Many of my students have asked me about the value of generative AI in college essays. Many of them ask fair questions. If AI can now create expressive, heartfelt sentences in seconds, what’s the value of writing your own essays? More importantly, does it really matter when it comes to admissions outcomes?
My answer will become clear over time. AI can improve language. You can create polish. It helps improve the sound of your writing. But that doesn’t create the personal context in which the actual college essay must occur.
The entire premise of AI is based on predictability. Simply put, it examines what exists, mathematically predicts what will happen next, and produces something that is consistent, complete, and often well-written. But a good college essay is about more than consistency, especially for the most selective universities. It’s about voice.
Voice is not just about style. It’s not just a word choice. It’s the personal background behind the writing. It is a strange development of memory, a hesitation, a small contradiction, a form of experience that belongs only to the student. AI has a hard time producing this part because it doesn’t live the student’s life. You can imitate a pattern, but you cannot create an underlying personal truth.
This distinction is most important in highly selective university admissions. For colleges with acceptance rates below 10%, brushing up isn’t enough. Many applicants already have a track record of excellent grades, test scores, extracurricular activities, research work, internships, recommendations, etc. At that level, the question is whether something in the application allows the student to stand out.
That’s where AI lighting can come into play.
too well tuned to what already exists
Over the past two admissions cycles, I examined a series of Common App essays (used to apply for undergraduate admission to U.S. universities) written by 167 students from India, the UAE, Thailand, Vietnam, and Singapore. These were not randomly applied. They were among the strongest students in the school system and had a realistic chance of getting into selective US universities.
As I read these essays and compared them to the results, one pattern stood out. Essays that felt original included people in them. The writing seemed to come from a living place.
Most of the AI-assisted essays were not bad. In fact, many were very good on a surface level. It was smooth. It was grammatically clean. They had structure. They had emotional dynamics. They often used the right words in the right places. But after reading many of them, one felt the same thing.
It’s similar to McDonald’s Happy Meal. it will satisfy you. Designed for easy consumption. It offers completeness, familiarity, and universal acceptance. But you don’t get the Michelin star experience. Nothing outstanding.
That is the problem with selective admission. The problem isn’t that AI writing is bad. In many cases, this is not the case. The problem is that it’s too well tuned to what already exists. It removes the strangeness, the hesitation, the asymmetry, the bit of personal phrasing, the context that only this student could have created. Doing so can improve your writing while weakening the person behind it.
This is also why human mentors remain important. Human mentors do not have the same universal scope as large language models. This restriction can actually be useful. A good mentor doesn’t create endless options. Good teachers listen, interrupt, ask questions, notice, and sometimes push students back into an uncomfortable state of thought. A mentor’s job is not to make your essay look like the best one. It’s about helping students find the truest version of their story.
If AI is used too early, it can interfere with that process.
An extreme example is a student opening an AI chat window before starting a brainstorm. They ask for essay ideas, possible structures, opening lines, themes, and drafts. Their hearts have not experienced the sorrow of creation. I’m not looking for options, testing arguments, or struggling to find the right form for the story. As a result, nothing new is often created.
Language models fill that gap. Predict likely possibilities. Choose the most suitable pattern. The next word is in a sense a prediction, so the essay begins to move towards the likely rather than the characteristic.
At the other end of the spectrum are students who construct their own arguments. they think. They get stuck. They tried one version and rejected it. They return to an experience and realize that their initial interpretation was too easy. They realize that the essay is actually not about the activity they had in mind, but about how they see the world.
That struggle is important.
There are moments in the writing process when students feel stuck and ask for help. Requests such as “I’m just asking for validation from the AI; the core idea is mine” often sound innocuous. But this is often the exact moment when the brain is building up creative tension. If AI enters the picture too soon, that tension will collapse. The student receives the answer before the act of thinking is complete.
Similar to the marshmallow experiment, students may need to delay “Language Model Assistance” until they have completed the brain-training parts, such as brainstorming, story construction, and first draft. When you follow it to the end, it forces your mind to create options. Those choices are the cause of the discrepancy. And divergence is central to creativity.
AI as a fine tuner
This doesn’t mean AI doesn’t have a role. That’s right. Once students have written a complete first draft, AI can help improve clarity, tighten language, point out repetitions, and suggest places where the argument is unclear. At that stage, you can add richness to the work without inheriting its origins. If AI is used late, it can amplify the essay. If used early, it can replace the thoughts that were supposed to create the essay.
This isn’t just a college essay question. It’s a matter of learning. The mind will not learn the first action if it is always outsourced. Don’t learn to sit in confusion. If you don’t see anything, it won’t learn to create options. It does not learn to build arguments from within itself.
Writing has always been a way to visualize your thoughts. The first draft is more than just a communication artifact. It is a record of how the mind approaches a problem. It shows what students noticed, ignored, exaggerated, misunderstood, and gradually understood. If AI steps in before this process takes place, the draft may look better, but the idea behind it may be diluted.
This is why I worry about early AI uses more than incomplete sentences. Incomplete writing can be improved. But if a student doesn’t have trouble finding ideas, the essay may never be completely hers.
Exactly how AI improves quality while preserving voice requires more rigorous study. You need to understand how the timing, usage, and intensity of AI support impact originality. But from what I’ve seen so far, the general direction is clear.
The danger is not that students use AI to write poorly. The danger is that we use AI too quickly and end up creating sentences that are smooth, acceptable, and memorable.
But the greater danger may lie elsewhere. The reason is that the very thinking muscles themselves are being used less and less. If students repeatedly outsource their early struggles, such as exploring ideas, the discomfort of not knowing what to say, and the act of constructing an argument from within, they may not only lose voice in their essays. You may lose the habit of making your distinctive voice in the first place.
(The author is the founder of ACadru, an interdisciplinary learning platform for high school and university students)
