Medical Schools Using AI to Screen Applicants — 2026 List

Medical Schools Using AI to Screen Applicants: The 2026 List

Premed forums are full of the rumor that "AI reads your application now." It's partly true—and the reality is more specific (and more interesting) than the rumor. A small number of medical schools have built machine-learning models that perform the initial screen: the first pass that decides whether a human ever reads your file. Most schools still don't. Nobody maintains a clear, sourced list of who does, so here's ours.

This page tracks US medical schools and selection systems that use AI to screen, score, or triage applicants—with what each system actually does and where it stands. For the harder question—is it fair that schools restrict your AI use while screening you with AI?—see the companion piece on the medical-school AI double standard.

The list at a glance

The honest headline: at the medical-school level, only two programs (NYU Grossman and Zucker) are confirmed live with published validation. The bigger, less-noticed story is in residency selection, where an AAMC-backed AI tool went live for every program in 2025.

The live medical-school systems

NYU Grossman School of Medicine

NYU built a "virtual faculty screener"—a machine-learning model that mimics how a human screener would sort an application into invite for interview, hold for further review, or reject. It was trained on 14,555 applications from 2013–2017 and the decisions of 99 faculty screeners.

In validation, it matched human reviewers reasonably well on the clear calls (area-under-curve around 0.83 for invite and reject) and noticeably worse on the ambiguous "hold" calls (~0.62–0.64). In a 2019 randomized trial of roughly 3,700 applications, the model and the faculty produced no significant difference in overall interview-recommendation rates. The school frames it as a response to workload: manual screening consumed "more than 6,000 hours of faculty time yearly." As the project's lead, Marc Triola, puts it, the model "doesn't get tired, it doesn't get grumpy"—but, he's careful to add, "the algorithm by definition includes the collective biases, flaws, strengths, and performance of the faculty from which it learned." Final interview and admission decisions remain with the human committee. (Source: Triola et al., Academic Medicine, 2023.)

Zucker School of Medicine at Hofstra/Northwell

Zucker's model is the literal first reader of an application's structured sections, recommending interview, reject, or "defer" (low-confidence cases are routed to humans). It narrows roughly 5,000 applications a year down to 1,500–2,000 for committee review. Notably, the system is fed the file with the applicant's name, birthplace, and photo removed—a deliberate bias-reduction step.

Reported accuracy was 88% on the test set (AUC 0.93) for the AI alone, and 96% in the combined human-plus-AI workflow. One number worth knowing: precision was about 0.63, meaning a real share of false positives before the human layer catches them. The published study also flagged equity concerns in its own data (e.g., African American applicants somewhat less likely to be invited), and its authors warned that reducing reviewer variability "does not eliminate systemic bias—it might in fact have the opposite effect." (Source: Keir, Woldenberg et al., JAMIA Open, 2023.)

The ones still in the lab

• George Washington (SMHS) is developing a tool meant to assess the entire application across roughly 13,000 applicants a year; admissions leaders stress it's decision support—"not something that will replace human review." No live deployment or validation is published.
• University of Cincinnati (COM) has published a framework for an essay-screening tool, but it's a conceptual paper with no accuracy results and no deployment. (A separate 2026 grant to Cincinnati is for AI in medical training, not admissions—don't conflate them.)
• UC San Diego appears only as a brief "in discussion" mention with no named system. Treat it as exploratory and unverified.

These four trace back to a single January 2025 AAMC feature; as of the most recent reporting, none of the pilots has been confirmed live.

The bigger story: AI screening went mainstream in residency

While undergrad and med-school coverage argues about two schools, AI screening quietly scaled across residency selection. In July 2025, the AAMC and Thalamus made "Cortex"—an AI-assisted holistic-review tool that normalizes transcripts, letters, and experiences from ERAS applications—free to every residency program, after a 2025 pilot. Roughly 1,500 programs (about 30%) used it in the first cycle, and it reportedly cuts screening time by about half. The AAMC is clear that ERAS itself doesn't use AI to sort or reject; Cortex assists human reviewers.

It hasn't been frictionless: a peer-reviewed audit found persistent errors in Cortex's transcript normalization, which the vendor publicly acknowledged and says it has since corrected (and it's standing up a community oversight board). And research on AI in residency selection has surfaced real warning signs—one 2024 head-to-head found an AI and a program director agreed on only 7.3% of selections, with the AI skewing toward higher-scoring, more-published, no-visa, and more White/Hispanic applicants.

The quiet expansion layer

Watch the application platforms, not just individual schools. In March 2026, Liaison (which runs the centralized application services behind many health professions) launched WebAdMIT Holistic and Predictive Insights, which auto-calculates a predictive "holistic score" per applicant. That puts AI-assisted scoring into the plumbing used across MD, DO, PA, dental, nursing, and pharmacy admissions—the broadest, least-noticed vector for this technology.

What this means for your application

• At most schools, a human still reads you first. Confirmed AI screening is rare (two med schools). Don't write to an algorithm.
• Where AI does screen, it's trained on past human decisions—so the conventional signals (rigor, coherence, evidence of the traits a school values) still matter; there's no secret keyword trick.
• The applicant-side rules still bind you. You're certifying your essays are your own work; for what each service and school actually permits, see what AI you can use on a medical school application, and browse the medical-school AI policy directory. The surest move is still to write and review in your own voice.

The undergraduate version of this shift—schools using AI to read admissions essays—is tracked in which colleges use AI to read essays; the med-school version is narrower but moving in the same direction.

---

Sources and methodology

We include only programs tied to a peer-reviewed study, an AAMC source, or a school's own statement, and we label each by status (live, piloting, framework, exploratory). Primary and best sources:

• Triola et al., Academic Medicine 98(9):1036 (2023) — NYU Grossman's screening model.
• Keir, Woldenberg et al., JAMIA Open 6(1):ooad011 (2023) — Zucker/Hofstra-Northwell's model.
• AAMC, "AI will now read your medical school application" (Jan 21, 2025) — the connective source naming GW, Cincinnati, and UCSD.
• AAMC × Thalamus press materials on Cortex (2025) and the Laryngoscope audit of Cortex accuracy.
• Liaison/PRNewswire announcement of WebAdMIT Holistic and Predictive Insights (March 2026).

This page is updated as schools confirm deployments; "piloting" and "exploratory" are not predictions of launch.

GradPilot helps medical school applicants review and strengthen their essays in their own voice—so the question of "who, or what, wrote this?" never comes up.

Related Reading

• Med Schools Ban Your AI — Then Use AI to Screen You
• Can You Use ChatGPT for Your Medical School Application?
• Medical-School AI Policy Directory
• Which Colleges Use AI to Read Essays?
• AI-Powered Medical School Essay Review: How It Works