Do Colleges Check Application Essays for AI? What Admissions Offices Put in Writing

We read fifteen official application documents end to end. Across the fourteen we could load live, not one names a detection product, and not one says a detector is pointed at your essay. The fifteenth is BYU's essay page, which says it uses software tools to analyse the admission essays of all students and may rescind an offer where an essay is found to have been generated by AI. Both of those are accurate. Which one governs you depends entirely on where you are applying, and that is why the answers online contradict each other.

That first sentence is narrower than it may look, and the narrowness is the whole point. Checking does happen. UC's own page says `UC runs plagiarism checks on applications`, which is software, pointed at what you submitted. What is missing from all fourteen is different and more specific: a named detector, or any claim that one is run on your essay.

Two other things follow, and they pull in opposite directions. A high score from GPTZero or ZeroGPT is not something any of these documents attaches a consequence to; none gives an AI-detection percentage or a threshold for an AI score. And the absence of a detector is not permission: Brown refuses AI-written content under any circumstances, in one sentence, with no scanning involved.

So the useful question is not "do colleges check". It is "what did my schools actually write down". Below is the wording, school by school, so you can check yours instead of averaging Reddit.

One quick redirect first. If you are a current student wondering whether your coursework gets scanned, you want Do colleges use AI detectors? It depends on what they bought, which is about Turnitin licensing inside a university. This page is about admissions, which runs on a different system, different paperwork and different published policies.

The Common App fraud policy is the document that binds every applicant who uses it. Its date line reads `Date of Last Update: August 2023`. Under the heading `Fraud`, the third bullet names AI directly:

Submitting plagiarized essays or other written or oral material, or intentionally misrepresenting as one’s own original work: (1) another person’s thoughts, language, ideas, expressions, or experiences or (2) the substantive content or output of an artificial intelligence platform, technology, or algorithm;

Read the verbs. Fraud here is defined as intentionally misrepresenting someone else's, or something else's, substantive content as your own. It is a statement about what you did, not about how a paragraph reads to a classifier.

The certification is the other half. Brown reprints it on its own integrity page, and Johns Hopkins reprints the same paragraph word for word on its admissions FAQ, which is how we know this is the applicant-facing wording rather than someone's paraphrase of it:

I certify that all information submitted in the admission process — including this application and any other supporting materials — is my own work, factually true, and honestly presented, and that these documents will become the property of the institution to which I am applying and will not be returned to me.

And Common App's applicant help center has a one-paragraph article called "Personal Essay". This is the whole article, last modified in September 2023:

Your personal essay must be your own work. Do not use another writer's work, and do not use Artificial Intelligence software (Chat GPT, Bard, etc.) to write your essay.

Now the part almost nobody quotes. The same fraud policy also says what Common App will not do:

Common App also will not independently judge, evaluate, conduct any verification of, or perform an investigation on any information provided on or about a student-applicant and/or application from a Reporting Member, except to the extent required to address misuse of the Common App Solution consistent with this Policy and our Terms of Use.

The sentence straight after it hands the decision to the college. The sentence straight before it leaves colleges free to run their own fraud checks without reporting what they turn up. We walked through that pair, and the procedure attached to it, in personal statement flagged as AI, so it is not repeated here.

So Common App writes the definition and runs a complaints process. It does not position itself as the thing that checks. Its Terms of Use, 44,730 characters last updated on 24 October 2025, do not contain the phrase "artificial intelligence" at all, and do not contain the standalone token "AI" either. (`fraud` appears in the same text, which is how we know the counter was reading the page.)

We read fourteen documents end to end on 15 September 2026, logged out, in a browser rather than through a search engine. Six sit on Common App's own domains: its fraud policy, its terms of use, its one-paragraph "Personal Essay" help article, and three of the AI-guidance articles published in the same help center (two of which carry a college's name, NYU and Babson). Two are University of California documents. The remaining six are the published admissions AI policies of Caltech, Brown, Cornell (undergraduate), Cornell (Graduate School), Georgia Tech and Johns Hopkins. Together that is 166,734 characters, measured with every collapsible panel counted, which matters more than it sounds like it should.

The last row is the control. The counter found 112 AI mentions in the same texts, so it was not simply failing to read the pages.

One process note that is worth passing on, because it would have made this table wrong. Two of the fourteen keep their AI policy inside a collapsed accordion: Georgia Tech's Statement on AI, and the Johns Hopkins entry titled "What is your policy on the use of AI in an application?". Read the way a browser shows them, Johns Hopkins' page is 3,160 characters and contains no AI policy at all. Read with the panels counted, it is 38,406 characters and contains the clearest permission rule in this set. If you go looking for your own school's policy and a search on the page turns up nothing, open the accordions before you conclude there is nothing there.

All four occurrences of `detection` are accounted for, and none of them is an office announcing that it scans essays:

1. Common App's fraud policy, describing the colleges' side: `members may do so consistent with the member’s own internal fraud detection protocols.`
2. NYU: `NYU does not use AI detection technology as part of the review process.`
3. and 4. Cornell's Graduate School, in a sentence that says it does not rely on such software.

Be careful with what this does and does not establish. It is a fact about these fourteen documents, not about all of American admissions. A school that says nothing might be doing nothing, or might be doing something it has not described. And the four occurrences do not all point the same way: three are an office explaining that it does not use or does not rely on detection software, but the first one runs the other direction. Common App's line assumes member colleges have internal fraud detection of some kind and simply declines to require them to report every case. It takes college-side checking for granted rather than disclaiming it, which puts it alongside UC's page rather than alongside the other three.

The narrow reading that survives all of that: among the offices that published an AI policy at all, the published part is about conduct, and no document in this set names a detection product or says it runs one on your essay. Which is what makes the page in the next section matter.

Everything above is about fourteen documents we could load. There is a fifteenth, and it breaks the pattern hard enough that the pattern is not the story.

BYU's Essays & Activity page tells applicants this:

NOTE: We use software tools to analyze the admission essays and activity responses of all students. We may rescind the admission offer of any student whose essay is found to have been generated by AI or otherwise plagiarized.

On the same page, the commitment applicants agree to:

You may not pay anyone to help you with your application, and you may not use generative AI tools (like ChatGPT) as you compose your responses.

Read what that first quote does and does not say. It says software, over the essays of all students, not a sample. It names AI-generated writing as a ground for rescinding an offer. It does not name a product, does not call the software an AI detector, and does not mention a percentage or a threshold anywhere. "Software tools to analyze" is as specific as it gets. So the distinction the rest of this page has been drawing still holds, and it holds even here: what triggers the consequence is the essay being found to have been generated by AI, not a number.

Three things about how we got this, because they change what the count above is worth.

It is an archived copy, not the live page. `enrollment.byu.edu` has returned a CloudFront block to us on every attempt, seven of them, across two BYU hostnames and four different paths. The text above comes from the Internet Archive's snapshot of 1 September 2026, at this permalink. You can open it and read the whole page. We cannot confirm what the live page says today, and if you are applying to BYU you should read it on BYU's own site rather than trusting a two-week-old snapshot or us.

It is therefore not in the count. The table above is fourteen documents read live on one day with one method, which is what makes it something you can redo. Folding in a snapshot taken on a different date through a different route would quietly break that. So the numbers stand at fourteen, and this page sits outside them, and the honest summary is: across the fourteen we could load, no document names a detection product or says one is run on your essay; the one we had to reach through an archive says software runs over every essay.

Worth saying what the exclusion costs, because the answer is nothing. We counted the archived page too: `Turnitin` 0, `GPTZero` 0, `detector` 0, `detect` 0. Folding BYU into the table would not move a single figure in it; it would only add 7,686 characters and two more mentions of AI. The exclusion buys back a count you can redo in an afternoon and pays nothing in evidence for it.

And it was hiding in a collapsed panel too. This is the third in the same corpus, after Georgia Tech and Johns Hopkins. Rendered the way a browser shows it, that BYU page is 1,895 characters and the NOTE is not in it at all; the string `NOTE: We use software tools` returns nothing, and neither does `AI`. With the panels counted it is 7,686 characters and the NOTE sits at character 3,421. (Both figures are whitespace-normalised and both exclude the Internet Archive's own banner, so they are the same ruler.) A reader who opened that page, skimmed it, and used the browser's find function on "AI" would have come away believing BYU says nothing about AI at all.

That last point is the one to carry away from this whole article. Three of the sharpest policies we found were invisible on page load.

Policy pages are not the only place an admissions office speaks. Deans post to admissions blogs, universities put out newsroom pieces, and a lot of what applicants repeat to each other traces back to one of those rather than to a policy. Georgia Tech's guidance started life as a post on its admission blog by Rick Clark, dated 27 July 2023, and most of that paragraph is still sitting on the application review page today, lightly reworded.

So we swept the blogs and newsrooms too, using site-scoped searches on 15 September 2026. Three specific schools get named in this rumour constantly, so they got checked by name:

• UCLA. A site-scoped search of `admission.ucla.edu` for "artificial intelligence" or "AI" returns three documents, and all three are marketing PDFs: two viewbooks, both naming an alumna as an artificial intelligence pioneer, and a careers list naming "AI machine learning specialist" as a job outcome. No AI policy for applicants on that site. The UCLA newsroom returns one item, a media round-up linking an outside newspaper story, not a UCLA statement.
• UC San Diego. Same search on `admissions.ucsd.edu` returns course-preparation and department pages where "artificial intelligence" is the name of a research area. `today.ucsd.edu` returns one item, a 2023 story about graduate-school mentoring, and nothing on applications and AI.
• Johns Hopkins. Its news site returns nothing for AI detection and admissions. Its actual policy is the accordion entry on the admissions site, quoted further down this page, not anything in the newsroom.

We did this specifically because a claim circulates that those three schools switched AI detectors off in admissions after measuring the error rate. We could not find that anywhere on their own sites, in policy or in news. What UCLA and UC San Diego do have is the UC Statement of Application Integrity, which is systemwide and which we quote in full two sections down. What Johns Hopkins has is its own permission rule, which does not mention detection either way.

A fourth school was in this sweep and is worth naming separately. `news.byu.edu` returns two items, from 2014 and 2006, neither about AI. BYU's policy is not in its newsroom; it is on the admissions page quoted earlier, which we could only reach through an archive. If a school's live site blocks you, the Internet Archive is a page of that site, with a date on it, and it is worth trying before you conclude a school has said nothing.

Two honest caveats. A site-scoped search is only as good as the index behind it, and an office can change what it does without publishing anything. Also: none of the fifteen documents carries its policy in an image, chart or flowchart. We checked, because a policy that lives only in a diagram is invisible to any text search, including ours. The one picture on the Johns Hopkins page is a photograph of the campus.

NYU publishes an article for applicants inside Common App's help center, last modified 24 July 2026. It opens:

NYU does not use AI detection technology as part of the review process. We trust that ­applicants are submitting their original work and expect prospective students to uphold the same practices and standards as all NYU students when incorporating AI in their academic work.

Further down the same article: `While applicants may utilize AI as a tool during the college application process, the final submission must be their own authentic creation.`

The longer statement is Cornell's, and it is worth reading twice. This is the Graduate School, updated 1 September 2026, so it does not govern first-year undergraduate applications:

Applications are always subject to an integrity review, which may include assessment of voice and consistency across materials, and verification through interviews or short writing exercises. Cornell does not rely on automated AI-detection software as primary evidence of a policy violation, as such tools can produce inequitable false positives, particularly for non-native English speakers and other writers whose style differs from the norms built into detection software. Concerns about AI use are evaluated through holistic, human-centered review, not algorithmic screening alone.

Two things in there matter. The first is the stated reason. Cornell does not say detectors are useless; it says the errors land unevenly, and it names the group they land on, non-native English speakers. We went through the published evaluations behind that in why non-native English writers get flagged by AI detectors more often.

The second is the replacement. Cornell is not saying it will take your word for it. It names what it does instead: consistency across your materials, and verification through interviews or short writing exercises. Preparing for that looks nothing like preparing for a percentage.

UC is not on the Common App, and its wording is looser than most applicants expect.

From the Statement of Application Integrity, last updated August 2023, under `Personal Insight Question Responses`:

Students may receive advice on content and editing, including the use of generative artificial intelligence software to assist with readability, but content and final written text must be their own.

The next sentence is about plagiarism, not AI: `Students must not plagiarize or copy from sources in print or online nor use any other person’s words or published works.` And then the operative one:

UC conducts regular screenings to verify the integrity of the responses, may request authentication of the content or writing as the student’s, and will take action when it is determined that the integrity of the response is compromised.

"May request authentication of the content or writing as the student's." Two documents in this set reserve the right to make you answer for the writing yourself, and this is the blunter of the two. The other is Cornell's Graduate School, quoted above, which names `verification through interviews or short writing exercises`. Same move, different words. This is also the line that tells you what to keep.

A practical note, and the reason people miss this rule: it is not on the page where UC explains the Personal Insight Questions. That page runs 12,580 characters and contains zero mentions of AI. The Statement of Integrity is a one-page PDF linked from the bottom of the UC application sign-in page. If you are writing PIQs this month, open that PDF and read the paragraph headed "Personal Insight Question Responses" before you do anything else. It is four sentences long.

Separately, UC's Office of the President published an interview with two campus admissions directors in October 2024 under the heading `Does UC screen for AI in applications or use it to review them?`. The answer runs three paragraphs and ends:

And, while using AI as a tool is one thing, using a completely AI-generated answer is another — and one that is equivalent to academic dishonesty. UC runs plagiarism checks on applications, and if your PIQs are found to have been generated by AI with unattributed sources, you could be disqualified from UC admission entirely.

Plagiarism checks, not AI detection. The word "detect" does not appear anywhere on that page. That is a news article rather than policy, so treat the PDF as the binding text, but the distinction it draws is consistent with the PDF: a similarity check against sources is a different mechanism from a classifier guessing at authorship. If you have been comparing numbers from different tools and wondering why they disagree, GPTZero vs Turnitin covers what each kind of tool is actually measuring.

This is the part that makes the internet look contradictory. The offices are not disagreeing about detection. They are disagreeing about what counts as help.

Brown uses the most absolute wording in this set:

Brown affirms this definition and reiterates our own independent policy that the use of artificial intelligence by an applicant is not permitted under any circumstances in conjunction with application content. While an applicant may use artificial intelligence to assist with spelling and grammar review, in the same way as any other platform that supports basic proofreading, the content of all essays, short-answer questions and any other material submitted by an applicant must be the work of that individual.

Caltech publishes two lists for its Fall 2027 applicants. The unethical list has four items and the ethical list has three. Counted top to bottom, the four unethical ones are: copying and pasting directly from an AI generator; relying on AI generated content to outline or draft an essay; replacing your unique voice and tone with AI generated content; and translating an essay written in another language. The page also states that failing to comply `may result in the rescission of your admission to Caltech`.

Cornell's undergraduate FAQ draws the same line in its own words: ethical uses are `researching colleges, brainstorming essay topics, and reviewing the grammar and spelling of your completed essays`, while unethical uses include `using generative AI to outline, draft, or write your essays; translating an essay written in another language; and creating images for a required portfolio`.

Georgia Tech sits at the other end. Its Statement on AI, tucked inside a collapsed panel on the application review page, says AI `could be a helpful collaborator, particularly when you do not have access to other assistance to help you complete your application`, and asks only that `your ultimate submission should be your own` and that you `should not copy and paste content you did not create directly into your application`.

Babson, in its Common App help-center article: `If you choose to use generative AI, we recommend that you use this tool to enhance your original work and not create it.`

Johns Hopkins is the only one that puts a clock on it, which makes it the rule that maps most directly onto what a lot of applicants actually did:

It is Hopkins' policy that you may use AI tools for brainstorming or refining ideas before the writing process begins and for basic proofreading of your own writing, such as spelling and grammar review, but all material submitted within the application must be your own original work.

The next sentence borrows Common App's wording: `Submitting "the substantive content or output of an artificial intelligence platform, technology, or algorithm" is not permissible and is considered an instance of application fraud in accordance with Common Application policies.` Note the timing in that first sentence. Brainstorming and refining ideas are allowed before the writing process begins; proofreading is allowed on your own writing. The window in between is the part that is yours.

Cornell's Graduate School does something none of the undergraduate offices do. It requires disclosure:

If you use any GenAI tool while preparing your application, even for a permitted purpose, you must disclose it.

You add a short AI Use Disclosure to the statement of purpose naming the category of tool, what you used it for, and which sections it touched. You do not have to name the product.

Notice that using AI to translate an essay written in another language is listed as unethical by both Caltech and Cornell, and is not mentioned at all by Georgia Tech or Babson. If you drafted in another language, that one line is worth more to you than everything else on this page.

None of these documents describes a percentage triggering anything. They describe people asking questions. Three concrete examples, all from the documents above:

• Brown verifies `a small number of credentials each year for a randomly selected sample of admitted students who have chosen to matriculate`, by asking your school counselor to confirm selected factors. `Requests are sent to counselors in May with a response deadline no later than July 1.` Brown adds that those selected `need not worry`, because selection is random rather than suspicion-driven.
• UC `may request authentication of the content or writing as the student’s`.
• Cornell's Graduate School names `assessment of voice and consistency across materials, and verification through interviews or short writing exercises`.

There is a second path, where a complaint reaches Common App rather than a college. That one has a written timetable attached, with a deadline for your side of it, and it ends in a determination that can be disclosed to every college on your list. It is set out step by step in personal statement flagged as AI, and running through it twice would not make it any truer.

The examples above give you concrete things to prepare: UC may ask you to authenticate the writing, and Cornell's Graduate School may use interviews or short writing exercises. Common App's complaints process has its own written response timetable. That timetable does not establish a right to notice or a chance to respond in a college's separate review; check the procedure of the office handling your application.

The common situation right now is that you wrote the essay, ran it through a free checker because everyone said to, got a number you did not like, and are now staring at highlighted sentences wondering what to do before the deadline.

Start here. None of the fifteen documents mentions an AI score, a highlighted sentence, or a third-party checker. None gives an AI-detection percentage or a threshold for an AI score. So none of them says anything about those highlighted sentences in either direction, and hunting through them for a safe number is hunting for something that was never written. Every one of them asks about your process instead, including BYU, whose consequence attaches to an essay `found to have been generated by AI`, not to a score. Your process is the thing you can actually work on this week, and these five steps are what the documents support.

1. Find your specific school's rule, in its own words, and open the collapsed panels. Search the admissions site for "artificial intelligence", then click every accordion and FAQ toggle before you decide there is nothing there. Three of the sharpest policies in this article were invisible on page load, including the only one that says software runs over every essay. The rules quoted above run from an absolute ban on AI in content, through an office calling AI a helpful collaborator, to BYU reserving the right to rescind an offer. The average of those is not a rule, and Reddit gives you the average.
2. Check what you actually did against that rule, line by line. Brown, Caltech, Cornell and Georgia Tech all write theirs as a list of activities, so this is a genuinely answerable exercise: did I outline with it, draft with it, translate with it, run a grammar pass with it. Write your own answer down. That is the question the certification puts to you, and it is the question a person would put to you later.
3. Keep whatever shows how the essay came to exist. Version history in Google Docs or Word, dated drafts, the counselor email with track changes, the notes you wrote before you had paragraphs. UC reserves the right to request authentication of the writing as yours, and Cornell's Graduate School names short writing exercises and interviews. Both are easy to answer if you can show the essay's history and impossible to answer if you cannot. Clearing out old drafts to tidy up your Drive is the one avoidable mistake on this list.
4. If your school requires disclosure, disclose in the form it asks for. Cornell's Graduate School wants a short AI Use Disclosure inside the statement itself, naming the category of tool and what it touched. Improvising a disclosure somewhere that does not ask for one is a different thing and tends to read strangely.
5. If you are already in a conversation about a flag, prepare the material rather than the argument. We wrote up what that looks like in flagged, but you wrote it yourself.

One last note about the number itself. Application essays are short by design. UC's instructions read `Each response is limited to a maximum of 350 words`, and the Common App personal essay is not much longer. Short text is where these tools are least stable, and that is a property of the tools rather than a comment on your writing. How accurate is a Turnitin AI score on short documents and short documents and the all-or-nothing problem go through the mechanism. If your score moved after you cut the essay down to the limit, does cutting words change your AI score covers that exact case.

BYU says it uses software tools. Does that mean it runs an AI detector? It does not say so, and the difference is worth holding on to. The page says `software tools to analyze the admission essays and activity responses of all students`. It names no product, never uses the word detector, and states no percentage or threshold anywhere. The consequence attaches to an essay `found to have been generated by AI or otherwise plagiarized`, and that same sentence also covers plagiarism, which is a different kind of check entirely. So the accurate reading is that BYU tells you software is involved and does not tell you what it is. (On whether Common App itself checks anything, its fraud policy answers that directly, and personal statement flagged as AI quotes the sentence.)

Will admissions officers see the score I got from GPTZero? There is no mechanism described in any of these documents by which a score you generated on a third-party website reaches an admissions office. You ran the check; the result is on your screen. If what you actually want to know is why an essay you wrote yourself came back with a high number, personal statement flagged as AI is about the causes rather than the consequences.

I used ChatGPT to cut a 900-word draft down to 650. Is that fraud? It depends entirely on whose rule applies, which is why the honest answer is not a yes or a no. Under Brown's policy, AI may assist with spelling and grammar review and nothing else in the content. Caltech lists relying on AI generated content to outline or draft an essay as unethical. Georgia Tech asks that you not copy and paste content you did not create and that the submission ultimately be yours. UC permits generative AI `to assist with readability` as long as the content and final written text are your own. Johns Hopkins allows brainstorming and refining ideas `before the writing process begins` and proofreading `of your own writing`; a 900-to-650 rewrite reads to me as falling outside both, though both phrases are right there for you to weigh yourself. And BYU's commitment is flat: `you may not use generative AI tools (like ChatGPT) as you compose your responses`. Six different answers to the same question, all published. Read yours.

A school has no published AI policy. What applies? First make sure it has not: three of the policies in this article sit inside collapsed panels and do not appear in the page text until you click. If there genuinely is nothing, and you applied through Common App, the fraud policy and the certification you signed still apply, and both are quoted above. Colleges without a public AI page have not opted out of those.

Can an admissions office tell that I used AI? No document in this set claims to be able to. Two say in writing that they do not use or do not rely on detection software. BYU says it runs software tools over every applicant's essays, but it does not describe what that software concludes or how, and the consequence it names attaches to an essay `found to have been generated by AI` rather than to any output of the software. What the rest describe is comparing your materials to each other, asking your counselor, and asking you. That is what to be ready for.

Should I tell them I used AI for grammar? Only if the school asks. Cornell's Graduate School requires an AI Use Disclosure for any tool, including grammar tools. The undergraduate pages quoted here have no disclosure field, and several of them list grammar and spelling review as explicitly fine, which means there is nothing to declare.