What does ChatGPT mean for application essays and college admissions?
View from a guesthouse in Chennai, India
This post was written by me, Kevin Martin, Flesh-and-Blood Human, with minor AI assistance in mid-December 2022
Two weeks into a three-week stay at Bali Silent Retreat, I woke up uncharacteristically early and indulged in phone scrolling before the sun rose. It floored me when I read The Atlantic declaring The End of High School English and that The College Essay Is Dead. Daniel Herman writes in the former article that ChatGPT "may signal the end of writing assignments altogether." At the same time, Stephen Marche forecasts in the latter that the "entire tradition [of undergraduate essays] is about to be disrupted from the ground up."
On my silent retreat, I had spent a collective 90 minutes on my phone the previous week. That morning and afternoon, I scrolled for over six hours that morning and afternoon, trying to digest the development and implications of ChatGPT. Despondency replaced my tranquil days of yoga and homegrown vegetarian food with fears about technological disruption. I wanted to run into the common's area and scream, "It's here! Our Skynet overlords!" I fired off emails from my computer to tech-savvy former students and colleagues. I implored, feeling helpless, "Is ChatGPT as disruptive and world-transforming as it seems?"
Composing this blog post helped me make sense of and find a middle ground to what could be a dawning fundamental juncture in human history. If you haven't heard about ChatGPT yet, you will soon, either because of your child's education or your profession.
I've dedicated the totality of my high school and college students and professional life to honing the craft of writing. I estimate that I have over 20,000 hours of disciplined practice and experience in communication, broadly defined. I'm grateful that my life is such that I cannot imagine a hypothetically more optimal one, and, initially, ChatGPT felt like an existential threat to what I've created and an otherwise promising future. A machine that can write a convincing Shakespearean sonnet in Gen-Z slang from the perspective of a raccoon seemed to threaten the core of my identity and career.
I'm a trash panda, a punk kid on the prowl,
Sneaking through the night, I'm on the hunt,
I'll take what I want, and leave you scowling,
'Cause I don't give a ****, I'm not your average runt.
For I am a raccoon, and I do not fear,
The humans who would keep me from my prize,
I'll take what I want, and leave them in the rear,
As I make my way back to my hidden guise.
So humans, beware, when you leave your trash,
For I am the raccoon, and I'll have my stash!
What is ChatGPT?
ChatGPT explains itself as if to a five-year-old:
“ChatGPT is a kind of computer program that can talk to you and understand what you are saying. It uses special rules and tricks to figure out what you are talking about and then come up with a response that sounds like something a person would say. ChatGPT is like a very smart robot that you can talk to, and it can learn new things and get better at talking the more you talk to it. You can even ask it questions, and it will try to give you an answer that makes sense.”
You will understand ChatGPT better if you spend half an hour submitting queries and requesting the AI to improve upon them. That raccoon sonnet took five minutes and a few iterations to generate. I've also tinkered with it to produce college essay samples that I will discuss later. The New York Times's Cade Metz explains the mechanics of ChatGPT and similar projects to a lay audience. Here is a detailed, critical TechCrunch article detailing upsides, downsides, and limitations.
While The Atlantic's Herman forecasts the End of English, the implications are wide-ranging. Initial anecdotes on Reddit report high school, college, and graduate students solving myriad problems thought reserved only for human creativity, critical thinking, and making connections across different domains.
The Guardian writes, "AI bot ChatGPT stuns academics with essay-writing skills and usability." News articles and blog posts on reputable outlets written by ChatGPT are already appearing, indistinguishable from human-generated content.
High school and college students are already using it to submit assignments across many subjects, including math, computer science, and the humanities. One physics professor reports it scores perfectly on an exam after some tinkering. It can write decent poems (1, 2) and communicate in many major languages, including Danish, despite denying it can. It can score well on AP free response questions, build a virtual machine, and pass tech corporate HR initial screening processes. Say goodbye to cover letters.
It is "scary good," saving time on complex projects by orders of magnitude relative to current AI solutions. One Redditor declares it "the single best learning tool I've ever encountered," and another reports it's exceptional at producing flashcards for study. Teachers use it to make customized lesson plans on specific topics that outperform their efforts and presumably decrease paperwork loads for submission to their administrations.
It can yield answers faster, more accurate, and less spammy than Google or Wikipedia, although there are occasional significant factual errors and issues with systemic bias. Stack Overflow has already banned submissions from ChatGPT due to repeated inaccuracies. Ian Bogost at The Atlantic cautions that the hype surrounding ChatGPT is "misplaced" and "dumber than you think."
He continues, "ChatGPT lacks the ability to truly understand the complexity of human language and conversation. It is simply trained to generate words based on a given input, but it does not have the ability to truly comprehend the meaning behind those words. This means that any responses it generates are likely to be shallow and lacking in depth and insight." It cannot write compositions anywhere near the nuance, quality, and accuracy as The Atlantic or even free-form blog posts like this.
Nevertheless, it has the potential to disrupt myriad industries, including law, computer science, media, and legions of content marketing freelance writers. (I felt insecure about this sentence and cross-checked with ChatGPT, who, after disassembling the grammatical constructions better than I could, reassured me, "The sentence is well-constructed and easy to understand." Whew!)
Virtually anyone participating in, is studying to join, or has a career that depends on the knowledge economy – hundreds of millions of people worldwide – may be impacted by ChatGPT and its subsequent iterations and related projects.
ChatGPT and College Admissions
For this post, I'm obviously most interested in what ChatGPT means for college essays, the admissions process, and education more generally.
Suppose AI can generate passable and occasionally excellent writing samples. What does this mean for applicant essay submissions, the humans who review them, and the institutions that set academic standards and award degrees?
Conversations around potential plagiarism and academic honesty are already occurring. Some students report completing homework in minutes that might have taken them a week. Another Redditor submitted more than a hundred scholarship applications after scraping the portals and inputting the ChatGPT-generated essays, although the comment eludes me now. Consequently, combining ChatGPT with free or inexpensive off-the-shelf products can produce novel or unexpected outcomes.
Essay submissions are central to grading in high school and college disciplines. If ChatGPT can write a passable essay comparing and contrasting the political theory of Thomas Hobbes and John Locke, what does that mean for the kinds of political theory classes I took at UT? Here is a sample response from ChatGPT:
Next, I asked what would happen to college admissions. Its answer reassured me:
It repeated the same answer when asked the second time.
I tried a follow-up:
When I pressed it by modifying "every" to "some," it reasserted its original conclusion. "The essays generated by ChatGPT are unlikely to be as well-written or well-organized as those written by human applicants…. it is unlikely that using essays generated by ChatGPT would improve an applicant's chances of being admitted to a university."
I think ChatGPT overestimates human writing ability, but at least the AI is charitable to its subjects. That aside, It followed up another query with: "Yes, I believe that humans can distinguish essays written by ChatGPT from those written by other humans."
Its assertion gave me food for thought and questioned the implicit assumptions about whether ChatGPT is the end of high school English or the undergraduate essay. One limitation of ChatGPT is that it has a narrow range of possible responses to a given topic. I imagine that if even two students in a single humanities high school or college class submitted ChatGPT variants, the professor could immediately tell by comparing them. Like in the Hobbes/Locke response, even when you prompt it to use specific quotations, it yields similar answers.
I'm optimistic that ChatGPT, in its present iteration, will not undermine essay submission and grading. Herman's claim that English is Over seems reactionary and overblown. It's more like Cliffnotes on steroids than a Turing Test-capable machine capable of thousands of distinct Hobbes/Locke responses. For humanities and social science prompts that could produce hundreds of potential responses, or for AP Exam Document-Based Questions that require on-the-fly synthesis, ChatGPT doesn't seem yet capable of fooling a professor if more than one student submitted a ChatGPT submission. When ChatGPT begins writing Haruki Murakami or Margaret Atwood novels indistinguishable from the real things, we should worry.
I tested it with queries about German philosopher Friedrich Nietzsche of whom I've read the complete works and most of his major works multiple times. Like technical experts reporting inaccuracies, it provided some insight into Nietzsche while obscuring other critical aspects or committing outright fabrications. At first glance, the responses seem sound. However, upon further scrutiny, particularly for a writer like Nietzsche who communicates mainly through parables, riddles, wordplay, and metaphors, ChatGPT produces responses that wouldn't pass a professor's sniff test. It seems better suited for technical writing or responding to queries with a single, knowable, concrete answer, like coding, math, or physics.
In my recent book Surviving the College Admissions Madness, I nixed a proposed chapter on technological disruption. There didn't seem to be much there "there" yet. I can imagine a universe where General Artificial Intelligence or reaching the Singularity will make obsolete human readers, student submissions, admissions committees, or perhaps all of education and the knowledge economy. But we're not automated into oblivion yet.
ChatGPT is the first time I've raised my eyebrows at a technological development with potentially sweeping implications for college admissions and higher education. I figured these developments for five or ten years away, so I admitted it from my book, but I can chalk that up to my laity.
A former transfer client and current UW-Seattle CS student, whom I also assisted with his graduate application, seemed more bearish on current capabilities and future developments. "There are fundamental limitations not apparent in the superficial displays of ability that ChatGPT and generative AI models seem to garner among popular media." He also expects plagiarism tools and counter-AI to catch up quickly to snag ChatGPT submissions, particularly as people continue sharing ChatGPT examples online.
He raised analogies that rang true. Following exponential growth and subsequent hype in quantum computing and self-driving cars, current technology has seemed to hit a point of diminishing returns. Self-driving trucks excel on interstate highways yet can't navigate complex urban traffic reliably. If an engineering manual is the NLP equivalent of a straight highway, college essays are the wrong ways down oneway streets in unfamiliar downtowns. Despite being an early adopter and long-time hodler of cryptocurrency, I’m also increasingly skeptical if there are seismically disruptive use cases for blockchain technology. Consider that few people use Alexa or Siri voice recognition software beyond basic queries about music or weather rather than a HAL 3000 intelligent assistant initially envisioned by Bezos.
Will Natural Language Processing software utilizing Large Language Models like ChatGPT be like self-driving cars – a promising yet limited tool – or the harbinger of a genuinely earth-shattering transformation of humanity? At the moment, it seems better suited as a study aid or assistant rather than a substitute for human efforts. Nevertheless, it's still too early to tell, but we can't afford not to pay attention.
What does ChatGPT mean for brainstorming, writing, revising, proofreading, and submitting college essays?
Reddit's Applying to College has already caught on to the potential implications by asking whether the AI can produce better essays than them. Predictably though, the essays A2C thinks are amazing, admissions counselors or professors chimed in that those submissions are vague, platitudinous, and cliched.
As an aside, I double-checked that platitudes is a word, and ChatGPT provided an immediate and more thorough answer with a redacted example sentence than if I consulted dictionary.com. Then I wondered if it was redundant to use platitude and cliché in the same sentence. ChatGPT suggested it "is generally not a good idea… but they have slightly different connotations." Maybe I could have substituted with the more common-use trite, but whatever; I'll let the redundancies live. Along with that semicolon that ChatGPT protested.
I've already used ChatGPT in my revision work to navigate marginal grammar cases and precisely when you should use "of" following "myriad" (answer: only when myriad functions as a noun and not an adjective, apparently.)
What's undeniable is that ChatGPT is excellent at producing first drafts, but an initial draft is only one step among many toward an eventual submission.
I've experimented with plugging in resume bullet points from an activity, and it spits out decent essays about extracurriculars, volunteering, sports, and so on. If you prompt it to exclude any fabrications or go beyond the inputs, it provides an accurate summary suitable for supplemental questions like "why do you want to study your first choice major?" or "describe an extracurricular that is meaningful to you."
I plugged in various scenarios like "write a 650-word essay about a moment of realization at my rural farm home following a harvester accident that impeded my psychology ambitions yet offered perspectives on finally pausing school and joining a traveling circus, written in a lighthearted tone, using apples as a metaphor for growth." Plug it into ChatGPT yourself to see the response. Submit follow-ups to see how and whether you can improve the overall quality.
I queried it to improve certain aspects like the introduction or selected sentences to modify semantics or syntactic nuance. It produced a solid essay that would easily pass for a top 10-20% submission among a hundred randomly selected Common Application essays in a reviewer's portal. As much as I've tried with this and other examples, I can't prompt it to elevate a top 10% essay or into the to top 1% or one of the rarified one-in-a-thousand essays bound to floor your reviewer and get you into your dream school despite very below average academics. First drafts often require disproportionate time, so I can envision a role where ChatGPT helps students to get the ball rolling.
For the moment, the best ChatGPT submission is far inferior to the least bad submissions of my clients'. You can input ChatGPT queries and compare them with client essay examples from a few years ago – my client submissions have only improved since I published that post.
An obvious ChatGPT limitation for a memoir or autobiographical writing is that the machine doesn't and can't ever know you. Whatever inputs you supply can't capture nuance. The submissions are vague, bland, and filled with cliché, although that isn't much different from the typical college essay submission, which isn't very good. All of its responses are also bland with predictable, despite flawless sentence structures. It's free of any head scratchers common to previous chatbots. Curiously, it never uses punctuation besides periods and commas.
The :why are you applying to X university" ones are the worst. At first glance, they seem compelling with sentences like "As a lifelong resident of Texas, I have always known that UT Austin is one of the top universities in the state, and I am excited about the opportunity to become a part of the vibrant and diverse community on campus."
Inserting other universities provides a similar submission. It’s all filler and fluff. At best, it provides a template for you to build around or substitute vague assertions with your personal experiences and specific resources, extracurriculars, courses, research opportunities, and so on that you'd like to pursue. But that isn't different from the legions of essay examples already available online. I anticipate plagiarism software to quickly identify and pick up on these once ChatGPT examples become ubiquitous online. The students most likely to benefit are international applicants for whom English is not their first language or especially lazy STEM students who loathe writing and are willing to risk a plagiarism check.
Arguably the most important aspect of college essay-building involves brainstorming. Often, the best client essays emerge from ideas they hadn't considered, or they hint at something toward the end of our Zoom call, "I don't know if this would work, but…" before proceeding to share something super interesting or quirky. Ideas omitted cannot necessarily be inputted into ChatGPT. I inserted some of my former client brainstormings into the software, and again it produced solid, if unexceptional and bland responses. There remains a role for an experienced educator to help students navigate between topics.
I can imagine software that scrapes every "50 Essays that got into Harvard" and equivalent thousands of college essay examples paired with a questionnaire and ChatGPT that might be able to generate a top-5% or even top-tier essays, but I haven't seen this yet. I feel it'll suffer from the same input limitations as even a steroids version of ChatGPT
Moreover, although ChatGPT can improve sentences and paragraph quality to some degree and help evaluate word selection, it cannot big picture review or edit for structure. Should you switch the second and third paragraphs, for example, or remove the fourth and combine a chunk of it with the introduction? ChatGPT demurs on most of these queries.
NLP proofreading software already exists, Grammarly. The UW student asserts that it's the most effective NLP use case, more helpful and accurate than voice assistants or customer service chatbots. I've used it for over six years and can testify that it has improved significantly over the years. I'm currently rewriting my Your Ticket book updated for the third edition. Grammarly has caused me a lot of insecurity about how many tiny, edge-case grammar or sentence structure instances it has flagged despite thoroughly writing, editing, and Grammarly proofreading it twice before. Proofreading remains outside the scope of ChatGPT. Standing alone, Grammarly isn't especially helpful for building college essays other than taking an essay 95% complete and taking it over the finish line.
Finally, perhaps the most overlooked aspect of college essays are submitting them. A pernicious condition inflicts many of my clients, overediting. They hesitate to press submit, agonize over minor punctuation concerns, and feel analysis paralysis about the overall quality. Some clients delay for days, weeks, and sometimes months before hitting submit. I have an extensive PDF dedicated to the topic to try and get ahead of these tendencies.
ChatGPT is certain to amplify overediting because, should a determined student wish to input ChatGPT for their Common App essay, they're likely to have tens of versions. Which one should they select as the final, final one? ChatGPT cannot tell you. That requires the discretion of someone like me with extensive editing and admissions experience or a trusted education professional to help brainstorm, process drafts, edit, and assess for overall quality.
Overall, ChatGPT seems like a useful tool, but it cannot substitute for a student's effort or room for the assistance of qualified humans. The closest analogy for a disrupted profession seems to come from professional translators who use AI to complement their efforts rather than a substitution for human judgment and discretion.
My former client at UW reassures me, "For your business, I don't think that ChatGPT will be able to tailor essays to accurately hit the points that college admissions offices want as well as you can. The ChatGPT prose is really good, but I don't know if it can go beyond answering the prompt. I believe your services go beyond answering the prompt and successfully induce the reader's sense of immersion in the essay in such a relatively small number of words."
To plug my own value-added and why we should work together. Essay building tends to be a relatively small part of how I assist. Most families hire me superficially to help maximize their admissions chances and address admissions questions. They learn quickly that I also help juggle and assign tasks, identify which essays or applications to do in what order, and minimize wasted time. I help alleviate inevitable stresses, build a reasonable college list, set realistic admissions expectations, and anticipate any issues. Sometimes, a worried parent or child at 2am just needs to be told that they're doing great, their essays are perfectly fine, and everything will be okay.
An AI can never provide these soft benefits testified by my former client testimonials. College applications and the admissions process will always be stressful for applicants to UT and top-100 universities, even if essays become marginally easier to craft. As long as universities receive record application numbers and artificially limit enrollment spaces, services like mine will likely remain. Even if college admissions was fully automated without human reviewers or even students needing to submit applications – like a totally unified and centralized database – there will still be independent consultants to help families navigate the gatekeepers. Admissions staff are likely to lose their jobs to automation before independent consultants. However, ChatGPT may push out newcomers or inferior consulting options.
Still, our tech overlords have put me on notice. I will continue to rethink my career trajectory, what services I offer, and to whom.
A summary of what ChatGPT cannot (yet) due regarding college essays
Discover which topics you could potentially write about and which essay topics your content best matches
Identify the optimal balance or weight to give to some experiences over others within an essay
Provide attention-grabbing introductions that it weaves throughout the essay that connects to a compelling conclusion
Adequately supply metaphors and symbols, which are a hallmark of exceptional essays
Connect essays with one another to form an overall content portrait that details your competencies and interests
Distribute your rough draft content among the varied long and supplemental essays
Although you can ask it to provide humor, pessimism, lightheartedness, anguish, and so on, the emotional valence feels forced or not quite right. In short, in no way does it write like a teenager in a semi-formal tone
Choose among the best alternative samples for which single essay to submit. If you have 20 ChatGPT essay options, how will you know which is optimized for your admissions audience?
Although it can diversify syntax if you ask, the sentence structures tend to be formulaic and relatively static
It summarizes and combines distinct content well, particularly for extracurricular or major-specific experiences, but it struggles with nuance
Provide sources for its claims for fact-check assertions for "issue of importance" type essays
Going beyond the prompt to connect with readers on an emotional level
It makes stuff up unless you give it explicit instructions not to unless you supply it with precise details of what to include