Key findings

1Practice pays off over weeks: some applicants gained 20+ points by their 50th mock interview
2The most dedicated go far: the top 1% did 115+ mock interviews, and the top 0.2% did 300+
3Applicants who started earlier scored higher, mostly because they practiced more
4Don't spend your time on "Why medicine?": it's already your best answer
5The 80/20 of interview prep is ethics: 1 in 5 practice answers, but over a third of the weakest
6Substance beats polish, and better substance comes with better delivery
7The #1 mistake is not answering the question: 1 in 4 practice answers missed a core part of it
8Say more, not less: answers under 100 words a turn miss the question 40% of the time

Finding 1

Practice pays off over weeks: some applicants gained 20+ points by their 50th mock interview

We compared each applicant with their own first three mock interviews. Among those who reached about 50, the share of their answers scoring 80 or higher nearly doubled, from 17.6% to 33.8%, and the share scoring under 70 fell from 19.8% to 8.5%.

The biggest gains were individual. Some applicants gained 20 or more points by their 50th mock interview, all of them people who started near the bottom. The average gain was smaller, about 3 points, because many applicants started high and stayed there: where you start varies far more than how much you gain.

Progress takes weeks, not a day. On applicants' busiest practice day, usually 7 mock interviews, their last answers scored the same as their first, and we found no single session where things suddenly click.

points gained by some applicants by mock 50: 20+; points gained by some applicants by mock 50
answers scoring 80+, first mocks vs about the 50th: 17.6% to 33.8%; answers scoring 80+, first mocks vs about the 50th
points, the average gain: About 3; points, the average gain

Scores rise slowly with practice, and where people start varies more

Average AI rubric score at each session, with the middle half of applicants shaded. n = 2,619 applicants at session 1 and 117 at sessions 51-100.

Average
Middle half of applicants (25th to 75th percentile)

Source: Confetto practice data, Dec 10, 2024 to Oct 9, 2026. Earlier rubric versions standardized and shown on the current scale.

What to do about it

Give yourself weeks, not a weekend. Practice steadily, retry the questions you struggled with after reading the feedback, and judge your progress over many sessions, not one day.

n = 142 applicants who reached about 50 mock interviews (sessions 45 to 55 against their own sessions 1 to 3); 116 applicants with 50+ sessions for individual gains; 2,619 applicants at session 1 in the chart; 729 applicants for the busiest-day test.

How we measured: Scores from earlier rubric versions are standardized and shown on the current scale (mean 75.5, 1 SD = 8.7 points). Later points include only people who kept practicing. Large individual gains came from applicants who started low, and part of any low start is a bad day.

Finding 2

The most dedicated go far: the top 1% did 115+ mock interviews, and the top 0.2% did 300+

Most applicants practice a little: the median applicant did 3 scored mock interviews and 1 in 10 did 27 or more. At the far end, the top 1% did 115 or more, and a handful, about 1 in 500, did more than 300.

The chart follows the 117 applicants who did 50 or more, the same people at every point. Their average rose from 72.5 at their first mock interview to 77.1 by their 51st to 100th, and the weaker end of the group rose most: the bottom of their middle half went from 70.8 to 75.8.

mock interviews for the median applicant: 3; mock interviews for the median applicant
for the top 1%: 115+; for the top 1%
for the top 0.2%: 300+; for the top 0.2%

Applicants who did 50+ mock interviews kept improving

The same 117 applicants at every point: average AI rubric score, with their middle half shaded. 72.5 at session 1, 77.0 at sessions 51-100.

Average
Middle half (25th to 75th percentile)

Source: Confetto practice data, Dec 10, 2024 to Oct 9, 2026. Earlier rubric versions standardized and shown on the current scale.

What to do about it

Volume alone isn't the goal. Space your sessions out, read the feedback, and spend extra sessions on the question types where you lose the most points.

n = 2,677 medical school applicants with at least one scored mock interview; 5 did 300 or more. The chart shows the 113 to 119 applicants with 50 or more sessions.

How we measured: Scores from earlier rubric versions are standardized and shown on the current scale. People who practice this much chose to, so they are not typical applicants.

Finding 3

Applicants who started earlier scored higher, mostly because they practiced more

Applicants who started a month or more before their interview scored about 2.3 points higher in their final week than those who started in the last week. They had also done far more practice: a median of 31 sessions before the final week, against none.

When we compared applicants with the same amount of practice, the start date made no difference. Starting early helps because it gives you time to practice more.

Most applicants don't start early. The median applicant did their first mock interview 10 days before the interview, 40.2% started less than a week out, and 38.0% of all practice happened in the final 7 days.

points in the final week for month-plus starters: +2.3; points in the final week for month-plus starters
median start before the interview: 10 days; median start before the interview
start less than a week out: 40.2%; start less than a week out

Most applicants start practicing within two weeks of the interview

Median start: 10 days before the interview. n = 336 medical school applicants who saved an interview date.

Less than a week
A week or more

Source: Confetto practice data, Dec 10, 2024 to Oct 9, 2026

What to do about it

Start at least two to three weeks before your first interview, so there is time to practice, read the feedback and practice again.

n = 248 applicants with sessions in their final week (135 started under a week out, 35 a month or more); 336 applicants for start times; 8,773 sessions from 393 applicants for the final-week share.

How we measured: Based on the one interview date each applicant saved, which may be their next interview rather than their first. Scores are standardized across rubric versions.

Finding 4

Don't spend your time on "Why medicine?": it's already your best answer

Two in three medical school applicants (67.6%) told us at sign-up that "Why medicine?" is a question they struggle with. In practice it was their best answer: 3.4 points above the same applicants' other answers (95% CI +2.4 to +4.4), and it missed a core point in 15% of answers, against 45% for ethics.

Where "Why medicine?" answers fall short is sameness, not substance. One in four (24.4%) leaned on a stock line such as "help people", "always been passionate" or "since I was young", three times the rate of any other question type. Those lines did not lower the score; they just make an answer sound like many others.

say they struggle with "Why medicine?": 67.6%; say they struggle with "Why medicine?"
points above their own other answers: +3.4; points above their own other answers
of answers lean on a stock line: 24.4%; of answers lean on a stock line

What to do about it

Rehearse your "Why medicine?" answer a few times, swap any stock line for something only you could say, and move on to the questions that cost more points.

n = 5,895 sign-up survey answers; 134 applicants compared with themselves (current rubric); 767 first answers from 266 applicants for stock lines.

How we measured: Within-applicant comparison of AI rubric scores; it holds before the Oct 2 scoring upgrade. "Why medicine?" questions are found by their wording, and stock lines by a fixed phrase list on the first answer, from speech-to-text.

Finding 5

The 80/20 of interview prep is ethics: 1 in 5 practice answers, but over a third of the weakest

Ethics questions were 19.9% of practice sessions but 35.6% of the answers that missed a core point, 38.2% of answers scoring under 60, and 34.1% of all points lost below applicants' own averages. The same applicants scored 3.9 points lower on ethics than on their other answers (95% CI -5.1 to -2.6).

Doing more of the same didn't close the gap. Applicants who did five or more ethics sessions scored 71.7 on them, about the same as those who did one or two (70.8), and nearly half of ethics answers in every group still missed a core point.

Ethics sits with role-play (69.4) and healthcare policy (70.8) at the bottom of the question themes; personal growth (77.9) and teamwork (76.1) are at the top.

of practice is ethics: 19.9%; of practice is ethics
of missed core points are ethics: 35.6%; of missed core points are ethics
points against their other answers: -3.9; points against their other answers

Ethics, policy and role-play questions score lowest

Average score out of 100 by question theme, with the share of sessions missing a core point. 17 themes had 100+ sessions; 108 to 1,528 sessions each.

Lowest-scoring themes
Highest-scoring themes

Source: Confetto practice data, Aug 26 to Oct 9, 2026

What to do about it

Put your extra practice into ethics, and change how you practice it: use a simple frame (the issue, who is affected, the options, your decision and a next step) and check after each answer which core point you missed.

n = 3,840 current-rubric sessions from 466 applicants (764 ethics sessions from 212 applicants) for the shares; 190 applicants compared with themselves; 50 applicants with 5+ ethics sessions and 115 with 1 or 2; 17 themes with at least 100 sessions and 30 applicants each in the chart.

How we measured: Ethics means questions tagged ethics, medical ethics or research ethics. Applicants who practice ethics a lot may start weaker, so this shows that more of the same practice hasn't closed the gap, not that practice can't help.

Finding 6

Substance beats polish, and better substance comes with better delivery

We compared the top quarter of sessions (scores of 81 or higher) with all the others. The widest gap is in content and completeness: 85.7% of the maximum points against 71.7%. A separate AI review of video frames found almost no difference in body language (61.0 against 60.4 out of 100) or facial expression (63.7 against 62.3).

Substance and delivery move together. When the same applicants gave answers with stronger content, their professional communication score was 5.2 points higher (95% CI +4.4 to +5.9), and it was higher for 73% of them. Body language barely changed (+0.7). Clear thinking seems to come out as clearer speaking, not as better posture.

The best answers differ in substance, not body language

Top quarter of sessions (score 81+) against all others. n = 1,079 and 2,836 medical school sessions.

Top quarter (score 81+)
All other sessions

Source: Confetto practice data, Aug 26 to Oct 9, 2026

What to do about it

Spend most of your preparation on what you say. Work on posture and eye contact once your answers reliably cover the question, not before.

n = 1,079 top-quarter and 2,836 other sessions (video review for 1,074 and 2,812); 328 applicants compared with themselves (their stronger-content sessions against their weaker ones).

How we measured: Content is half of the rubric, so part of the content gap is built in, and professional communication is scored by the same AI review as content, so the two are not independent. The video review runs separately.

Finding 7

The #1 mistake is not answering the question: 1 in 4 practice answers missed a core part of it

When our AI scorer reviews an answer, it lists five points about it and labels each one a strength, something to improve, or a core omission: a missing point that materially affects the answer. This season, 24.7% of medical school practice sessions had at least one core omission.

Those answers averaged 66.7 out of 100, against 78.4 for answers without one. Comparing answers of similar length, the gap is still 11.0 points.

The flag almost separates the weakest answers from the strongest. It appears on two thirds (66.7%) of bottom-quartile answers and on 2.7% of top-quartile answers.

of answers miss a core point: 24.7%; of answers miss a core point
average score when they do: 66.7; average score when they do
average score when they don't: 78.4; average score when they don't

What to do about it

Before you answer, name the one or two things the question is really asking. At the end, check you covered them. For an ethics question that usually means the conflict, who is affected and what you would do.

n = 3,915 sessions from 461 medical school applicants, Aug 26 to Oct 9, 2026.

How we measured: The flag and the score come from the same AI judgment, so they are not independent measures, and the rubric caps the score when content is weak.

Finding 8

Say more, not less: answers under 100 words a turn miss the question 40% of the time

A common worry is talking too long. In our data the bigger problem is the opposite. In standard eight-minute sessions, answers that averaged under 100 words per turn scored 71.5, while answers of 200 to 299 words per turn scored 77.6, a gap of 5.9 points (95% CI +4.2 to +7.5).

The same people show it too: comparing applicants' own short and long sessions, the longer ones scored 7.3 points higher. Short answers missed a core part of the question 40% of the time, against 13% for the fullest ones. The extra words add substance; structure stays about the same.

We found no length at which scores started to fall, though very few answers ran past 300 words per turn.

Short answers score lower and miss the core more often

Standard-length sessions (7 to 9.5 minutes). n = 1,765 medical school sessions, 226 to 649 per band.

Average score
Missing a core point

Source: Confetto practice data, Aug 26 to Oct 9, 2026

What to do about it

If your answers often finish in a few sentences, add the reason behind your point and one concrete example. Aim for a complete answer, not a long one.

n = 1,765 standard-length sessions (7 to 9.5 minutes) in the four bands shown, from medical school applicants (100 to 231 per band); 30 applicants had sessions in both the shortest and longest groups.

How we measured: Words per turn is the applicant's words divided by the number of answers in the session (main answer plus follow-ups), from speech-to-text. The gap holds before and after the Oct 2 scoring upgrade and after removing the scorer's small length adjustment.

More from the data

Check your format: 68% of the most-searched schools interview only online, 40% use an MMI

We checked the 60 schools whose interview guides applicants search most against each school's own pages in October 2026 (39 US MD, 15 US DO, 5 Canadian and 1 UK school). 41 of them (68%) interview only online, 5 (8%) only in person, and 11 (18%) offer both or let applicants choose.

24 (40%) use an MMI for at least part of the interview, 28 (47%) use traditional or one-on-one interviews, 4 use a panel and 4 a recorded video interview as the main step. 16 (27%) add a recorded or asynchronous step, such as Kira Talent responses, Acuity's early-phase MMI or a Spark Hire video, and 16 list Casper.

Across all US MD schools, the AAMC's own 2027 MSAR interview report, which schools fill in themselves, shows the same shift: about 4 in 5 list virtual-only interviews.

interview only online: 68%; interview only online
use an MMI for part of the interview: 40%; use an MMI for part of the interview
add a recorded or asynchronous step: 27%; add a recorded or asynchronous step

n = 60 schools, checked Oct 9, 2026. AAMC: Medical School Admission Requirements, Interview Procedures, 2027. How we measured: These are the 60 most-searched schools, not every school. Categories are our reading of each school's own wording, and formats can change during a cycle, so check your invitation.

Who is preparing: 68% target at least one MD school, 28% a DO school, 16% a Canadian school

At sign-up, medical school applicants pick the schools they are preparing for. Among the 5,660 who picked at least one, 68.0% chose a US MD school, 27.6% a DO school (15.5% both) and 15.7% a Canadian school.

Asked directly whether they are applying MD or DO, 70.3% said MD, 8.4% DO and 21.4% both. More say "both" than list both kinds of school, because most lists are short (a median of 2 schools).

chose at least one US MD school: 68.0%; chose at least one US MD school
chose at least one DO school: 27.6%; chose at least one DO school
chose at least one Canadian school: 15.7%; chose at least one Canadian school

n = 5,660 medical school sign-ups who chose at least one target school; 5,995 answers to the MD or DO question. How we measured: Confetto users, not all applicants. School types come from our school directory; target lists show interest, not full application lists.

Most start late and alone: 72% had never done a mock interview when they signed up

When medical school applicants sign up, we ask how far along they are. 72% had not done a single mock interview yet, and 53% said their earliest interview was next week.

Most practice without useful feedback: 41% get no feedback at all and 23% say the feedback they get isn't useful. 72% have no mock interview partner, and among those who do, the most common complaint is that partners are too nice or not honest (34%), ahead of scheduling (32%).

Only 14% say they know their specific weaknesses, and 22% are unsure of their schools' interview formats or didn't know they vary.

have done no mock interview yet: 72%; have done no mock interview yet
have their first interview next week: 53%; have their first interview next week
have no practice partner: 72%; have no practice partner
know their specific weaknesses: 14%; know their specific weaknesses

n = 5,859 to 5,960 medical school applicants per question (about 5,900), sign-ups up to Oct 9, 2026. How we measured: Self-reported answers to our sign-up survey from people already signing up for interview prep, which skews toward applicants with interviews soon. The wording has changed over time.

A typical practice week: 3 mock interviews and about 20 minutes

In weeks when they practiced, the median applicant did 3 scored sessions and about 22 minutes of interview practice. 62% of active weeks had 3 or more sessions.

Most practice is short-lived: 72% of applicants practiced in only one calendar week. In the 4 weeks before a saved interview date, the median was 2 sessions and 12 minutes a week.

sessions in a median practice week: 3; sessions in a median practice week
of practice in that week: 22 min; of practice in that week
practice in only one week: 72%; practice in only one week

n = 2,600 active weeks from 1,682 medical school applicants, Dec 2025 to Oct 2026; 321 applicants for the final 4 weeks. How we measured: An active week is a calendar week with at least one scored session. Minutes are time in session, from the transcript.

Casper is a race against the clock: 83% of typed practice answers hit the timer

Confetto also offers Casper practice. We are independent and not affiliated with Acuity Insights, which makes the Casper test. In the first two weeks of data, 83% of typed practice scenarios ran out the 3.5-minute timer before the applicant submitted (429 scenarios from 56 applicants).

On their first typing test, about half of applicants (52.6% of 57) typed under 50 words per minute, the speed we suggest for saying enough in the time.

These samples are small, so read them as directional. A fuller Casper report is coming in January.

of typed scenarios ran out the timer: 83%; of typed scenarios ran out the timer
typed under 50 words per minute: About half; typed under 50 words per minute

n = 429 typed scenarios from 56 applicants; 57 applicants' first typing tests. How we measured: Confetto practice data, Sep 27 to Oct 9, 2026. The timer flag is set when the countdown ends, so some applicants may type to the end on purpose. The 50 words per minute guide is ours, not Acuity's.

Myths the data didn't support

"Longer answers lose points." They don't. Answers under 100 words per turn scored lowest (71.5) and we found no length at which scores started to fall.
"You get worse as an MMI circuit goes on." No sign of it: the last two stations of practice circuits scored the same as the first two (+0.5 points, 126 circuits, 56 applicants).
"Practicing late at night or the night before hurts." No sign of that either: late-night sessions scored the same as the same applicants' daytime ones (-0.3 points, 63 applicants), and so did practice in the final 24 hours.

Methodology and limitations

Data

Aggregated, read-only Confetto practice data pulled Oct 9 and 10, 2026. Confetto internal data: 54,839 mock interviews since December 2024 by 8,394+ medical school applicants. The analysis uses the sessions in our current database by applicants who chose medical school at sign-up, from Dec 10, 2024 to Oct 9, 2026: 35,952 started, of which we analyzed the 29,572 that were completed and AI-scored, from 2,675 applicants (5,297 of them since Aug 1, 2026). Each MMI station and each panel answer counts as one session. Score findings use sessions scored with the current rubric from Aug 26 to Oct 9, 2026, by applicants who chose medical school at sign-up: 3,915 sessions from 461 applicants, with a mean of 75.5 and a median of 76. A session is one interview question plus the AI interviewer's follow-ups; the median session lasts 7.4 minutes.

Inclusion

A scored session is a completed interview with an overall score. We excluded internal, staff and test accounts, and sessions whose account no longer exists (about 0.7%). Groups are reported only when they have at least 30 people (n ≥ 30), with applicant counts alongside, and numbers are rounded to one decimal place. No names, emails or answer text were used.

How scores are made

Scores come from our internal AI rubric, trained on medical school acceptance information. An AI model reads the whole transcript, including follow-ups, and scores three dimensions from 0 to 5 against anchored definitions: content and completeness (weight 50%), structure and organization (25%) and professional communication (25%). Code turns these into a score out of 100 on a fixed curve, caps the score when content is weak or the answer is under 120 words, and applies a small length adjustment (+1 to -3). The model also writes five analysis points, each labelled a strength, an improvement or a core omission. A separate AI model rates body language and facial expression from screenshots of the session.

Checks

The rubric changed on Aug 26, 2026, so score findings start there. Our scoring model was upgraded on Oct 2, 2026, which raised average scores by about 3.6 points. We checked the practice and answer-length findings on sessions from before the upgrade, and they hold. Practice findings that span earlier rubric versions use scores standardized within rubric version, week and scoring model, and compare each applicant with their own first sessions.

Limitations

The users are self-selected: applicants who chose to try an AI interview prep tool. They are not a representative sample of all applicants.
Scores estimate interview performance; they don't guarantee admission, and we don't track admissions outcomes.
The practice questions and their themes are Confetto's, not schools' real questions.
Survey answers are self-reported, and not every question was shown to every applicant.
Word counts come from automatic speech-to-text.

For journalists

You may use these findings and charts with credit to Confetto and a link to this page.

Media contact: hello@confetto.ai

Charts as images: PNG downloads of each chart will be added here. Until then, email us and we will send them.

Confetto is an AI interview-practice platform. Scores are AI rubric scores from practice sessions and do not predict admission. Not affiliated with any medical school, the AAMC or Acuity Insights.