Should You Use AI to Write Your Secondaries?

Thirty secondaries land in your inbox in eight days. Every one of them wants 500 words on why their school, specifically, is the one. You have a job. You have a life. You have a machine in your pocket that will produce 500 competent words in eleven seconds.

So you have thought about it. Everyone has thought about it. Let's get you a real answer, because this question has better evidence behind it than almost anything else in the application.

What the Rules Say

Start with the AMCAS certification, since most people click past the thing they are signing.

The AAMC's certification statement asks you to affirm that all your writing is your own. It then names AI tools right alongside mentors, peers, and advisors as acceptable help for brainstorming, proofreading, or editing, provided the final submission is "a true reflection of my own work." The AMCAS Application Workbook says the same thing even more plainly, telling applicants they may use artificial intelligence tools for brainstorming, proofreading, or editing their essays.

That surprises people. AMCAS is the permissive end of this spectrum.

Here is the catch that matters for this article. That certification governs your AMCAS application: personal comments, work and activities, MD-PhD essays. Your secondaries are school documents, and they run on school rules.

Those rules are all over the map. Mount Sinai publishes a plain-language list of ethical and unethical uses, putting researching schools, brainstorming topics, and grammar review in the acceptable column, and putting outlining, drafting, or writing your essays in the unacceptable one. West Virginia asks for applications completed without the assistance of AI platforms. UC Davis prohibits using AI to generate responses. Georgetown's language is the bluntest of the bunch, stating that the only person who may be doing the writing is you. A policy tracker covering 72 medical schools found 45 with no AI-specific policy at all, deferring to the AMCAS language, and 27 with explicit rules, a handful of which ban AI outright.

One more thing I ran into while researching this, and it is worth your attention. A site selling AI "humanizing" services publishes a much stricter version of the AMCAS certification, worded as an outright ban, that does not match what the AAMC publishes. Read the AAMC page yourself, then read each school's own instructions. Do not take a vendor's word for what you are signing.

What the Data Says

People are doing this, and the numbers are surprisingly well documented.

Researchers at a US medical school analyzed 6,000 application essays, comparing the 2021 to 2022 cycle with the 2023 to 2024 cycle. They ran everything through GPTZero to score how likely each essay was to be human written.

Two details in that study matter more than the headline.

Secondary essays scored lower than personal statements, which means more AI use on exactly the essays you are reading this article about. And the human-authorship score showed no relationship with admissions outcomes. Whatever those applicants were doing, it did not appear to buy them anything.

Now the question everyone asks next: can readers tell?

Not reliably. In a study published under the title Death of the Personal Statement, 17 medical school application readers scored 309 essays and identified authorship correctly 56 percent of the time. A coin gets 50. The readers also rated the AI essays more highly, and they tended to assume human authorship for the essays they scored best. Studies on the residency side, in otolaryngology and spine surgery fellowships, found the same pattern.

So here is the honest landscape. Reviewers mostly cannot spot it. Detectors are inconsistent. More applicants are doing it every cycle, especially on secondaries. And the best available evidence says it did not improve anyone's odds.

Sit with that last one for a second. The whole appeal of this shortcut is that it buys you an edge. The data says it buys you time and nothing else.

Where This Leaves You

Three reasons I would keep the drafting human, and then what to do instead.

The risk runs in both directions. Getting caught is one problem. Getting wrongly accused is the bigger one. In a 2023 paper in Patterns, Liang and colleagues at Stanford ran seven detectors over essays written by humans and found something ugly.

Every essay in that study was human written. The detectors were nearly perfect on US eighth-grade essays and catastrophic on TOEFL essays, because they lean on statistical measures that also happen to track how much linguistic variety a writer uses. The authors say plainly that these tools should not be used to evaluate people, especially non-native English speakers.

Keep that in mind when reading the prevalence study, which found that needing a visa predicted a lower human-authorship score. That could be a real behavior difference. It could also be the same bias showing up in the data. Nobody can tell you which, and that ambiguity is exactly why these tools make shaky evidence.

Generic writing is the specific way secondaries fail. In post-cycle interviews reported in Academic Medicine, medical students described AI-written statements as vague and not concise. Vague is fatal here. A secondary exists to answer one question: why this school, in a way nobody else could write. A model that has never walked through that hospital, never sat in that clinic, and never met your patients will hand you 500 smooth words that could belong to anyone.

You signed something. Medicine runs on attestations, from a certification statement now to a chart note later. Starting that habit by signing something untrue is a rough opening move, and it is the part no detector or policy loophole touches.

So here is the version I would run.

• Use the permitted lane on purpose. Brainstorming, structural feedback, proofreading, researching a school's programs, building a tracker for 30 prompts and deadlines. AMCAS names most of this as acceptable, and it saves real hours.
• Pre-write the common prompts in May and June. The guy who applied to 151 schools pre-wrote 14 of them before a single secondary arrived, and it is the single biggest time saver in this entire process.
• Draft ugly and fast, then edit. A rough paragraph in your own voice beats a polished one in nobody's, and editing is where the permitted tools genuinely earn their keep.
• Keep a specificity file per school. Three concrete details each: a program, a faculty member, a community partnership. That file is the thing AI cannot fake and readers cannot ignore.
• Check each school's policy before you write. They differ, they change, and the instructions are usually right there in the secondary itself.

One honest caveat on everything above: every prevalence study here depends on detectors that we just established are unreliable. Treat 12.3 percent as an estimate with wide error bars, not as a measurement.

Your secondaries are the only place in the entire application where a school hears you think. Make sure it is you doing the thinking.

Sources

• AAMC, AMCAS Certification Statements
• AAMC, AMCAS Application Workbook
• Spies, Ratts and Hagemann, Estimating the Prevalence of Generative AI Use in Medical School Application Essays
• Death of the Personal Statement, Journal of the American College of Surgeons
• Liang et al., GPT Detectors Are Biased Against Non-Native English Writers, Patterns (2023)
• Rethinking the Personal Statement in the AI Era, Academic Medicine
• Distinguishing AI-Generated and Human-Written ERAS Personal Statements in Otolaryngology
• GradPilot, Medical School AI Policies Tracker
• GradPilot, The Most Detailed AI Policies in U.S. Admissions
• Shemmassian, AI in Medical School Admissions