How many programs should I apply to? 1,001 cycles say that is the wrong question.
Sending more applications did not produce more interviews — not in any band, not for any applicant type. The number that actually predicts matching is the interview count, and it stops paying at about ten.
· 1,001 completed cycles
Every applicant asks it, usually in August, usually of someone one year ahead of them. The answers are folklore: sixty for a safe specialty, a hundred if you’re worried, more if you’re an IMG. Nobody has the denominator.
We do, for 1,001 completed cycles. The answer is not a number. It is that the question has almost no predictive power, and the thing it is standing in for does.
More applications, same interviews
Median interviews by applications sent, US MD seniors. The bar spans the 25th to 75th percentile; the dot is the median.
The band that sent 76–120 applications came away with a median of 10 interviews. The band that sent 1–15 came away with 12. Five times the applications, no more interviews.
This is cross-sectional, not an experiment. Nobody was randomly assigned to send 100 applications. The people who sent 100 are overwhelmingly the people who expected to need 100 — a weaker application, IMG status, a prior unmatched cycle, a competitive specialty without the numbers for it.
Application count is a symptom of expected difficulty, not a lever that produces interviews. The honest reading is not “apply to fewer programs.” It is “applying to more programs will not rescue an application that is not getting interviews — and the extra hundred will cost you about $1,200 finding that out.”
Ten interviews is where the curve stops paying
Share who matched, by interviews attended.
From 1–3 interviews to 10–12, the match rate climbs from 38% to 97%. From 10–12 to 17-plus, it does not move. Every interview after roughly the tenth is buying insurance against a tail risk you have already mostly bought off.
That is the number worth managing. It is also the one nobody can tell you in August, which is exactly why the conversation defaults to application count — it is the only quantity you control in the moment.
The useful version of the advice: apply broadly enough to clear ten interviews given your own invite rate, and spend the marginal hour on the things that change the invite rate rather than on applications forty through ninety.
The gap nobody prices
Split the same data by applicant type and a second finding falls out, one that has nothing to do with strategy. It needs one restriction first, and the restriction matters more than the table.
Across all specialties these rows are not comparable. The non-US IMG sample is 88% pediatrics; the US MD sample’s largest single specialty is ophthalmology at 18%. Comparing them measures a specialty difference wearing an applicant-type label. So this is pediatrics only — the one specialty where all four cells have data, and where every row is the same specialty as every other.
| Applicant type | n | Median apps | Median IVs | IVs per 10 apps |
|---|---|---|---|---|
| US MD | 91 | 25 | 18 | 7.2 |
| DO | 40 | 41 | 19.5 | 4.8 |
| US IMG | 12 | 85 | 21.5 | 2.5 |
| Non-US IMG | 23 | 86 | 11 | 1.3 |
A US MD applying pediatrics gets 7.2 interviews for every 10 applications. A non-US IMG gets 1.3, and sends a median of 86 applications against the MD median of 25. Same specialty, same cycle: 5.6× the work per invitation, and three and a half times the applications to get there.
For this group the folklore about applying broadly is right, and for a reason the folklore never states. They are not applying broadly out of anxiety. They are applying broadly because their per-application yield is a fraction of everyone else’s, and the fee schedule charges them the same per application as the applicant getting five times the return.
Two limits worth holding onto. The US IMG cell is 12 people — treat it as a direction, not a number. And every pediatrics applicant in this sample matched, all four cells at 100%, which is not what pediatrics does. It is what a sample of people who chose to post their outcomes does, and it is the clearest evidence in this post of the survivorship problem described below.
What this data is, and what it is not
1,001 completed cycles, cycle years 2025 and 2026. 985 of them are seeded from public outcome threads and a published outcomes dataset; the rest were reported here. Same corpus behind our signal-lift tables.
The sample matched at 88.9%, well above the real rate. People who match are likelier to post their numbers. That inflates every level on the second chart — the shape of the curve survives the bias, the heights do not. Read “97%” as “the plateau,” not as your odds.
Self-reported application and interview counts, months after the fact. Expect rounding to fives and tens.
No specialty split. Cutting 1,001 rows by 22 specialties leaves cells too thin to publish. A dermatology applicant and a family medicine applicant are in the same bars here, and they should not be. That is the next version of this post, and it needs more cycles than we hold.
The specialty split is thin because outcome data is the hardest thing to collect and the easiest thing to skip. Five minutes after your cycle ends makes the next applicant’s version of this post better than yours was.
Share your cycle →