The Blank Page Was Never the Hard Part
9 min read
It happens to nearly every junior and senior at roughly the same point in October. The SAT is done, the activity list is built, and the only thing left between you and the submit button is the essay. You open the Common App, you read the five prompts, and somewhere in that moment something becomes clear: the question you have been dreading — how do I write this? — is not actually the scary one. The scary one is the question before it. What do I say?
That is the question that keeps people up at night. Not the grammar, not the structure, not finding the right opening line. It is the harder, quieter problem of having too many true things about yourself and not knowing which one to commit to. The summer at the shelter. The year everything at home went sideways. That moment in rehearsal that nobody else was there for. The difference between how it feels to actually love something and how it feels to be supposed to love it.
Choosing between those things — that is the craft of a college essay. Not the sentences. The sentences are the easy part.
Which is why there is something worth examining honestly about what AI just did, and what it did not do. Most of the conversation has it backwards.
Before late 2022, getting from a blank page to a polished admissions essay was genuinely unequal in a way that did not get talked about enough. Students could either grind through months of drafts on their own — imperfectly, stressed out, under deadline — or families could pay for a private college counselor. In major cities, those counselors ran $150 to $450 an hour. The firms that advertised "Ivy outcomes" charged more. The National Association for College Admission Counseling found that for every public school counselor in the country, there were 408 students. At a well-funded private school, that ratio could be 1 to 80.
The blank page was, in other words, partly a money problem. Help crossing it was expensive, and the expense was real.
Then AI arrived and erased the blank page for everyone. Type a prompt into ChatGPT at 2 AM, get back a grammatically correct, properly structured, emotionally competent college essay in forty seconds. Free. Available to anyone with a phone. The expensive counselor, suddenly, costs nothing.
That sounded like a big deal. And it was — just not in the direction most people expected.
What the machine produces is the median essay. It is correct, it has a clear arc, it pivots in the right place, it ends with a forward-looking sentence about what you hope to bring to campus. It sounds like a lot of other essays, because it was trained on a lot of essays that worked, and "what worked on average" is exactly what it was built to produce. That is not a bug. That is the whole point. But there is a big difference between "what worked on average" and "what will work for the most competitive spots," and collapsing those two things is where the story started to unravel.
Researchers at Cornell and Carnegie Mellon spent 2026 studying roughly 81,000 college admissions essays submitted between 2020 and 2024. They estimated how much AI assistance each essay likely used, then tracked the outcomes. The results were not what the "AI levels the playing field" story predicted. Higher estimated AI use was linked to worse predicted admissions outcomes. And it was not evenly distributed — the effect was concentrated among lower-income students. The same students AI was supposed to help most.
Let that sit for a second, because it matters.
The mechanism, once named, makes sense. The expensive college counselor was never actually selling a polished essay. The polishing showed up on the invoice. But the real service was the conversation before the writing — the hour where someone helped you figure out which of your true things was worth saying. That is the excavation. AI skipped it completely and handed you the polish, and the part that got lost was the only part that was ever doing the real work.
The counterargument deserves a fair hearing, because there is a version of it that is genuinely reasonable.
The argument goes: the essay form is just exhausted. It has a shape — opening scene, reflective turn, forward-looking close — and that shape is now so thoroughly learned by machines that writing inside it is no longer an act of expression. You are just filling in a template the machine has already claimed. So the creative move is to abandon the form entirely and find spaces the machine hasn't mapped yet.
Forms do get tired. The five-paragraph essay was useful scaffolding until it became a cage. There is real truth in the idea that shapes can calcify.
But here is what that argument quietly requires: it requires believing the machine has perfected the essay form. Not just learned it — perfected it. Reached the top of what it can carry.
That is a very different claim from what actually happened. "Perfected" and "regressed to the mean" are not the same event. They are actually opposites. A perfected form means nothing is left at the ceiling — the best has been achieved, further effort is wasted. A form regressed to the mean means the floor has been raised to the middle and the ceiling is still exactly where it was, just standing in a much emptier room.
The machine raised the floor. It did not touch the ceiling. And the Cornell/CMU data shows this directly: if the median essay were the perfected essay, converging on it would be safe. It would at least hold steady. Instead, it performed worse. The admissions market had already started pricing the median down. That is not what a dead form looks like. That is what it looks like when the floor becomes free and the ceiling becomes the only thing anyone will pay for.
The equity piece still feels unresolved, and it deserves the honest version before moving on.
The strongest version of the other side says: okay, even if the machine does not produce the ceiling, at least it gives everyone a floor — and that floor used to require money that was distributed unfairly. The student who could not afford the counselor at least has something now. And the excavation problem — figuring out which true thing to say — was always easier for students with college-educated parents who could model what a compelling story sounds like, who could tell you that the thing that happened in rehearsal might be worth telling. AI does not help with that. But it does not hurt with it either. So why frame it as a loss?
That question is fair, and the concern is partly right. The excavation was never equally accessible, and AI's failure to help with it is a real gap.
But the Cornell/CMU data makes something specific visible that complicates the "at least the floor is higher" reading. AI did not leave lower-income students at the same disadvantage while improving their prose. It gave them a prose upgrade and a competitive penalty in the same package. Students who already had support systems — counselors, college-educated parents, schools with writing centers — kept those supports and added AI on top. Students who had only AI ended up with the median essay, which turned out to perform worse than their honest, unpolished, true-to-themselves essay would have. The democratization argument assumed polish was the scarce resource. In this particular market, it never was. What was scarce was the thing that gets a student from "I have no idea what to say" to "I know exactly what to say" — and that is precisely what AI does not provide.
There is one more move worth addressing: the "escape forward" version. Fine, if the essay is colonized, just do something the machine cannot do yet. Abandon the form. Find the genre it has not learned.
There is something to this — forms do exhaust and finding new shapes is real creative work. But look at what happened to the definition of creativity in that sentence. Creativity just got redefined as whatever the machine hasn't learned yet. Under that definition, creativity has no home. It is permanently running one step ahead of convergence. The moment the machine catches up, the form is dead and you have to flee again. That is not creativity. That is a very anxious game of territory.
And it repeats the same mistake in new clothes. The essay is not the thing being made. The person is the thing being communicated — The essay is the instrument. A machine that produces the median arrangement of sentences has not used up the space of things a person might genuinely need to say about themselves. It has made the average version of saying them free and abundant. The space above average is untouched, and it was always the space that mattered.
So here is where this lands.
The blank page was never the obstacle you thought it was. It felt like the obstacle because it was visible. Staring at it felt like the work. But what the blank page was actually doing — when a student sat with it long enough without anything to fill it in — was forcing the real question: What do you actually have to say?
That question did not go away. AI just made it possible to never ask it, and for a lot of students, that is the trap.
The craft of a college essay was never the sentences. It was knowing which memory is the true one — not the most impressive one, not the one that fits the arc neatly, but the one that, when you try to leave it out, makes you feel like you have written somebody else's essay instead of your own. It was knowing what to cut, and being brave enough to cut it, even when the thing you are cutting is the part that makes you sound most accomplished. Two students could make all those choices differently and both be completely right, because there is no single right answer for what is most true about a person.
A machine cannot find your true thing. It was not built to. It was built to produce the median of everyone else's, which is useful for many things but not for this.
The blank page is gone. The hard part is right where it always was.