Do college admissions check for AI?
Mostly not with software — because the software doesn't work. A Stanford study found seven AI detectors misflagged 61.22% of human-written TOEFL essays. Admissions offices instead rely on the fraud attestation you sign, experienced readers noticing a mismatch with your other writing, and reports from counselors. Common App can terminate your account and notify every college on your list.
Why — the first-principles explanation
The honest answer starts with why detection is largely off the table. AI detectors work by measuring perplexity — how predictable each word is. AI text is made of statistically likely words, so it scores low. The problem is that plain, clear human writing also scores low, and that turns the tool into a bias machine. Stanford researchers ran seven detectors over 91 TOEFL essays written by real humans: 61.22% were misclassified as AI, 19% were unanimously flagged by all seven, and 97% were flagged by at least one — while the same detectors performed near-perfectly on essays by U.S.-born eighth-graders. The same study showed that asking a model to "elevate" its text with literary language defeated the detectors entirely.
Sit with what that means, because it's the whole ballgame: the detector fails the honest non-native student and passes the polished cheater. An admissions office deploying that tool would systematically punish international applicants for writing simply, while waving through anyone who told ChatGPT to use fancier words. Turnitin itself revised its false positive rate from 1% up to 4% after real deployment. No competent office wants that legal and reputational exposure over an essay.
So enforcement moved to mechanisms that actually work. Attestation is the big one: you certify the work is your own. That's not a formality — it's the thing that converts "suspicious essay" into fraud, which is a category with real teeth. Common App's fraud policy explicitly covers misrepresenting as your own work the substantive content or output of an AI platform, and on a fraud determination it can terminate your account and report you to every college on your list. The signature, not the scanner, is the risk.
Human readers are the second mechanism, and they're better than people assume — not because they can spot "AI writing," but because they can spot incoherence across your file. A reader sees your essay next to your graded work, your recommendation letters describing you, and your short answers. Generated prose tends to be fluent, generic, and strangely absent of you. When the voice in the essay doesn't match the person in the rest of the file, that's noticeable — and it's noticeable in a way that doesn't depend on any AI at all.
Third: reporting. Counselors, teachers, and rival applicants report. Common App investigates on report, not by scanning.
An example that makes it click
Imagine a bakery hiring a pastry chef and asking each applicant to bring in a cake.
They don't own a machine that detects store-bought cakes — and honestly, such a machine would be terrible. It'd flag every genuinely simple, well-made cake as suspicious while passing an elaborate supermarket one. So they don't bother.
Instead, three things happen. You sign a card swearing you baked it. A chef who's tasted ten thousand cakes looks at yours next to the messy one you made in the trial round last week and says, hm, these aren't the same hands. And sometimes another applicant mentions they saw you leaving the shop on Fifth Street. None of that is detection technology. All of it works better than the machine would — and the card you signed is what turns a good cake into a lie.
How to do it
- Read the Common App fraud policy at commonapp.org/fraud-policy before you submit — it's the document that defines AI misuse as fraud and sets out account termination and college notification.
- Check each school's specific rule. Yale permits AI grammar and spelling review while treating AI-generated substantive content as fraud; Brown permits no AI use in application content at all. These are genuinely different.
- Recognize the real check: your essay is read alongside your graded work, recommendations, and short answers. Consistency of voice across your file is what a reader actually evaluates.
- Write with specifics only you have — real names, exact moments, particular failures. Generic fluency is what reads as generated, and it's also what makes a weak essay regardless.
- Keep drafts and version history. Detectors are unreliable enough that a false accusation is a real possibility, and contemporaneous evidence of your process is your best rebuttal.
- If accused on the basis of a detector score, ask what the evidence is and note the Stanford finding — 61.22% of human TOEFL essays misflagged, 19% flagged by all seven detectors.
Key facts
- A Stanford-led study tested seven AI detectors on 91 human-written TOEFL essays: 61.22% were misclassified as AI-generated, 19% were unanimously flagged by all seven, and 97% were flagged by at least one.
- The same detectors performed near-perfectly on essays written by US-born eighth-graders, demonstrating systematic bias against non-native English writers.
- The Stanford study showed detectors could be bypassed by prompting a model to elevate text with literary language — meaning polished AI text passes while plain human text fails.
- Turnitin revised its stated false positive rate from 1% to 4% after real-world deployment; Turnitin is used by over 16,000 academic institutions.
- Common App's Fraud Policy, updated in 2024, defines fraud to include misrepresenting as one's own the substantive content or output of an artificial intelligence platform; a fraud determination can terminate the account and be reported to all colleges on the applicant's list.
- Common App investigates on report rather than by scanning submissions; enforcement relies on attestation, experienced readers, and third-party reports.
- Detectors measure perplexity — how predictable each word is — which is why non-native writers with lower lexical diversity are flagged at high rates.
▶ The 60-second explainer (script)
Do college admissions check for AI? Mostly not with software — and the reason is the interesting part. AI detectors work by measuring perplexity, which is how predictable each word is. AI text is made of likely words, so it scores low. The problem: plain, clear human writing also scores low. Which turns the detector into a bias machine. Stanford researchers ran seven detectors across ninety-one TOEFL essays written by actual humans. Sixty-one percent were flagged as AI. Nineteen percent were flagged by all seven. Meanwhile the same detectors were near-perfect on essays by American eighth-graders. And the same study found that asking a model to use fancier literary language defeated the detectors completely. Sit with that. The detector fails the honest non-native student and passes the polished cheater. Any admissions office deploying it would systematically punish international applicants for writing simply. Turnitin itself raised its false positive rate from one percent to four after real deployment. Nobody wants that exposure over an essay. So enforcement moved to things that actually work. First, attestation. You sign that the work is yours. That's not a formality — that signature is what converts a suspicious essay into fraud, which has real teeth. Common App can terminate your account and report you to every college on your list. The signature is the risk, not the scanner. Second, human readers — and they're better than you think. Not because they spot AI writing, but because they spot incoherence across your file. Your essay sits next to your graded work and your recommendation letters. When the voice doesn't match the person, that's visible without any technology at all. Third, reports. Counselors, teachers, rival applicants. Common App investigates when someone reports — not by scanning.
What authoritative sources say
People also ask
Do colleges run my essay through an AI detector?
Most don't, because the tools are unreliable. Stanford found seven detectors misflagged 61.22% of human-written TOEFL essays while clearing native-speaker eighth-graders almost perfectly. Deploying that against international applicants is a legal and reputational problem no office wants.
So how would they ever catch AI use?
Three ways: the attestation you sign, which makes misuse fraud rather than just suspicion; human readers noticing your essay's voice doesn't match your graded work and recommendation letters; and reports from counselors, teachers, or other applicants.
Why do detectors flag non-native speakers so heavily?
They measure perplexity — how surprising each word is. Non-native writers use more common vocabulary and simpler syntax, which produces low perplexity. That's mathematically the same signature the detector treats as AI.
What happens if Common App finds fraud?
It can terminate or suspend your account and report the finding to every college on your My Colleges list. One determination reaches all your schools at once — which is the closest thing to a blacklist that actually exists.
Can I use AI at all on my application?
Depends on the school and you must check. Yale explicitly permits using AI to review grammar or spelling while treating generated substantive content as fraud. Brown permits no AI use in application content whatsoever.
The same question, asked other ways
- Do colleges check for AI in application essays?1,600/mo