Will AI Help or Hurt Your College Essay?

In order to understand whether AI will help or hurt your essay, we first need to define what specifically makes a winning essay. That way, we know what we're optimizing for and can assess whether AI helps or hurts. Luckily, admissions officers have already answered this question for us:

If you pay close attention to the words that these admissions officers use to describe winning essays, you'll notice a pattern: "who you are", "unique voice", "meaningful to you". You'll also notice other words that are either conspicuously absent or explicitly less emphasized: "well written".

The one important characteristic that winning essays have in common is realness. They reflect who the applicant genuinely is, and help the admissions staff understand how that person will fit into the campus community. Good writing is nice, but it is not the ultimate goal. Admissions officers know this, and are very good at discerning when a well written essay lacks realness.

So, will AI help you write a "real" essay?

When an LLM like ChatGPT or Claude writes something, it uses a set of inputs and rules to choose its words.

First, every word is assigned a score based on all of the words that preceded it (your prompt, what it's already written, any context you've given it, etc). These scores come from training. During training, the model read billions of text samples and learned statistical associations. For example, after the phrase "my sick Grandpa had spent the week \_\_\_\_," the words "resting at home" are far more likely to follow than "training for a UFC fight". Therefore, those words get a higher score.

Then, through a process called Reinforcement Learning from Human Feedback (RLHF), models like ChatGPT are refined by having human raters score their outputs on qualities like "helpfulness" and "quality." Over millions of these rating cycles, the model learns to produce text that the widest possible audience would rate as "good writing" (engaging, interesting, literary, etc). At this step, ChatGPT might decide that "bedridden" is a more specific, vivid word to finish the sentence about Grandpa than "resting at home".

Finally, the output is shaped by a setting called "temperature," which controls how much randomness the model introduces when selecting words. At low temperature, the model almost always picks the most statistically probable next word, producing text that is coherent, predictable, and bland. At high temperature, it's more willing to reach further down its probability rankings and pick a less likely word, producing text that sounds more creative and surprising ("My grandfather's body had become a kind of anchor, pinning him to the mattress while his mind still wandered the garden outside his window").

But here's the critical nuance: high temperature doesn't give the model new information, new experiences, or new ideas. It's still choosing from the same set of statistical associations it learned during training. It's the equivalent of shuffling a deck of cards. You'll get a different hand every time, but you're still drawing from the same 52 cards. None of them are really yours.

LLMs have been trained, at a fundamental level, to write the kinds of things that other people tend to write and approve of. If you ask it to generate writing about an experience at summer camp, it'll write about the statistically average experience at summer camp, with some temperature "flair" to fool you into thinking it's unique and interesting, which is exactly the essay that hundreds of other applicants also might be submitting, and exactly the essay admissions officers have learned to recognize.

AI is an excellent tool for writing and sharpening thinking in domains where convention, clarity, mass appeal, synthesis, or factual depth are the goals. Examples include professional emails, legal copy, marketing copy, technical documentation, research reports, strategy briefs, etc. because there are established patterns for success in each of those domains that can be captured by the training data and optimized for by the models.

AI can also be a helpful tool for writing the admissions essay if it's used in a way that supports rather than replaces the process of looking inward and expressing what feels true and meaningful. For example, applicants can prompt AI for reflective journaling prompts and ask for guidance when they get stuck. LLMs are excellent at generating probing follow-up questions because that's a well-established pattern. The answers, though, come from you, which is where the realness lives.

But for the reasons outlined above, applicants should not hand AI the pen. The undergraduate admissions essay is a unique form of writing where the entire purpose is to reveal an individual human being. It is, by design, the one part of your application that should not be conventionalized, because the more it sounds like it could have been written by anyone, the less likely it is to result in a successful admit decision.