A groundbreaking study has uncovered a troubling trend in college admissions: low-income students are turning to artificial intelligence (AI) tools like ChatGPT more frequently than their higher-income peers when crafting their personal essays. This shift, analyzed across over 81,000 applications to a selective U.S. university, reveals not just increased AI adoption among those seeking fee waivers—a common proxy for financial need—but also a stark disparity in outcomes. While AI promises to level the playing field by providing accessible writing support, the research suggests it may instead exacerbate inequities, as essays flagged for heavy AI influence fare worse for lower socioeconomic status (SES) applicants.
The findings come at a pivotal moment for higher education admissions. With the end of race-conscious affirmative action and a growing reliance on essays to gauge fit and authenticity, admissions offices nationwide are grappling with how to evaluate writing in an era of generative AI. As applications for the Class of 2030 pour in, this study highlights the urgent need for colleges to adapt their processes to ensure fairness across diverse applicant pools.
Unpacking the Cornell Study: Methodology and Data
Researchers from Cornell University and Carnegie Mellon University, led by Ph.D. candidate Jinsook Lee, examined de-identified essays from five admissions cycles (2020-2024) submitted via the Common Application. The dataset spanned 81,663 essays, focusing on linguistic shifts before and after the November 2022 launch of ChatGPT.
To detect AI influence, the team developed a distribution-based detector. This tool compares an essay's token-level likelihoods against reference distributions: one from pre-2023 human-written essays and another from synthetic essays generated by GPT-4o using Common App prompts. It outputs a mixing proportion (α), where 0 indicates fully human-like writing and higher values signal greater LLM (large language model) involvement.
Socioeconomic status was proxied using fee-waiver status, a reliable indicator of financial hardship approved by the Common App for low-income applicants. Controls included GPA, SAT/ACT scores, first-generation status, school type, and extracurricular honors. Logistic regressions and difference-in-differences models linked AI use to admission probabilities, revealing nuanced patterns stratified by SES.
Key Findings: Surging AI Use Among Low-Income Applicants
Pre-2023, estimated AI use hovered around 0.03-0.05 across groups. By the 2024 cycle, it jumped to 0.08-0.10 overall, with low-SES applicants showing a mean α of 0.102—28% higher than high-SES peers at 0.080. High-use essays (α > 0.13) were 22.7% among low-SES versus 18.7% for high-SES, a statistically significant overrepresentation.
Linguistic analysis showed post-AI convergence: reduced lexical diversity (e.g., lower type-token ratios, MTLD scores), shorter sentences, and simpler complexity—hallmarks of generic AI output. These changes were most pronounced among low-SES writers and ultimately rejected applicants, suggesting overreliance on free-tier tools without polishing.
"High-income students have counselors, teachers, and premium AI subscriptions," Lee noted. "Low-income students might rely solely on basic ChatGPT, yielding poorer results."
The Admissions Penalty: Why AI Hurts Low-SES Students More
Despite similar baseline admission rates favoring high-SES (23.6% vs. 12.9% pre-AI), the gap widened post-2023 to 14 percentage points. Each unit increase in α reduced admission odds by 62% for high-SES but 83% for low-SES—a 1.85 times harsher penalty.
Even after adjusting for credentials and stylometrics (e.g., word length, diversity), the disparity persisted. Mediation analysis indicated surface features partially explain it (7.1% attenuation), but low-SES AI essays often matched or exceeded high-SES in polish—implying bias in perceived authenticity.
AJ Alvero, co-author, warned: "Essays should showcase idiosyncrasies. AI templates erode that, disproportionately harming those without editing support."
For context, a 2024 EdWeek survey found one-third of seniors used AI for essay help, while prior Stanford research linked essay style to household income more than SAT scores.Read the full study here.
Photo by Marcus Ganahl on Unsplash
Resource Gaps Driving AI Reliance
Low-income students often attend under-resourced schools lacking advanced placement courses or writing tutors. National data shows only 40% of low-SES high schools offer robust counseling, versus 80% in affluent areas. AI fills this void, but free versions produce formulaic prose lacking personal voice.
High-SES applicants leverage paid tools (e.g., Claude Pro at $200/month), private coaches ($300/hour), and iterative feedback. A 2025 Kaplan survey revealed 70% of admissions officers suspect AI more in unpolished essays, common among resource-poor applicants.
College Responses: Policies, Detection, and Innovations
U.S. colleges vary widely. Yale, Cornell, and Brown prohibit AI-generated content outright, viewing it as academic dishonesty. Others, like Harvard and Stanford, permit grammar checks but ban full drafts. About 50% use AI for initial screening (e.g., UNC's essay scorer since 2019), flagging for human review.
Detection relies on tools like Turnitin (imperfect, 30% false negatives) and human intuition: AI essays are "generic, emotionally flat." Admissions officers report spotting patterns like repetitive structures or unnatural phrasing.
To counter, video essays are rising—Georgetown, NYU, and 20+ others now accept 1-2 minute clips. These resist AI (deepfakes lag) and reveal personality. Inside Higher Ed covers evolving policies.
Post-Affirmative Action: Essays Under Scrutiny
Since the 2023 Supreme Court ruling, essays bear heavier weight—up 25% in holistic reviews per NACAC. Yet, AI undermines this, homogenizing narratives and amplifying SES biases. Low-SES voices, already underrepresented (Pell Grant admits down 2% at elites), risk further marginalization.
Stakeholder Perspectives: From Students to Admissions Leaders
- Students: Low-SES applicants cite time pressures; 45% in a 2026 ACT report viewed AI positively for equity.
- Admissions Officers: 78% worry about authenticity (Kaplan 2025); many advocate rubrics emphasizing voice over polish.
- Educators: Counselors push workshops on ethical AI use, blending tools with personal reflection.
Solutions: Fostering Authentic, Equitable Admissions
- Transparent policies and AI literacy programs.
- Hybrid prompts rewarding specificity (e.g., "Describe a family tradition").
- Expanded fee waivers, free coaching via platforms like AcademicJobs scholarships.
- Portfolio assessments or interviews for low-SES applicants.
Colleges like MIT test AI-disclosure questions, boosting trust.
Looking Ahead: AI's Evolving Role in Higher Ed
By 2030, 90% of offices may use AI for triage, per projections. Yet, with ethical guidelines (e.g., NACAC's 2026 framework), it could enhance equity—personalizing feedback for all. For now, authenticity reigns: Write your story, revise thoughtfully, and seek human input.
Explore tips for standout applications and test prep resources.