For the thousands of first-generation, low-income (FGLI) students navigating the labyrinthine process of applying to college, the admissions essay has historically been a tool of agency. It is the one place where a student can humanize their data, contextualize their struggles, and present a narrative that transcends their GPA or standardized test scores. However, in the 2025–2026 admissions cycle, a new, cold phantom has entered the room: the specter of Artificial Intelligence. As an educator who works exclusively with FGLI applicants, I have watched the anxieties of my students shift. Once dominated by fears surrounding the Supreme Court’s 2023 affirmative action ruling or the rise of "trauma-dumping" in personal statements, the current discourse is now defined by a single, pervasive question: How do I avoid sounding like a machine? The New Admissions Anxiety: Decoding the "AI Literacy" Expectation The anxiety stems from a profound lack of transparency. The Common App, the primary portal for millions of applicants, explicitly classifies AI-generated content as a form of fraud. Yet, the institutions receiving these applications offer no clear roadmap regarding how they detect such content. This creates a high-stakes guessing game where students are forced to study the specific prose style of Large Language Models (LLMs) not to use them, but to ensure they don’t accidentally mimic them. This "AI radar" has become an invisible, subjective gatekeeper. Students are now obsessively editing their work to avoid the "hallmarks" of generative AI—the overly polished syntax, the rhythmic but sterile transitions, and the generic, platitudinous moralizing that often defines chatbot output. They are editing for their own humanity, terrified that an automated detection system will flag their genuine voice as synthetic, or worse, that an admissions officer will find their prose "too perfect" and assume a shortcut was taken. Chronology of a Digital Divide The integration of AI into the educational landscape did not happen in a vacuum, but the speed of its adoption has created a stark, tiered reality: Pre-2023: Admissions essays were primarily measured against traditional metrics of writing ability, authenticity, and narrative arc. 2023–2024: The public release of advanced generative AI tools sparked immediate alarm. Colleges scrambled to update academic integrity policies, often conflating "using AI to brainstorm" with "using AI to cheat." 2025: The "AI Literacy" mandate emerged. Some universities began explicitly inviting students to use AI to "support their original ideas," while others maintained zero-tolerance policies. The Current Landscape: We have arrived at a fragmented reality. A student applying to five different schools faces five different sets of AI policies. At one school, using Grammarly might be encouraged; at another, it could be a trigger for a deeper, potentially biased investigation into the student’s writing process. The Socioeconomic Infrastructure of AI Proficiency The core issue is not just the technology itself, but the pedagogical infrastructure surrounding it. We have long recognized that the admissions process favors the wealthy—SAT prep courses, private college counselors, and access to elite extracurriculars like robotics camps have always acted as force multipliers for affluent applicants. AI literacy is simply the newest, and perhaps most insidious, iteration of this advantage. In affluent, well-resourced school districts, AI is being integrated into the curriculum. Students are taught prompt engineering, the ethics of algorithmic bias, and how to use AI as a collaborative partner. They have teachers who can review their work and advise them on how to navigate the "gray areas" of these policies. Conversely, in underfunded, rural, or urban school districts, students are often left to navigate AI entirely on their own. They lack the institutional guidance to differentiate between "AI-assisted brainstorming" and "AI-generated prose." A December 2025 study by the College Board, surveying over 10,000 stakeholders, confirmed this disparity: private school teachers report significantly higher access to generative AI tools and are more than twice as likely to work in environments with clear, supportive policies governing their use. Supporting Data and Systemic Inequality The disparity in oversight is glaring. I recently worked with a student who submitted a draft that, to any trained eye, clearly bore the stylistic fingerprints of Claude. When I questioned him, he was shocked. "My counselor approved this," he told me. How could a professional miss the hallmarks of AI? The answer lies in the systemic neglect of public education. In many of the districts where I work, guidance counselors are managing caseloads exceeding 700 students. They do not have the time—or in many cases, the training—to act as an AI editor. Meanwhile, a counselor at a private prep school with a 1:50 ratio can provide granular, iterative feedback on a student’s ethical use of technology. This gap is not merely about who has a ChatGPT Plus subscription; it is about "cultural osmosis." Wealthy students learn to navigate these tools through guided instruction, while FGLI students often learn via trial and error. When that trial results in a rejection letter—often without an explanation or a path for appeal—the student is effectively culled from the process, and the cycle of inequality is reinforced. The Myth of Detection and the Bias of "Human" Review Colleges often lean on the perceived reliability of detection software, yet the science remains deeply flawed. Detection tools are notoriously inaccurate and have been shown to exhibit systemic bias against non-native English speakers, often flagging their unique syntax or phrasing as "AI-generated." When human officers take over, the process becomes even more subjective. Admissions officers are rarely trained in digital forensics. Instead, they rely on their own internal, biased metrics of what "sounds like a student." If an admissions officer holds a prejudiced view of how a student from a specific background should write, they are more likely to view deviations from that norm—whether AI-assisted or not—with suspicion. Because colleges are not required to disclose their detection methodologies, these biases remain unchecked. There is no transparency regarding: The software used: Are they relying on Turnitin, GPTZero, or proprietary, in-house tools? Data training: What datasets are being used to "teach" these systems to spot AI? The "Human-in-the-Loop" standard: How much weight does an automated flag carry in the final decision-making process? Implications for the Future of Holistic Review The promise of "holistic review" in admissions was that it would allow for a more nuanced understanding of the whole person. However, if that review is now filtered through opaque, unreliable, and biased AI-detection mechanisms, the holistic process is dead. We have replaced the "SAT gap" with an "AI-literacy gap," and the results will be a less diverse, less representative student body. If we continue down this path, the admissions essay—once a platform for the underprivileged to speak their truth—will become a minefield. Students will play it safe, stripping their writing of the very vulnerability and unique perspective that makes them compelling candidates. They will produce sterile, "human-sounding" prose, not because it is better, but because it is safer. A Call for Transparency Colleges demand that students be transparent about their writing process, yet they operate in the shadows regarding their own. It is time for a radical shift in accountability. Institutions must: Publish Clear Policies: Stop hiding behind vague "academic integrity" statements. Colleges should explicitly state if and how AI-detection software is used. Disclose Detection Rates: Universities should release data on how often applications are flagged and the demographic breakdown of those flags. Standardize Training: If admissions officers are to act as arbiters of "human vs. AI," they must be trained in the limitations of these tools to prevent systemic bias against non-native speakers and students from under-resourced districts. Adopt a "Human-First" Policy: If an application is flagged, there must be a transparent, human-led review process that allows the student to explain their writing process before a decision is made. If a college cannot prove that its detection methods are equitable, it has no moral authority to use them as a basis for rejection. We are currently asking students to prove their AI literacy in an environment that refuses to demonstrate its own. The digital divide is no longer a future concern; it is the current reality of our admissions system, and without immediate intervention, it will continue to erode the foundations of opportunity for the next generation. Finn Anderson is the director of writing at Leadership Enterprise for a Diverse America (LEDA) and a lecturer in the Undergraduate Writing Program at Columbia University. Post navigation Beyond the Silence: Rethinking Early Expression in Multilingual Classrooms
