How Many Med School Applicants Use ChatGPT?
Published estimates of AI use in medical school and residency essays, with sample sizes, and the one study that checked admissions outcomes.
How Many Med School Applicants Use ChatGPT?
No published count of applicants exists that we could verify. What exists are detector estimates of essays. At one US medical school, 12.3% of 2023–2024 essays scored as likely AI-written, against 2.7% of 2021–2022 essays, written before the tools were widely available (6,000 essays; preprint posted October 22, 2024).
Every figure below is one study's count of essays its tool scored past a threshold the authors chose. None is a count of applicants. The one study that checked admissions outcomes found the score did not predict who was admitted. This page lays out each published estimate with its sample, its method and its date, then what the reader studies found. It sits inside our medical and health application essays hub.
Every study page cited below was retrieved on November 7, 2026. Detector names are left out. We use only figures whose sample size appears in the study's own published record.
The estimates, side by side
| Study | Sample | What counted as AI | Finding |
|---|---|---|---|
| Hagemann, Ratts and Spies, medRxiv (preprint, posted October 22, 2024) | 6,000 essays submitted to one US medical school, 2021–2022 and 2023–2024 | A commercial detector's likelihood-human score below 0.5 | 2.7% (2021–2022) → 12.3% (2023–2024) |
| Journal of Surgical Education (published online June 18, 2025) | 1,490 personal statements from US MD and DO applicants to one general surgery residency, 2022–23 and 2023–24 Match cycles | Flagged by both of two commercial detectors | 13 (1.7%) → 155 (21.2%) |
| JBJS Open Access (November 19, 2025) | Interviewed applicants to one orthopaedic surgery residency: 66, 58 and 69 personal statements, and 231, 179 and 250 letters, over three cycles | Any originality score below 100 on one commercial detector | Personal statements 65.2% → 60.3% → 43.5%; letters 5.6% → 5.0% → 17.6% |
Read each row for its threshold before you compare them. The medical school study counted essays under 0.5 on a 0-to-1 scale. The surgery study counted only statements both detectors flagged; each detector alone flagged far more (10.2% and 36.6% for one, 2.6% and 22.5% for the other, in the two cycles). The orthopaedic study counted any score under 100 as AI, which its methods state: "LOR and PS with an originality score less than 100 were classified as likely to have been generated or influenced by AI." Its results text gives 44 of 250 letters (17.6%) for the last cycle; its abstract says 17.8%.
The medical school study also reports that secondary essays scored lower on its likelihood-human measure (Phuman) than Personal Comments essays, and that "younger age, visa requirement, and higher GPA were significantly associated with lower Phuman."
Self-reported use. A 2025 Journal of Investigative Medicine survey of third- and fourth-year medical students asked about AI use in residency applications. Its abstract gives no sample size, and the journal's page returned HTTP 403 to our fetch, so we use none of its figures. We found no survey of medical school applicants with a sample size on a page we could open today.
What these numbers can and cannot tell you
Each row answers a narrower question than the title asks: how many essays at one school or one program crossed one tool's threshold in one cycle. The thresholds differ, so the rows cannot be added or averaged, and none of them names an applicant.
Our view: the statistic that matters to an applicant is not the share of essays scored as AI. It is the outcome finding: at the one school that measured it, the score did not predict admission, and in the two reader studies below, raters scored AI-generated and human statements about the same. What a reader can act on is what the statement says.
The countercase: the rise between cycles is real at the pool level in these studies. The medical school and surgery studies both report a large jump, the surgery study with two tools agreeing. A program deciding policy should weigh that trend; it still says nothing about which essays.
Do admissions readers care, and can they tell?
| Study | Sample | Finding |
|---|---|---|
| Hagemann et al. (above) | 6,000 essays, one medical school | "Phuman was not predictive of admissions outcomes in uni- or multivariate analyses." |
| Canadian Medical Education Journal, pediatric residency (December 22, 2025) | 30 AI-generated and 30 human personal statements; four faculty raters | "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." |
| Canadian Medical Education Journal, McMaster (December 22, 2025) | Essays by 24 undergraduates and recent graduates, plus essays generated by an AI tool alone; eight raters | "Analyses revealed no statistically significant differences in ratings across the four writer groups (p = .358)." |
Readers in these two studies scored AI-generated and human statements about the same and could not reliably tell them apart. The pediatric study adds that raters disagreed with each other: "There was considerable disagreement in scores between raters as indicated by negative ICCs."
Readers may still react to a disclosed AI label. That research comes from college admissions, not medicine; our page on how readers evaluate AI-assisted essays covers it. For which schools use AI on their side of the process, see medical schools using AI to screen applicants.
Letters of recommendation moved the other way
In the orthopaedic study, personal statements scored as AI fell over three cycles while letters rose:
| Cycle | Personal statements scored as AI | Letters scored as AI |
|---|---|---|
| 2022–2023 | 43 of 66 (65.2%) | 13 of 231 (5.6%) |
| 2023–2024 | 35 of 58 (60.3%) | 9 of 179 (5.0%) |
| 2024–2025 | 30 of 69 (43.5%) | 44 of 250 (17.6%) |
The authors also found "no significant difference in plagiarism scores across the 3 cycles for either LORs (p = 0.28) or PSs (p = 0.39)." This is one program's interviewed applicants with one detector, so it shows a direction, not a national rate.
What to do with these numbers
- Do not edit to a detector score. Edit for what the reader studies measured, the statement itself: specifics only you could write.
- Keep your drafts. Version history is the proof that holds up if your writing is ever questioned. Our page on an essay flagged as AI that you wrote shows how.
- Read the rule you sign. The certification in your application sets what help is allowed; the comparison of medical school AI policies collects them.
- For residency, read the program's own page. Our page on whether residency programs can detect AI covers the residency side.
Questions about AI use statistics
What percentage of medical school applicants use AI?
No published count of applicants exists that we could verify. The one medical school study estimated 12.3% of 2023–2024 essays scored as likely AI-written, against 2.7% of pre-availability essays, at one school.
Do admissions committees care if you use AI?
The one outcome study found the detector score was not predictive of admission. The rule that applies to you is the certification you sign and any school policy.
How many residency applicants use AI on their personal statement?
At one general surgery program, 21.2% of 2023–24 statements were flagged by both of two detectors; at one orthopaedic program, 43.5% of 2024–25 statements scored under 100 on one detector. Different thresholds, so do not average them. For what these numbers mean for non-native writers, see how IMGs can correct English without a rewrite.
Get a free review of your personal statement
Get a free Quick Review of your personal statement against GradPilot's AMCAS personal statement rubric. It checks whether your experiences show what you did and learned about a physician's work, and whether you explain how they changed your understanding. Residency applicants can use the ERAS personal statement rubric, which reads for why this specialty and the experiences you chose.
Sources
Retrieved November 7, 2026, except where noted. Detector and model names in the studies are left out of this page.
- Hagemann IS, Ratts VS, Spies NC, Estimating the Prevalence of Generative AI Use in Medical School Application Essays, medRxiv preprint posted October 22, 2024 (record retrieved through the medRxiv API; the web page returned HTTP 403 to our fetch): 6,000 essays, prevalence, baseline, secondary essays, outcomes.
- Artificial Intelligence-Assisted Narratives: Analysis of Surgical Residency Personal Statements, Journal of Surgical Education, published online June 18, 2025 (PubMed record retrieved through NCBI E-utilities): 1,490 statements, detector flags.
- Trends in Artificial Intelligence Usage in Orthopaedic Surgery Residency Applications, JBJS Open Access, November 19, 2025 (full text via PubMed Central): counts by cycle, classification threshold, plagiarism.
- Re-evaluating the role of personal statements in pediatric residency admissions, Canadian Medical Education Journal, December 22, 2025: 60 statements, rater accuracy.
- Investigating the threat of AI to undergraduate medical school admissions, Canadian Medical Education Journal, December 22, 2025: 24 writers, eight raters, ratings.
- Utilization and perception of generative artificial intelligence by medical students in residency applications, Journal of Investigative Medicine, 2025: no sample size on the pages we could open; no figure used.
Review Your Personal Statement
See how your AMCAS or secondary essay scores before you submit.