Which Colleges Use AI to Read Your Application First
Find out which colleges use AI in admissions for 2026-27, what those tools actually score, and how to write essays that hold up under both AI and human review.
By Jorbi TeamVirginia Tech's AI essay tool can scan around 250,000 essays in under an hour. A human reader doing the same work averages two minutes per essay, which works out to about 8,300 human-hours. That gap is why AI in college admissions went from rumored to operational faster than most students realize, and for the 2026-27 cycle, you need to know exactly which schools are using it and what it means for your application.
Skip the think pieces about where admissions is headed in five years. This is a school-by-school breakdown of who's actually running AI on your essays right now, what those systems are looking for, and what you should do differently because of it.
The Confirmed Schools Using AI in Admissions
Let's be specific, because the vague "some schools use AI" framing floating around Reddit isn't useful. Here are the confirmed institutions, with verified details on what they're actually doing.
University of North Carolina at Chapel Hill is the earliest major adopter on record, having used automated essay scoring since 2019. That predates ChatGPT by years. The Los Angeles Times reported that the system scores your Common App essay on grammar, writing style, and academic rigor on a 1-to-4 scale and generates "data points" pulled from your transcript. The tool is Slate's Reader AI for document summarization, paired with separate scoring software for essays. UNC faced significant backlash after The Daily Tar Heel broke the story in January 2025, including criticism from parents and applicants. The school updated its website with a clarification but declined further comment.
Virginia Tech is the most extensively documented case. For the 2025-26 cycle, the school replaced one of its two human readers per essay with an AI reader. Every one of your four short-answer essays gets scored by both one human and one AI on a 12-point scale. If their scores differ by more than two points, a second human reviewer steps in.
The tool is in-house, built over roughly three years. Juan Espinoza, Virginia Tech's Vice Provost for Enrollment Management, told VPM that the AI "does not experience fatigue or variation in attention" and is being used "to confirm, not to replace, the score provided by the person." The practical benefit for applicants: Virginia Tech is now delivering decisions about one month earlier than usual.
Caltech has the most novel approach. If you submit a research paper or publication as part of your application, you may be asked to complete VIVA, an AI-powered video viva voce. The tool is built by a company called InitialView; the AI generates questions about your submitted research, you answer on video, and Caltech faculty and admissions officers review the recording. VIVA screened roughly 10% of early applicants in the 2025-26 cycle.
Dean of Undergraduate Admissions Ashley M. Pallie told the LA Times the goal is to assess whether applicants "can speak about their work with understanding and interest." Translation: if you didn't actually do the research you're claiming, you will get caught.
Stony Brook University uses AI to process transcripts and summarize essays and recommendation letters, surfacing context that admissions officers should notice, things like family responsibilities or health challenges a student mentioned in passing. Senior Associate Provost Richard Beatty confirmed the system is being used to "highlight things an admissions officer should consider."
Georgia Tech uses AI for transfer applicant transcript review and flags Pell Grant-eligible students, suggesting the system has an equity-oriented function alongside efficiency. An expansion is planned for the coming years.
Beyond these five, a broader category is worth knowing about: schools using AI to detect AI-written content rather than to evaluate applications. GradPilot's database of 166 schools shows that about 28% use some form of AI detection or screening, including Duke, Carnegie Mellon, Columbia, BYU, and Bates. Penn State Smeal runs essays through iThenticate with a zero-tolerance policy; violations result in automatic denial. UCLA Anderson uses Turnitin.
Johns Hopkins, interestingly, has explicitly disabled its AI detection tools over concerns about false positives and the damage of wrongly flagging authentic essays.
What These AI Systems Actually Do to Your Essay
These tools aren't reading your essay the way a tired admissions officer at 11pm reads it. They're measuring it.
Scoring rubrics at schools like Virginia Tech and UNC reward concrete structure, clear argumentation, and grammatical consistency. An in-house AI trained on years of high-scoring essays will reward essays that fit that pattern and flag ones that don't, whether the weirdness comes from AI generation or just an unconventional narrative style.
The research data on detection accuracy is sobering. A Cornell study from January 2026 found that a classifier trained on 30,000 real Common App essays versus nearly 88,000 AI-generated ones achieved an F1 score of 0.998. Under controlled conditions, AI-written essays are highly detectable.
The practical caveat: detection tools used in admissions offices aren't operating under controlled lab conditions. Turnitin reports roughly a 4% false positive rate per sentence. For a 500-word essay, that could mean one or more sentences incorrectly flagged as AI-generated, with higher rates for international and ESL students.
There's also a perception problem that exists entirely apart from detection tools. A Foundry10 study from 2024 found that when a reader was told ChatGPT helped write an essay, perceived authenticity dropped from 3.98 to 3.09 on a 5-point scale, a 22% decline. The essay didn't change. The perception did.
The Policy Landscape Is a Mess, and That's Your Problem
The strangest part of all this is the asymmetry. Schools may be running AI on your application without telling you, and most have no formal policy on whether you're allowed to use AI in writing it.
GradPilot's State of AI Report covering 174 schools found that 117 of them, 67%, have zero explicit AI admissions policy. A Kaplan survey of 220 admissions offices found a nearly identical figure: 68% have no formal policy. About 30% of schools ban AI outright per Kaplan's data, up from 25% the year before. Only 2% explicitly allow it.
Penn is worth mentioning here because it's the only top-10 school to have publicly said it doesn't use AI in admissions review. Whitney Soule, Penn's VP and Dean of Admissions, told GradPilot directly: "No, we're not using it... I don't see that AI could substitute for a person's discretion or review." That's a meaningful statement when most schools simply don't answer the question.
The binding policy that applies to everyone, regardless of what any individual school says, is the Common App's fraud definition, updated in August 2023 and in effect for the 2026-27 cycle:
> "Submitting... the substantive content or output of an artificial intelligence platform, technology, or algorithm" as your own original work constitutes application fraud.
Common App CEO Jenny Rickard has acknowledged there's no fixed definition of "substantive," but fraud findings result in account suspension and notification to every school on your list. Yale has gone further in its own language, stating that violations can result in admission revocation or expulsion.
Silence from your target school about AI policy is not permission. The Common App clause applies regardless.
How to Write Your Application When AI Might Read It First
The practical guidance here pulls from admissions counselors, peer-reviewed research, and what we know about how these scoring systems actually work.
Write your first draft yourself, without any AI. The practical case for this is just as strong as the ethical one. Multiple counselors converge on this point: your first cold draft, written from memory without a thesaurus or AI suggestions, will produce the specificity that both human readers and AI scoring rubrics reward. The small sensory detail from that specific afternoon. The exact thing your coach said. AI can't generate those moments because it wasn't there.
Write in Google Docs from day one. Google Docs saves a timestamped version history that documents your essay developing sentence by sentence over days or weeks. If your essay is ever questioned, this history is your best evidence of authenticity. Pasting a finished AI draft into a blank document leaves no trail.
Run the read-aloud test before you submit. Read your essay out loud. If a sentence is hard to say in one breath, it's probably too long. If something doesn't sound like how you'd actually talk to someone, it needs revision. This is the single most reliable authenticity check available to you.
Don't inflate your vocabulary. Writing above your authentic level is one of the clearest signals both AI detection tools and experienced human readers pick up on. Using "fastidious" when you'd naturally say "careful" doesn't make the essay stronger. It makes it suspicious.
If you're applying to Caltech and submitting research, prepare to defend it out loud. Treat the VIVA the way you'd treat a dissertation defense. Know your methodology. Understand your findings. If you can't answer follow-up questions about work you submitted, the system flags it automatically.
AI is reasonable to use for brainstorming topics, generating follow-up questions to help you explore an idea, and checking grammar and spelling on a draft you already wrote. Caltech explicitly permits grammar and clarity edits with disclosure.
Here's the sharper reason to keep AI out of your actual drafting: the Cornell study found that GPT-generated essays linguistically align with male, continuing-generation, and higher-income applicants 65-92% of the time across measurable features. Lower-SES students who used AI in their essays were penalized 1.85 times more per unit of AI use than higher-SES applicants. The students who thought AI would give them an edge were the ones hurt most by it.
Frequently Asked Questions
Does AI read my Common App essay before a human does?
At most schools, no. The confirmed schools currently using AI to score or summarize essays are Virginia Tech, UNC-Chapel Hill, Stony Brook, Georgia Tech, and Caltech (for research applicants). That said, roughly 50% of admissions offices use some form of AI, including transcript processing and application chatbots, and 67% have no public policy about it. Most schools haven't said what they do.
Can colleges detect if I used ChatGPT to write my essay?
Under controlled conditions, yes, with very high accuracy. The Cornell study from January 2026 found AI classifiers can separate real Common App essays from LLM-generated ones with an F1 score near 0.999. Real-world tools like Turnitin have roughly a 4% false-positive rate per sentence, which is why Johns Hopkins disabled them.
Detection tools can flag you, but the more common outcome is a human admissions officer reading an essay that doesn't sound like a real person and marking it down. That's the risk most students underestimate.
Is it against the rules to use AI on my college application?
Submitting AI-generated content as your own work violates the Common App's fraud policy, which applies to all 1,000-plus Common App member colleges whether or not they have a separate stated policy. Using AI to brainstorm or check grammar is generally considered acceptable; using it to draft the essay itself is not. Yale and Penn State Smeal have especially explicit consequences, including admission rescission.
Which college is most aggressive about AI screening right now?
Caltech is the most distinctive, using an AI-powered video interview (VIVA) to verify whether research-submitting applicants actually understand the work they claim. Penn State Smeal uses iThenticate with zero-tolerance enforcement and automatic denial for violations. Among schools using AI to score essays, Virginia Tech is the most transparent about its process and the most publicly documented.
What to Do Next
Four concrete steps to take before you start writing this fall:
- Look up your target schools on GradPilot's AI policies database. It covers 174 schools and tells you whether each one restricts student AI use, uses AI to screen applications, or requires disclosure. Takes five minutes and tells you exactly where you stand.
- Open a Google Doc and write a rough, messy first draft of your Common App essay this week. Don't edit. Don't use AI. Just write from memory for 30 minutes. That raw material, your specific language and your specific details, is what you'll build the final essay from.
- If you're applying to Caltech and plan to submit research, schedule time to practice your verbal defense of that work. Read the VIVA guidelines on Caltech's site. Answer out loud, without notes, every methodological question someone could reasonably ask.
- Read the Common App's fraud definition in full. It's about 100 words. Knowing exactly what it prohibits (submitting AI output as your own) versus what it doesn't address (using AI to brainstorm) will save you a lot of anxiety and help you make smarter decisions throughout the process.
The schools using AI to read your essays are doing it in the open, or at least open enough that it's documented. That's actually useful information. Use it.