In February 2025, I wrote a Harvard Magazine column about my time as a tutor at the Harvard Writing Center, arguing that while AI tools should not substitute a solid writing education, they could be useful in helping accomplished writers refine their thinking. I’ve changed my mind.
In August, amid a large-scale restructuring of the Faculty of Arts and Sciences, Harvard closed the undergraduate Writing Center’s main office space and laid off its longtime director, Jane Rosenzweig, a skeptic of AI who consistently voiced concerns about outsourcing thinking to a chatbot. A few weeks later, Danoff dean of Harvard College David Deming made headlines for recommending “acceptance, or even encouragement” of AI in writing-based courses. When Tara Malhotra ’29, a current writing tutor, lamented the loss of the Writing Center in the Harvard Crimson recently, the website—in a poetic but likely unintended irony—punctuated her column with ads for Google Gemini: “Free for College Students.”
Now, the future of writing tutoring at Harvard seems uncertain. While the College says some of the resources offered by the Writing Center will be available through other programs, I imagine some students may give up on human tutoring altogether. According to a 2026 Gallup poll, 54 percent of U.S. college students say they use AI regularly for “editing or improving [their] writing”—more than for ideating (49 percent), doing research for papers (45 percent), and writing papers outright (36 percent).
During the two years when I was an undergraduate Writing Center tutor, the staff talked often about AI, in both existential and practical terms, and what to do if we suspected we were encountering it. Largely, we continued to engage the students suspected of using AI in their writing, hoping that through our sessions we could push them to think more critically about their work. My discussions with Rosenzweig—who, though skeptical of AI, always listened to her students’ varying opinions—inspired me to write my Harvard Magazine column, in which I argued that “AI has the potential to make good writers even better.”
At the time, I believed in the ability to use AI responsibly, primarily in an editing capacity. I felt confident that “I know what original, succinct, thoughtful writing looks like.” But this is becoming harder to discern as AI floods the internet. A recent Pew study shows that 35 percent of webpages created after 2022 include AI-generated text. The advent of LLMs is also leading to more homogeneity across all written text, human and AI alike, further diminishing the prominence of expressive, original writing. I’m increasingly worried about my own ability to distinguish thoughtful writing from the declaratory AI slop that pretends to say something profound but lacks real insight.
What is college writing for, if not cultivating students’ insight? Under Rosenzweig’s guidance, we were trained not to tamper with text, but instead to guide and shape ideas, to help someone searching for a compelling throughline in a sea of disjointed interestings. When students came into the center, we never read their essays ahead of time, and we never read the texts that they were referencing, even if they asked us to. We trusted that their interpretation of the source material was valid, which left us well-positioned to comment on the logic and structure of the argument, the backbone of good writing. It also meant that the intellectual work of the final essay was fully the students’ own. Often, we’d look closely at a quote they’d selected, then to their analysis. Does the evidence support the reasoning?
This was frustrating, tedious work. I almost never had a session that was shorter than the allotted hour. Many papers would come to the writing center polished and leave in pieces, but with a stronger argumentative throughline. Theoretically, AI could do this. But it is not intrinsically invested in the kind of feedback that fosters understanding, mastery, and ultimately conviction.
Rosenzweig, who studied AI and its effect on writing even before ChatGPT was made publicly available, presciently argued in a 2022 Boston Globe op-ed that the most worrisome output of AI writing is “that we will lose our commitment to the idea that we ought to believe what we say and write.”
The patient human tutor, both pedagogically trained and attuned to the allure of LLMs, can bolster a student without looking at their half-baked mess and saying, as a chatbot might, “You make a fascinating point—it’s not an essay, but a quiet exploration that offers genuine insight!” Best of all, when face-to-face with a peer tutor, a student can gain the confidence to push back, to determine that, no, they like their style and will go their own way. AI’s authority is supposedly predicated on the totality of knowledge available on the internet, and so it is hard not to trust its wisdom, at least with well-trod topics like how to write an essay. By contrast, we accept that humans are fallible, and that different humans have different ideas. There is no one way to make an argument.
And that was another remarkable feature of the Writing Center. Sometimes students would resist my feedback, insisting that their structure would make sense once they added a bit more reasoning. In doing so, they were often pushed to articulate their thinking more clearly, and to affirm their confidence in their position.
An LLM cannot do that, because it does not have its own opinions or thoughts. It is only trained to engage the user: the more you use it, the more you use it. That’s actually the antithesis of writing instruction, which aims to cultivate independent thought.
AI encourages students with endless platitudes, luring the would-be-writer with an illusion of assistance to degrade the inimitable self that makes their writing interesting. AI trains us, more than anything, to parrot the AI voice. I’d rather we teach students how to find their own.
Serena Jampel ’25 is a former Berta Greenwald Ledecky Fellow at Harvard Magazine.
