Why AI writing may hurt applications to highly selective colleges

Many of my students have asked me about the value of generative AI in college essays. Many of them are asking a fair question. If AI can now create expressive and resonant writing in a few seconds, what is the value of writing the essay yourself? More importantly, does it really matter in terms of admission outcomes?

My answer is becoming clearer with time: AI can improve language. It can create polish. It can help a sentence sound better. But it cannot create the personal context from which a real college essay must emerge.

The entire premise of AI rests on predictability. In simple terms, it looks at what exists, mathematically predicts what is likely to come next, and creates something that is coherent, complete, and often well written. But a strong college essay, especially for the most selective universities, is not just about coherence. It is about voice.

Voice is not just style. It is not just word choice. It is the personal context behind the writing. It is the odd turn of memory, the hesitation, the small contradiction, the shape of an experience that belongs only to that student. This is the part that AI struggles to produce because it has not lived the life of the student. It can imitate patterns, but it cannot create the underlying personal truth.

This distinction matters most in highly selective college admissions. For universities with acceptance rates below 10%, polish is not enough. Many applicants already have strong grades, test scores, extracurricular activities, research work, internships, and recommendations. At that level, the question is whether something in the application allows the student to stand apart.

That is where AI writing can become a problem.

Too well-adjusted to what exists

Over the last two admission cycles, I looked at a set of Common App essays (used for applying to universities in the U.S. for undergraduate admissions) written by 167 students from India, the UAE, Thailand, Vietnam, and Singapore. These were not random applicants. They were among the stronger students in their school systems and had a realistic shot at selective U.S. universities.

In reading these essays, and in comparing them with outcomes, one pattern stood out. The essays that felt original had a person inside them. The writing seemed to come from a lived place.

The AI-assisted essays were often not bad. In fact, many were quite good at the surface level. They were smooth. They were grammatically clean. They had structure. They had emotional movement. They often used the right words at the right places. But after reading many of them, one felt a sameness.

It is a bit like a McDonald’s Happy Meal. It satiates you. It is designed to be consumed easily. It gives you something complete, familiar, and broadly acceptable. But it does not give you the Michelin-star experience. Nothing stands out.

That is the problem in the context of selective admissions. The issue is not that AI writing is bad. Often, it is not. The issue is that it is too well-adjusted to what already exists. It removes the oddity, the hesitation, the asymmetry, the small personal turn of phrase, the context that only this student could have created. In doing so, it may improve the sentence while weakening the person behind the sentence.

This is also why human mentors still matter. A human mentor does not have the universal reach of a large language model. That limitation may actually be useful. A good mentor does not generate infinite options. A good mentor listens, interrupts, asks, notices, and sometimes pushes the student back into the discomfort of thinking. The mentor’s job is not to make the essay sound like the best possible essay. It is to help the student find the most truthful version of her own story.

AI, when used too early, can interfere with that process.

At one extreme are students who open an AI chat window before they have even begun brainstorming. They ask for essay ideas, possible structures, opening lines, themes, and drafts. Their mind has not gone through the pathos of creation. It has not searched for options, tested arguments, or struggled to find the right shape of the story. As a result, it often does not create anything new.

The language model fills in the gaps. It predicts likely possibilities. It chooses the most fitting pattern. Since the next word is, in some sense, a prediction, the essay begins to move toward what is likely, not what is distinctive.

On the other extreme are students who build the argument themselves. They think. They get stuck. They try one version and reject it. They return to an experience and realize that the first interpretation was too easy. They discover that the essay is not really about the activity they thought it was about, but about a way of seeing the world.

That struggle matters.

There is a moment in the writing process when a student feels stuck and wants help. The request often sounds innocent: “I am just asking AI for validation. The core idea is mine.” But this is often the precise moment when the brain is building creative tension. If AI enters too soon, that tension ruptures. The student gets an answer before she has completed the act of thinking.

Akin to the marshmallow experiment, students may need to delay the “language model assist” until they have completed the parts that stretch the brain: brainstorming, narrative build-up, and the first draft. When they do that right to the very end, they force the mind to create options. Those options are the source of divergence. And divergence is central to creativity.

AI as fine-tuner

This does not mean AI has no role. It does. Once a student has written a full first draft, AI can help improve clarity, tighten language, point out repetition, or suggest where an argument is unclear. At that stage, it can add richness without taking over the origin of the piece. Used late, AI may amplify the essay. Used early, it may replace the thinking that should have produced the essay.

This is not only a college essay problem. It is a learning problem. If the first move is always outsourced, the mind does not learn to make the first move. It does not learn to sit with confusion. It does not learn to create options when none are visible. It does not learn to build an argument from inside itself.

Writing has always been one of the ways in which thinking becomes visible. A first draft is not merely a communication artifact. It is a record of how a mind approached a problem. It shows what the student noticed, what she ignored, what she exaggerated, what she misunderstood, and what she slowly came to understand. When AI enters before this process has happened, the draft may look better, but the thinking behind it may be thinner.

This is why early AI usage worries me more than imperfect writing. Imperfect writing can be improved. But if the student has not struggled to find the idea, then the essay may never fully belong to her.

The exact point at which AI enhances quality while preserving voice needs to be studied more rigorously. We need to understand how timing, usage, and intensity of AI support affect originality. But from what I have seen, the broad direction is clear.

The danger is not that students will use AI to write badly. The danger is that they will use AI too early and produce writing that is smooth, acceptable, and forgettable.

But the larger danger may be elsewhere. It is in the increasing disuse of the thinking muscle itself. If students repeatedly outsource the early struggle – the search for ideas, the discomfort of not knowing what to say, the act of building an argument from within – they may not just lose voice in an essay. They may lose the habit of creating a distinctive voice in the first place.

(The author is founder, ACadru, a multi-disciplinary learning platform for senior school and college students)