AI in College Admissions Research: 4 Major Studies
4 studies on AI in college essays: detection F1=0.998, SES admit gap widened 31%, AI essays lose 22% authenticity, AI writes male and privileged.
4 studies on AI in college essays: detection F1=0.998, SES admit gap widened 31%, AI essays lose 22% authenticity, AI writes male and privileged.
Cornell + Stanford analyzed ~170,000 essays: AI aligns with male, continuing-gen, and high-EC writing 65-92% of the time. GPT-4 is more skewed than GPT-3.5.
Cornell + Stanford analyzed 35,789 Latinx UC applicants: 20% used some Spanish in their essays. Across ~26,000 GPT-3.5 and GPT-4 essays on the same prompts, 0% did.
22% of under-$50K applicants used AI on essays vs. 40% of $75-100K (Foundry10). Lower-SES users are penalized 1.85x more per unit of AI use (Cornell).
Cornell tested 8 LLMs: telling ChatGPT your race or first-gen status doesn't make essays sound like you. For Black applicants, it actively backfired.
Foundry10 surveyed 425 teachers: 31% used AI for college rec letters. They rate their own use as more ethical than students do.
CASPA is the centralized PA school application used by ~90% of US PA programs. $185 first program. Cycle opens April 30. Full timeline, fees, components.
From Harvard's one-word AI policy to a professor who compares ChatGPT to Janet from The Good Place, here are the most striking, funny, and thoughtful things professors have written about AI in their syllabi.
Some professors are rejecting AI detection entirely — citing inaccuracy, surveillance concerns, and trust. We found their actual syllabus language from 210 course policies.
We analyzed the actual language of 210 course-level AI policies across 181 institutions and 75 disciplines. Here's what professors' word choices reveal about how academia really feels about AI.