
A one-person company program has you ship a product to real users; an AI side hustle course sells you lessons. How to tell them apart before you pay.
Can I build a real product with AI if I can't code?
Yes, if you pick a small problem and learn by shipping. What AI handles, what it doesn't, and a first-week plan for non-programmers.