Can Med Schools Tell ChatGPT Wrote Your Secondary?

Can Med Schools Tell ChatGPT Wrote Your Secondary?

No central application service whose rules we could read on November 7, 2026 describes an AI scan of secondary essays. Secondaries go to each school through its own portal, under its own instructions. The published evidence points somewhere else: the person who reads your secondary is the person who will ask you about it at interview.

Arguing about detectors is the wrong fight. One study inside a real medical school's admissions files found that secondaries scored as more AI-like than primary essays, and that the score predicted nothing about who got an interview or an offer. Two rating studies found raters scored AI-written and human-written text about the same, and in one of them faculty named the source correctly only 59% of the time. So the useful question is not "will a tool flag this?" It is "can I stand behind every sentence when someone asks me to keep talking?" If you are drafting now, run each answer through a secondary essay review before it goes out.

Every TMDSAS page and study quoted below was checked on November 7, 2026. The AAMC's AMCAS pages blocked our automated check that day, so this page does not quote the AMCAS certification. Read it yourself on the AAMC certification statements page before you rely on any summary of it.

The short answer, document by document

Our view: treat each school's secondary instructions as the governing rule for that school's secondary, and read them every time. A school that writes its own AI line into the secondary portal has written the rule you sign. If it writes nothing, the certification inside the application is what you sign, so read it there. For school-level policies we track, see medical school AI policies.

What TMDSAS makes Texas applicants certify

Texas applicants sign a certification that does name essays. The TMDSAS application certification page (EY2027 application) has you certify that your personal statement, optional essays and activity descriptions are your own and "have not been composed or generated by a third party." The same page says irregularities are documented and reported to the AAMC, ADEA and other professional organizations, and you authorize your schools "to verify the information I have provided." The full wording, beside every other health service's, is in health application AI policies compared.

Two things matter for a Texas reader. First, the sentence names the TMDSAS essays, not the schools' secondaries. TMDSAS's after-submitting page says "Any questions relating to a secondary application, documents, and/or fees required should be sent directly to the school, NOT TMDSAS." Second, the TMDSAS page never says "AI." It says "generated by a third party." Our reading: text an AI tool wrote for you is text you did not compose, so it sits inside that sentence even though TMDSAS has not said so in public. Nothing on the page describes a scan. It describes documentation, reporting and verification.

What the study inside one medical school's files found

The only published look inside a real admissions office's essays that we found is by Spies, Ratts and Hagemann, published in JMIR Medical Education on September 3, 2026 (an earlier version appeared as a preprint in October 2024). The authors ran "6000 essays from 2364 applicants submitted to a US medical school" through a commercial AI detector. They compared the 2021–2022 cycle, before these tools were widely available, with 2023–2024.

What they report:

• Personal Comments essays had a median human-likelihood score of 0.77 in 2023–2024 against 0.83 in 2021–2022.
• About 12.3% of 2023–2024 essays and 2.7% of 2021–2022 essays scored below 0.5 on that human-likelihood scale. The 2.7% are essays from the baseline cycle, before wide availability of these tools.
• "Essays submitted as part of the secondary application demonstrated lower Phuman values than those of the American Medical College Application Service (AMCAS) "Personal Comments" essays."
• "Phuman was not predictive of interview invitation or acceptance in adjusted applicant-level logistic regression analyses."

And the authors' own conclusion: "AI detection is imperfect. We do not recommend deploying AI detection for individual applications in live admissions cycles."

The scope is one US school, two cycles and one detector, run by researchers after the cycles closed.

Can a human reader tell?

Two blinded studies published in the Canadian Medical Education Journal in December 2025 asked raters to read AI and human writing side by side.

• Pediatric residency statements (Curry et al.): 30 AI-generated against 30 human statements, rated by four pediatric faculty. "There was no significant difference in mean scores between AI and human-written statements. Raters could only identify the source of a letter (AI or human) with 59% accuracy."
• Medical school admissions essays (Choi et al., McMaster): eight raters scored essays from four writer groups, including essays generated solely by an AI tool. "Analyses revealed no statistically significant differences in ratings across the four writer groups (p = .358)."

Neither study was run on secondaries, and one is residency, not medical school. Both point the same way: the reader grades what the text says, not who typed it. Our view: a secondary fails or passes on whether it says something specific and true about you and this school. That is also the thing an interviewer can test.

Where AI-written secondaries actually fail

Our view: these are the four places an AI-written secondary breaks, and none of them needs a detector.

1. The interview follow-up. An interviewer reads your "describe a challenge" answer and asks what you said to your supervisor the next morning. If the story's details came from a prompt, you have nothing to add. The sibling page on what to say if an interviewer asks whether you used AI covers that moment.
2. Cross-school sameness. A tool asked for twenty "why us" answers can return twenty versions of one paragraph. The secondary system that survives 30 schools works because the reused parts are your stories, not your adjectives.
3. Mission-statement echo. An AI draft built from the school's website restates the school's own mission back to it. The reader wrote that mission.
4. Invented specifics. A tool fills gaps with plausible program names, clinic names or tracks. A wrong one tells the reader you never checked.

Constructed example: a "why us" answer an interviewer can probe

Constructed example, written by us, not a real applicant statement. Context (invented): Coppervale School of Medicine, a fictional school with a student-run street-medicine clinic.

Version A (52 words): Coppervale's commitment to serving underserved communities deeply resonates with me. I am drawn to its innovative curriculum, collaborative culture and emphasis on holistic care. As a future physician, I hope to contribute my passion for service and my dedication to health equity, and I am confident Coppervale is where I can thrive.

Version B (57 words): For two years I drove the Saturday supply van for a shelter clinic in Dayton, and I learned that most visits stalled on refills, not diagnoses. Coppervale's student-run street clinic keeps a refill log between visits. I want to help run that log, and I want to learn why your team chose paper over the county's portal.

Version A invites one follow-up: "What do you mean by holistic care?" Any answer works, so the answer tells the interviewer nothing. Version B invites three: "What did a stalled refill look like?", "What would you change in our log?", "Why do you think we chose paper?" Only the person who drove the van can answer the first. That is the gap between a secondary you wrote and one you were handed.

For the rewrite itself, line by line, see how to fix a secondary that sounds like AI.

A ten-minute self-check before the next secondary goes out

Run this on every answer. Each check has a pass condition.

1. Tell it aloud. Say the story in 60 seconds with two details that are not on the page. Pass: you can do it without looking.
2. Name the school-only fact. Point to the sentence in your "why us" answer that names something only this school has. Pass: the sentence breaks if you swap in another school's name.
3. Survive one follow-up. Write the question an interviewer would ask next, then answer it in one sentence. Pass: the answer adds a fact.
4. Check every proper noun. Program names, clinic names, tracks. Pass: each matches the school's own website today.
5. Read the school's instruction. Pass: you have read the authorship or AI line in this school's secondary portal, or confirmed it has none.
6. Keep your drafts. Pass: your earlier versions are saved with dates, so the work is yours on the record.

Our view: if an answer fails checks 1 to 3, rewrite it from your own notes before you submit it, whatever any tool says about it. The countercase: an applicant who wrote every word but writes in a flat, formal register can fail check 2 too. The fix is the same: add the fact only you have.

Questions about AI and medical school secondaries

Does AMCAS check secondaries for AI?

Secondaries are not filed through AMCAS. Each school collects its own. We could not verify the AMCAS certification text on November 7, 2026, because the AAMC site blocked our automated check, so read it on the AAMC page.

Do medical schools use AI detectors on secondaries?

No central service we could check describes a scan of secondaries, so the answer sits with each school. Check the school's secondary portal and its admissions pages. The one published study of a school's essays ran the detector after the cycles closed, for research, and its authors advise against using detection on individual applications.

Can Texas medical schools detect AI essays?

TMDSAS's certification covers the personal statement, optional essays and activity descriptions, and bars text "composed or generated by a third party." It describes documenting and reporting irregularities and verifying information. It does not describe a scan.

Is it cheating to use AI for medical school secondaries?

It depends on what the school's instructions say and what you used it for. A school that bans outside help bans it. Where a school is silent, its application certification is what you sign, so read it there. On our reading, text an AI tool wrote for you is not your own writing under the TMDSAS wording quoted on this page. Copied text is a separate question: does AMCAS check for plagiarism.

Reusing drafts across schools? Run the scrub list for school-specific tells before each submission.

proposed: "Human help sits under the same sentence; see where outside help becomes authorship." (closing)

proposed: "For every published estimate and its sample, see published estimates of AI use in medical school essays." (after the 6,000-essay study mention)

Get a free review of your secondary essay

Get a free Quick Review of your secondary essay against GradPilot's community service and mission secondary rubric. It reads for a named community, service shown as work you did, what the community changed in you, and mission fit backed by evidence. Writing a different prompt? Use the diversity and perspective secondary rubric or the adversity and challenge secondary rubric.

More secondary guides:

• Pre-write medical school secondary essays: a system for drafting before the invitations arrive.
• MD secondary prompts by school: the questions you will actually get.
• Can you use AI for your medical school application?: the primary-application rules, system by system.
• Medical school essays hub: every guide in one place.

Sources

Retrieved November 7, 2026, except where noted.

• TMDSAS, Application Certification (EY2027 application): written-passages certification, irregularities reporting, verification authorization.
• TMDSAS, After Submitting Your Application: secondary questions go directly to the school.
• AAMC, AMCAS certification statements: linked for readers; not verified (automated check blocked, HTTP 403).
• Spies NC, Ratts VS, Hagemann IS, Estimating the Prevalence of Generative AI Use in Medical School Application Essays, JMIR Medical Education, September 3, 2026 (2021–2022 and 2023–2024 cycles): all study figures and the authors' conclusion.
• Curry B et al., personal statements in pediatric residency admissions, Canadian Medical Education Journal, December 2025: rater scores and 59% source identification.
• Choi J et al., AI and the rating of applicant essays, Canadian Medical Education Journal, December 2025: no significant rating differences across four writer groups.