The honest answer isn't "more" or "less" -- it's that AI changed what the essay is actually being evaluated for, and the real research on this, which now includes several genuine academic studies rather than just admissions-blog speculation, points in a more specific and more useful direction than either simple answer would.

What happens when a reader suspects AI was involved

A 2024 study from the research group Foundry10 tested this directly: 425 high school teachers and 523 college applicants read identical essay paragraphs, each attributed to one of three sources -- ChatGPT assistance, professional admissions coaching, or no outside help -- and rated them on a 5-point scale. When readers were told ChatGPT was used, perceived authenticity dropped from 3.98 to 3.09 (roughly a 22% decline), perceived competence fell from 4.04 to 3.42, and perceived ethics dropped from 4.04 to 2.86 -- all statistically significant, and notably, the applicant's name, race, and gender didn't change the size of the penalty. The essay itself didn't get worse; the label attached to it did the damage. That's a striking result, because it means the cost of being perceived as AI-assisted is severe even when the underlying writing is identical.

What admissions officers actually say -- and how few colleges have an actual rule

Kaplan's 2025 survey of admissions officers at 220 of the country's top national, regional, and liberal arts colleges found real skepticism: 50% consider AI assistance in essays unethical, 26% are neutral, and only 14% view it favorably. But there's a real gap between that sentiment and actual policy -- 68% of colleges have no official policy at all on whether applicants can use generative AI to write their essay (2% explicitly allow it, 30% explicitly ban it). Attitudes loosen considerably for narrower uses: 27% of colleges allow AI for brainstorming and 21% allow it for feedback on a draft, with bans in the single digits for those uses. The takeaway: officers are wary of AI writing the essay outright, more permissive of AI as a brainstorming or editing tool, and most schools still haven't formalized a rule either way -- which puts real weight on using judgment rather than assuming a clear line exists everywhere.

Detection is technically very good -- in a research lab

A January 2026 Cornell study paired 30,000 real Common App essays with nearly 88,000 AI-generated synthetic essays from eight frontier language models and trained classifiers to distinguish them. The classifiers hit F1 scores of 0.998-0.999 -- essentially perfect separation. The mechanism is intuitive once you see it: AI-generated essays cluster tightly around a narrow set of abstract words ("challenge," "growth," "resilience"), with far higher similarity to each other than real human essays have to each other. Worth being precise about what this does and doesn't prove: it shows AI writing is statistically distinguishable at scale in a research setting, not that any individual admissions office is running this exact detection pipeline on your specific essay -- but it confirms admissions readers' instinct that heavily AI-assisted essays tend to sound alike in a detectable way.

The part that should concern you more than getting caught

A separate, more troubling Cornell study analyzed over 81,000 de-identified Common App essays submitted to one selective engineering program between 2019 and 2024, using fee-waiver status as a proxy for socioeconomic status. Before ChatGPT existed, the admit-rate gap between higher- and lower-SES applicants was 10.7 percentage points. After ChatGPT became widely available, that gap widened to 14.0 percentage points -- even though lower-SES applicants used LLM assistance 28% more often, they paid a per-unit penalty for that use that was 1.85 times larger than higher-SES applicants paid for the same degree of use. A related study found that AI-generated essay text aligns with male-coded, continuing-generation, and high-income-ZIP-code writing patterns 65-92% of the time -- and among a set of Latinx UC applicants, 20% wrote parts of their real essays in Spanish, while 0% of AI-generated comparison essays did. Put together: AI isn't a neutral equalizer for applicants without access to expensive essay coaching -- in this data, it reproduced the voice of already-privileged applicants and penalized the students who leaned on it most.

Some schools are responding by cutting the essay, not doubling down on it

Not every school is reacting by scrutinizing essays harder. As covered in our piece on colleges dropping supplemental essays, Tulane explicitly cited the rise of AI writing tools as a reason for eliminating its supplemental essay -- when a polished, on-topic response can be generated in seconds, a short prompt stops reliably telling a school anything school-specific about the applicant. It's not just a U.S. phenomenon: UCAS, the UK's centralized university admissions system, announced it's replacing its long-form personal statement with three shorter, more specific prompts starting in 2026, a change explicitly framed around AI concerns -- shorter, more targeted prompts are simply harder to answer generically well.

So: more important, or less important?

Neither framing quite fits what the evidence shows. The essay hasn't become less scrutinized -- if anything, readers are leaning on it harder as the place they test for a genuine, specific, idiosyncratic voice, precisely because AI made generic-but-polished writing cheap and abundant everywhere else in the application. But any individual essay that reads as AI-generated is now actively, measurably penalized rather than just quietly ignored, and the schools responding to AI by eliminating the essay entirely are making a different bet: that a short writing sample isn't worth keeping if it can no longer reliably distinguish applicants. The essay's evidentiary value specifically as proof of authentic voice went up. Its reliability as a fair, neutral measuring stick across applicants with unequal access to AI tools and coaching went down -- and that second part is the more serious problem hiding underneath the first.

What this means for you

  • Write in your own voice, even where it's rougher than a polished AI draft would be. The research consistently shows readers penalize essays that read as AI-generated far more than they reward technical polish -- authenticity is the scarcer, more valuable signal right now, not smoothness.
  • Using AI for brainstorming or feedback on a draft is viewed far more permissively than having it write your essay -- but check each specific school's policy if one exists, since a third of colleges explicitly ban AI use for the writing itself.
  • Don't assume AI assistance levels the playing field. The available research found the opposite in real application data -- AI-influenced essays skewed toward already-privileged writing patterns, and the "AI use" penalty landed hardest on lower-income applicants, not evenly across the board.
  • If a school on your list has dropped its supplemental essay, your personal statement is carrying more relative weight there -- not less scrutiny, just concentrated onto fewer words.
  • If "write in your own voice" feels easier said than done, start with a structured brainstorming technique rather than a blank page -- genuine voice tends to show up in specific, real material, not in trying to sound authentic in the abstract.

Sources