Can Colleges Detect AI-Written Essays?

The short answer: mostly no, and the schools that talk about it at all mostly say they aren’t trying. But the question is aimed slightly wrong. The real risk of an AI-written essay was never a detector flagging it. It’s what happens when a human reads your file, and what your signature on the application actually promised.

What detection can and can’t do

AI-text detectors judge statistical texture — how predictable your word choices look to a model. That texture overlaps badly with real students: polished human writing gets flagged, lightly edited AI writing sails through, and multilingual writers get flagged most of all. OpenAI retired its own AI-text classifier in 2023 for exactly this reason: it couldn’t tell reliably. No vendor since has published accuracy numbers good enough for a decision as consequential as an admissions offer, and admissions offices know it.

That’s why almost no college claims to screen application essays with a detector. In the thirty-plus published policies we track in the school AI-policy directory, exactly one school addresses detection head-on — by disavowing it.

What schools actually say

UW–Madison is the candid one: “We will not disqualify an applicant found to have used or suspected of using AI in their admissions essays, nor are we running essays through any system to detect if AI was used.” The same statement strongly discourages “simply feeding AI a prompt for their essay” — not because they’ll catch you, but because it wastes the one channel where they hear your voice.

Virginia Tech flips the question: it announced that its own office reviews each essay with one human reader and one AI reviewer, the model “designed to confirm, not replace, the human reader’s score.” AI on the reading side, nothing published about applicants’ own AI use. Most schools publish nothing in either direction — of the sixty-eight schools in the directory, roughly half are silent on AI entirely.

The risk is rescission, not detection

Here’s the part the “can they catch me” framing misses. When you submit the Common App, you sign an affirmation that your essay is your own work. Schools with published AI rules attach real consequences to breaking it: Villanova states that applicants “found to be submitting writing that is not their own may be denied or have their admission rescinded,” and Brown prohibits AI use “in conjunction with application content” outright. Rescission doesn’t require a detector. It requires a mismatch someone eventually notices — and offers can be withdrawn after you’ve enrolled.

And mismatches are what human readers are good at. Your file is full of writing samples that aren’t the essay: short answers, the activities list, graded work behind your transcript, sometimes an interview. A 780-word personal statement in a voice none of the rest of your file can produce doesn’t need software to look wrong. Admissions readers see thousands of seventeen-year-olds’ essays a season; the season’s ChatGPT house style is something they recognize the way you recognize a stock photo.

So what’s actually safe?

Write the essay yourself, and know your schools’ published lines before you use AI for anything around it. The lines differ more than most students expect: the University of Washington explicitly lists an “AI tool” alongside parents, counselors, and tutors as an acceptable reviewer of your writing, and Bowdoin permits AI to “review grammar or provide feedback after you have written a draft.” At the other end, Brown’s prohibition covers feedback too. The directory quotes each school’s own words with sources and dates, so you can check yours in a minute.

The durable rule, everywhere: the words you submit must be yours. Feedback that leaves you the author sits inside most published policies. Text you didn’t write sits outside all of them — whether or not anything ever “detects” it.