Are Colleges Cutting Admissions Readers? What Records Show

Are Colleges Cutting Admissions Readers? What the Public Records Can — and Can't — Show

The promise that "a human reads every word of your essay" is quietly expiring. The strange part: you can't prove whether it's already gone at the school you applied to.

Start with the fact that broke the spell. For the 2025–26 cycle, Virginia Tech pairs one human and one AI reader on every application essay — the AI replaced one of the two humans who used to read each file. It scans roughly 250,000 essays in under an hour, a second human steps in only when the AI and the human disagree by more than two points on a twelve-point scale, and the office estimates it saved about 8,000 staff hours. "The AI does not get tired. It doesn't get grumpy. It doesn't have a bad day," Virginia Tech's vice president for enrollment management told the Associated Press in December 2025.

That a machine now reads admissions essays is, at this point, well documented school by school — and we keep a running roster of which colleges use AI and the double standard of banning student AI while deploying it on your file. This piece asks the next question, the one almost nobody has tried to answer: if AI is reading, are there fewer humans being hired to read? And could you find out if there were?

The short version: the headcount has not visibly collapsed — yet. But the work each human reader does is shrinking, the budget pressure is real, and when I went looking for hard numbers in the public records that are supposed to make public universities transparent, the records turned out to be structurally blind to exactly this question. Here is what is knowable, what isn't, and what it means for the essay you're about to submit.

First, the part that's confirmed: AI is reading

Before the headcount question, the baseline. AI-assisted reading is no longer a pilot at a handful of schools; it's a documented practice at named institutions, mostly large public universities under volume pressure:

The plumbing underneath matters as much as the marquee names. Technolutions Slate — the dominant admissions CRM, used by more than 1,900 offices — shipped a "Reader AI" feature in 2025 that summarizes essays, letters, and transcripts inside the file reader. It's opt-in, so adoption is uneven and largely undisclosed, but the capability now sits one toggle away in the software most American admissions offices already run. We've written separately on what Slate's Reader AI actually does and on Georgia Tech becoming the first to publish an AI admissions policy.

Graduate and professional admissions moved first in some respects: the AAMC reported in January 2025 that medical schools use AI to triage applications, with one school screening roughly 5,000 applications down to 1,500–2,000 for human committee review. The pattern across every confirmed case is the same official line — AI assists, humans decide. That line is true at the final-decision stage. It is much shakier at the first-pass stage, which is where most applicants are actually sorted.

One stat you'll see everywhere deserves a flag: the claim that "87% of colleges let AI make admissions decisions" traces to a single 2023 commissioned survey of 399 respondents, and it's contradicted by every on-the-ground report since. Treat it as a contested marketing figure, not a fact.

So are there fewer human readers?

Here the evidence forks, and honesty requires holding both halves at once.

In raw numbers, colleges are still hiring readers. Seasonal and contract reader postings were plentiful for 2025–26 — Cornell, Brown, Boston University, Spelman, Harvey Mudd, Washington & Lee (advertised at $30/hour), Utah, NYU. The University of California still scales up dramatically in season; UC San Diego's reading team reportedly grows from around 50 to roughly 300 during the crush. Record application volume keeps demand for warm bodies high.

But the work each reader does is shrinking, and at least one school has said the quiet part out loud. Virginia Tech didn't add an AI reader on top of two humans — it replaced one of them. The University of San Francisco was blunter: it told a reporter that AI "might allow it to be less reliant on the 10 outside readers it hires each year." No named school has yet confirmed actually cutting its reader pool — every reduction so far is phrased as may or might — but the direction is unambiguous, and the people inside admissions aren't pretending otherwise. The founder of NACAC's AI special-interest group put it on the record: "Ten years from now, all bets are off. I'm guessing AI will be admitting students."

So the truthful answer to "are colleges cutting admissions readers" in mid-2026 is: not as headcount, not yet, not visibly — but as hours-of-human-judgment-per-applicant, absolutely. A school can keep its reader count flat and still hand the machine the first, formative read of your essay. That's the shift that matters, and it's invisible in a job-posting count.

The squeeze: why this is happening (it's arithmetic, not villainy)

The reason to resist a conspiratorial frame is that the pressure is structural and easy to document. Four forces are converging:

• More applications. Common App first-year applications rose about 8% in the 2024–25 cycle (10.2 million applications from 1.5 million applicants) and were up another ~9% year-over-year through December 1, 2025. With roughly 80% of four-year colleges still test-optional, essays and the rest of the holistic file get read for nearly everyone.
• Fewer actual students. The "demographic cliff" is real and arrived on schedule: U.S. high-school graduates peaked in 2025 and are projected to fall ~13% by 2041 (WICHE). The paradox that breaks admissions offices: more applications from a shrinking pool, because anxious students hedge by applying to more schools.
• Less money. New international enrollment fell about 17% in fall 2025 — the largest non-pandemic drop in over a decade — costing an estimated $1.1 billion and roughly 23,000 jobs (NAFSA), on top of federal research-funding cuts and state budget pressure. International tuition was the high-margin revenue that quietly subsidized a lot of admissions operations.
• Crushed capacity. EAB reports that applications per open admissions position are down ~80% and that admitted-students-per-counselor has climbed from around 500 to 750 during yield season.

More files, fewer students, less revenue, thinner staff. That's the whole engine. The schools automating the read mostly say so directly — they're adopting AI because the math left them no slack, not because they set out to remove humans. (For the longer evidence trail on what's driving adoption, see our roundup of the major studies on AI in admissions and the enforcement gap between AI rules and reality.)

I tried to count the readers. The public records are blind.

Here's where it gets genuinely strange — and where this story stops being a recap and starts being reporting.

Public universities are supposed to be countable. They file employee data; their salaries are public; mass layoffs trigger notices. So the obvious move is to pull the records and watch the admissions-reader headcount rise or fall over time. I went through the candidate sources one by one. Almost none of them can isolate admissions readers, and the contract readers — the ones AI most directly displaces — are invisible in all of them.

Two structural problems sink the naive approach. First, the federal data is too coarse and now actively degraded: NCES, which runs IPEDS, was gutted in 2025, so the human-resources releases are late and error-prone — the most recent reliably comparable year is still fall 2022. Second, and more fundamental: the application "reader" is, increasingly, a seasonal contractor paid by the hour for three months a year. That worker doesn't show up in IPEDS (built for permanent staff), often doesn't appear in a payroll snapshot taken on a single date, and never triggers a layoff notice. When a school stops rehiring its reader pool because the AI now does the first pass, nothing in the public record changes. The disappearance is invisible by construction.

What the records can tell you is the part nobody markets: admissions reading was already a high-churn, low-paid job. CUPA-HR found that roughly 71% of admissions counselors have been in their role three years or less. The human reader was a revolving door before AI showed up; AI is now quietly absorbing the turnover instead of refilling it.

What the budget filings DO show

If readers are invisible, the broader cuts are not. Inside Higher Ed tracked more than 9,000 higher-ed job cuts and buyouts in 2025, and admissions/enrollment offices were not spared — though usually as one unit among many rather than singled out:

• Western Washington University consolidated enrollment management, advising, and outreach under the provost and eliminated roughly 100 positions.
• New Jersey City University, merging into Kean, is eliminating its admissions, registration, and advising offices outright.
• Penn State is centralizing recruiters under a "Unified Enrollment Management Initiative."
• Eastern Illinois University cut 44 positions and named declining international enrollment as "the single most significant factor."

Hiring freezes hit public flagships across 2025–26 (Maryland, Oregon, UT-Arlington, Washington), and admissions is covered by inclusion — frozen, not spotlighted. The macro picture is consistent with quietly thinning reader pools; it just can't prove it at the reader level. That gap between the obvious pressure and the unprovable specifics is the whole story.

What this means for your essay

Step back from the institutional plumbing and this lands on one applicant, with practical consequences:

1. The "a human reads every word" assumption is dead at high-volume publics. Your essay's first reader may be a model trained on past essays and a rubric (Virginia Tech, UNC). At selective and graduate programs, humans still read — Duke has publicly refused AI screening — but even there, AI may pre-summarize your file before a person sees it.
2. Two different AIs may be judging you in opposite directions. A scoring model rates your essay's quality while a detection model decides whether you used AI to write it. Optimizing for one can trip the other. (We've covered whether colleges actually run detectors and what schools really spend on detection.)
3. Don't write to the machine. This is the actionable finding. A Cornell and Carnegie Mellon study found AI-written essays homogenize toward the same template and are detectable with near-perfect accuracy — and that lower-income students, leaning hardest on free AI tools, converged most and still got rejected at higher rates. Generic gets penalized twice. (More: how AI flattens essays into a male, privileged voice and the word-fingerprints that give AI away.)
4. Mechanics now carry algorithmic weight at some schools. UNC's tool scores grammar, sentence variability, and length. Unconventional structure or non-native phrasing can be docked by the machine before a human supplies context.
5. Keep your drafts. AI detectors falsely flag real writing — especially from multilingual students — and the burden of proof flips to you. Version history is your defense.

What we don't know (and won't pretend to)

Honesty about the limits is the point of this piece, so the caveats are part of the finding, not a footnote:

• No clean national headcount trend exists for admissions readers. Job-board totals are noisy snapshots; the federal data can't isolate the category; seasonal readers are uncountable. Anyone claiming a precise "colleges cut X% of readers" figure is guessing.
• The confirmed AI-reading cases cluster around a single reporting wave (largely a December 2025 AP story). They're real and verified, but they aren't proof of universal adoption.
• A few sources here were paywalled and reconstructed from search summaries; specific quotes should be re-verified against the primary article before anyone repeats them as gospel.

The honest headline isn't "colleges fired the humans." It's narrower and, in a way, more unsettling: the human read is being hollowed out one essay at a time, the system that's supposed to let the public watch can't see it happening, and the only people who could tell you — the admissions offices themselves — mostly aren't.

---

GradPilot is an independent essay-feedback platform. This article reports on publicly documented practices and public-records limitations; it does not allege wrongdoing by any named institution, all of which describe their AI use as assistive with humans making final decisions. Figures are sourced and dated to mid-2026 and will shift as the 2026–27 cycle and delayed federal data land.

Sources

• Associated Press, "AI may be scoring your college essay" (Dec 2, 2025) — Virginia Tech, Georgia Tech, Caltech, NACAC.
• Daily Tar Heel, UNC–Chapel Hill essay auto-scoring (Jan 2025).
• AAMC News, "AI will now read your medical school application" (Jan 21, 2025).
• Technolutions Slate — Reader AI feature documentation (2025).
• NAFSA, Fall 2025 International Student Enrollment Snapshot & Economic Impact (Nov 2025).
• WICHE, Knocking at the College Door (11th ed.) — demographic projections.
• Common App, End-of-Season Report 2024–25 and December 2025 deadline update.
• EAB, admissions-staffing analyses (applications-per-position, admits-per-counselor).
• CUPA-HR, The Higher Ed Admissions Workforce (Apr 2023) — headcount, tenure, turnover.
• NCES/IPEDS Human Resources component; Inside Higher Ed on the 2025 NCES cuts and on 2025 sector layoffs.
• Cornell/Carnegie Mellon study on AI-written admissions essays (Cornell Chronicle, Sep 2025; Inside Higher Ed, May 2026).