Essay Tips

AI and the College Essay: How Admissions Officers Are Adapting (And Why Detection Is Dead)

T20 Hub Research TeamCollege Admissions Intelligence

June 28, 2026•10 min read

You’ve heard the advice a hundred times: don’t use ChatGPT to write your college essay. But here’s what nobody is telling you — the colleges themselves are quietly giving up on catching you if you do.

The admissions world is in the middle of an identity crisis. The personal essay, long considered the one place where your authentic voice could shine through a sea of test scores and extracurriculars, is under existential threat. And the institutions that depend on it are scrambling to figure out what comes next.

The Detection Tools Don’t Work (And Everyone Knows It)

Let’s start with the uncomfortable truth. AI detection tools — Turnitin, GPTZero, Copyleaks, and the rest — are not reliable enough for high-stakes admissions decisions.

The numbers are damning:

61.3% of TOEFL essays written by non-native English speakers were falsely flagged as AI-generated in a Stanford study. That means if you’re an international student or English isn’t your first language, you’re more likely to be wrongly accused than a native speaker using ChatGPT.
A 2026 study published in Computers & Education found that a hybrid editing strategy lets 88% of AI-generated content evade detection. In other words, if a student makes even minimal edits to AI output, detectors are nearly useless.
Different tools produce wildly different results on the same text. There is no consensus “truth” — just competing algorithms with competing false-positive rates.

Vanderbilt University figured this out early. They instructed their admissions staff not to use Turnitin’s AI detector at all. Instead, they compare drafts, check for stylistic consistency, and cross-reference essays against the rest of the application. Johns Hopkins quietly disabled their detection tools too.

Duke went even further: they eliminated numerical essay scoring entirely in the 2023-24 cycle, citing the rise of generative AI. They still read essays, but they no longer rate writing quality — only content and insight into who the applicant is as a person.

The Voice-Matching Problem: Why Detection Is a Dead End

Here’s where it gets really interesting — and really concerning for anyone betting on detection.

Today’s LLMs can do something that would have seemed impossible two years ago: they can write in your voice. Not a generic “student essay” voice. Your voice.

A student can now feed ChatGPT ten samples of their writing from the past four years — school papers, journal entries, emails, whatever — and say: “Write my college essay about [topic], in my voice.” The model learns their vocabulary, sentence structure, rhythm, quirks, and even their characteristic mistakes.

Detection tools rely on statistical markers like perplexity (how “surprising” the word choices are) and burstiness (how much sentence length varies). Voice-matching prompts specifically neutralize both of these signals. The output isn’t generic AI text — it’s text that looks, statistically, like it came from the same person who wrote those ten samples.

And it’s not just voice matching. Students are using iterative refinement prompts:

“Make this sound more like a 17-year-old.”
“Add a few grammatical imperfections.”
“Remove the overly polished transitions.”
“Vary the sentence length more.”

Add a growing cottage industry of “AI humanizer” tools — GPTHumanizer, WalterWrites, and others — that rewrite AI text specifically to bypass detectors, and you have a situation where detection is fighting a battle it cannot win.

As one admissions officer put it: “If you can’t tell the difference, does the difference matter?”

What Schools Are Actually Doing (The Surprising Part)

If detection doesn’t work, what’s a college to do? The smart ones are pivoting — away from trying to catch AI and toward approaches that are structurally resistant to it.

Schools Using AI to Read Your Essays

Yes, you read that right. While students are told not to use AI to write essays, some colleges are using AI to read them.

Virginia Tech debuted an AI-powered essay reader for the 2025-26 cycle. Their system pairs one human reader with one AI reader per essay on a 12-point scale. If the scores differ by more than 2 points, a second human steps in. The result: Virginia Tech can scan roughly 250,000 essays in under an hour (versus about two minutes per essay for a human), saving 8,000+ staff hours and releasing admissions decisions about a month earlier.

UNC-Chapel Hill has been quietly using AI to score essays since 2019 — making them one of the earliest adopters. They spent nearly $200,000 on a system called Project Essay Grade (PEG) that scores essays 1-4 on writing mechanics. Humans can override the scores.

Stony Brook uses a tool called EddyAI that has automated more than 60% of incoming transcripts and also summarizes essays and recommendation letters for reviewers.

And here’s the part most people don’t know: Slate, the dominant admissions CRM platform used by 2,000+ schools, offers a built-in Reader AI powered by OpenAI that can summarize essays, highlight key topics, and pre-score written materials. It’s off by default — but the toggle is there in thousands of admissions offices. Almost none disclose whether they use it.

Caltech’s AI “Viva”: The Most Innovative Approach

Caltech did something genuinely new for the 2025-26 admissions cycle. Students who submit research projects with their applications must complete VIVA — a tool built by InitialView where an AI generates targeted questions about your research, you answer on video (like a mini dissertation defense), and Caltech faculty review the recording.

The point is authenticity. If you claim to have conducted neuroscience research, the AI will ask you about your methodology, your findings, your reasoning. You can’t ChatGPT your way through a live video interrogation.

Caltech’s Dean of Admissions Ashley Pallie put it this way: “It might seem strange to use AI to get more of a human voice, but I think of it as a way to bring more authenticity into the fold.” VIVA screened roughly 10% of early applicants in its first cycle, and Caltech plans to expand it.

The Video Essay Revolution

The fastest-growing alternative to the written essay is the video introduction. Since its 2023 launch, 30+ schools have signed up for InitialView’s Glimpse platform, expecting 50,000 to 150,000 video submissions in the current admissions cycle.

Students record 90-second unedited “talking-head” videos from their phones. No fancy editing, no polished productions. Schools using it include Brown, Duke, Vanderbilt, UChicago, Claremont McKenna, and many more.

Terry Crawford, CEO of InitialView, is blunt about why: “When people ask me why colleges are using this, I think there are three reasons: ChatGPT, ChatGPT, and ChatGPT. Essays don’t really have value anymore — that’s just the reality.”

But video introductions aren’t a perfect solution either. They favor affluent applicants who can afford coaching. They introduce appearance and race bias. And in the post-affirmative-action legal landscape, seeing what applicants look like creates potential liability. Goucher College pioneered video applications back in 2014 and eventually abandoned them — not enough predictive value.

Live Writing and Proctored Assessments

Some schools are going back to basics: make students write in real time.

Sciences Po in France has reintroduced timed entrance exams.
Pioneer Academics requires every applicant to complete a 30-minute interview: 20 minutes of oral questioning followed by 10 minutes of proctored writing with screen-sharing and camera on.
Rotman Commerce (Toronto) and Waterloo Engineering use Kira Talent for timed video and written prompts.
Bowdoin offers optional live interviews and requires students who submit videos to respond spontaneously to a random question — harder to script.

The Policy Landscape: A Patchwork of Confusion

If you’re applying to Top 30 schools right now, here’s the reality:

Policy Stance	Schools
Allow limited AI (brainstorming, grammar only)	Caltech, Cornell, Vanderbilt, Carnegie Mellon, WashU, Emory, UVA, UC system
Explicitly ban all AI	Brown, Georgetown
No official policy	Princeton, MIT, Harvard, Stanford, Yale, Duke, Johns Hopkins, Northwestern, UPenn, UChicago, Columbia, Dartmouth, Rice, Notre Dame, UMich, UNC, USC, NYU

That’s right — 70% of Top 30 schools have no official AI policy. Most rely on the Common App’s fraud policy (which classifies AI-generated content as fraud) and honor pledges that students sign when submitting their applications.

Brown takes the hardest line: “The use of artificial intelligence by an applicant is not permitted under any circumstances.” Georgetown makes applicants sign a statement that AI use is prohibited and reserves the right to rescind admission.

Caltech and Cornell take a more nuanced approach. Their rule of thumb: AI can do anything you’d ask an adult to help with. You might ask your mom to check your spelling, but you wouldn’t ask her to write the whole draft.

The Equity Question Nobody Wants to Answer

This whole debate has an uncomfortable underbelly: college admissions have never been fair, and AI makes that worse in both directions.

On one hand, students from under-resourced schools without college counselors may use AI as their only source of writing feedback. Blanket bans risk punishing the students who need help the most. As Pioneer Academics founder Matthew Jaskol notes: “For a long time, students from well-resourced schools have benefited from structured application support, while others have had little guidance. AI, if used responsibly, can help level a playing field that has never really been equitable.”

On the other hand, wealthy students already have essay consultants charging $10,000+ per application cycle. One report found that a fourth of Harvard’s Class of 2027 relied on paid help. For these students, AI is just one more tool in an already-unequal arsenal.

What’s Coming Next

Based on where things are heading, here’s what the admissions landscape will look like in the next few years:

Detection tools will fade. Voice-matching and hybrid writing make binary detection obsolete. Schools that still rely on detectors will quietly abandon them.
Video and live components will expand. More schools will require or strongly encourage video introductions, timed writing, or live interviews.
Essays will get shorter and more specific. The UK’s UCAS is already replacing traditional personal statements with three structured short-answer sections in 2026. Expect U.S. schools to follow — shorter, targeted prompts are harder to AI-generate convincingly.
“Anchored” writing will become standard. Schools will increasingly require graded class papers as a baseline comparison for application essays.
The essay’s weight will decline. Duke already reduced it. Others will follow as the signal-to-noise ratio deteriorates.
Process-based evaluation will grow. Portfolios, research logs, multi-stage assignments, and revision histories — things that are hard to fake in a single ChatGPT session.

What This Means for You

If you’re a student applying to college right now, here’s the practical takeaway:

Don’t try to outsmart the system — out-authentic it.

The colleges that are winning the AI arms race aren’t doing it with better detectors. They’re doing it by asking questions AI can’t answer: What did you learn from your research? What does your voice sound like when you’re speaking off the cuff? What does your writing look like when you have 10 minutes and a proctor watching?

The best strategy isn’t to avoid AI entirely or to use it cleverly. It’s to use AI for what it’s good at — brainstorming, organizing ideas, checking grammar — and then write the essay yourself, in a voice that no model can replicate because it’s built from experiences no model has had.

Your unique perspective, your specific memories, your weird quirks — those are the things that make an admissions reader lean forward. And those are the things that no prompt, no matter how sophisticated, can fake.

Sources

This analysis draws on reporting and research from AP News, Washington Monthly, College Essay Advisors, GradPilot, Pioneer Academics, Forbes, Virginia Tech News, and peer-reviewed studies published in Scientific Reports (Nature) and Computers & Education. For the complete research report with all citations, see our AI Essays Admissions Research page.