Should AI Read and Assess Essays? — A Virginia Tech Case Study
Admission/enrollment leaders. Should AI be used to read and assess essays? Discuss…
Throughout my years leading Ravenna, a software platform for K-12 admissions, I often looked to higher ed for ideas worth borrowing. I recently saw a presentation about a college that's using AI to review its application essays. I know this is a third-rail topic in our industry, but the case study is one K-12 admissions directors should make time for.
Juan Espinoza, VP for Enrollment Management at Virginia Tech, walked through how his team built an AI tool that helps read the essay portion of their applications. The way they did it matters more than the fact that they did it.
Three things stood out — and to me, these are the three any AI-driven initiative has to get right: human-in-the-loop, guardrails, and transparency.
Every essay is still read by a human alongside the AI. When they rolled out the new system, they actually tightened the disagreement threshold that triggers a third reader, from a 4-point gap down to 2. Final admission decisions remain 100% human. The human role here is load-bearing.
The guardrails were serious. The LLMs run on-premise, trained on seven years of their own admissions data, with no applicant identifying information exposed. When they saw bias in early versions, they moved from a single model to an ensemble of three to triangulate the scoring.
The transparency was proactive. Juan went to the press himself rather than letting the story leak. His framing stayed with me: when a family submits an application, there's an implicit trust that it will be "handled with care" — and he didn't want to violate that trust by hiding AI's role.
The results: roughly 8,000 hours of human reading time saved, decisions to students out several weeks earlier, and — this is the part that surprised me — lower disagreement between AI and human readers than between two human readers.
Worth noting the obvious: Virginia Tech is a research-one university with a faculty linguist and an AI lead embedded in this work from day one. Most K-12 schools don't have that bench. But the principles travel — and higher ed often runs the hard experiments a few years before K-12 catches up.
I led with the third-rail question for a reason: to engage you in thinking about the three elements of any important AI initiative at your school.
Human-in-the-loop. Guardrails. Transparency.
The bar Virginia Tech set isn't about resources — it's about discipline. Any school adopting AI can hold itself to the same three.
If you're thinking through any AI initiative for your school, this case study is worth your time. Thank you to Juan Espinoza for going public with this work, and to Emily Pacheco and the AI in College Admission community for hosting.
