Interview Season, AI Reviewers, and a 19 Million Dollar Question

Training, research, and the specialty itself. Three things worth knowing this week.

1. Residency and Fellowship · What happens after September 23
2. Research · The reviewer reading your paper may not be human
3. Cardiothoracic Surgery News · Lifetime earnings, and a new POAF guideline

Part One · Residency and Fellowship

The Eleven Days Before Your Inbox Changes

Last week I wrote about why September 23 is the date that actually matters in this application cycle. It is now eleven days away. This week, what happens immediately after it.

Programs begin file review on September 23 at 9 a.m. ET. The first interview invitations follow within days, not weeks, and they do not arrive in a single organized batch. They arrive in waves. Some specialties coordinate a shared release date. Most do not.

Here is the part that catches people off guard.

At many programs, interview slots are allocated first come, first served after the invitation goes out. Desirable dates can fill within hours. The applicant who sees the email at 4:15 p.m. and responds at 4:20 has a meaningfully different set of options than the one who sees it at 9 p.m. This is not a measure of how much a program wants you. It is a queue.

The invitation tells you a program is interested. The speed of your response determines what that interest is actually worth to your calendar.

A second structural feature worth understanding: most programs run hold lists and trickle invites. A first round goes out, then additional invitations follow as applicants decline or new dates open. Silence in early October is not a rejection. It frequently means a program has not finished releasing its slots. Applicants who treat the first wave as final data make ranking and outreach decisions on incomplete information.

Practically, the season runs October through January, with December and January the busiest. Match Day is March 19, 2027.

Part Two · Research

The Reviewer Reading Your Paper May Not Be Human

Last month I wrote about watermarking and what it means for disclosing your own AI use. This is the same question from the other direction, and it is considerably less discussed.

21%

That figure comes from an analysis by Pangram Labs, a firm specializing in AI content detection, covering the complete review set for ICLR 2026, released in November 2025. It is the most concrete public estimate available. Medicine has no audit of this scale, which is itself worth noting.

What medicine does have is guidance. In March 2026, members at the Committee on Publication Ethics Forum advised on a case about precisely this problem, in which a journal described a rise in review comments that appear detailed and specific but raise issues unrelated to the actual content of the paper. Detection tools have returned rates above 80 percent on some of these reviews. When editors ask, many reviewers deny the use or simply do not respond, which extends review times and adds work for editorial teams.

A review that looks thorough and says nothing about your paper is a different failure mode than a review that is short or harsh. It is harder to appeal, because on the surface it reads as diligence.

There is a confidentiality dimension that authors rarely consider. Elsevier's generative AI policy, updated in June 2026, states plainly that reviewers should not upload a submitted manuscript or any part of it into an AI tool, because doing so may violate author confidentiality and proprietary rights, and where the manuscript contains identifiable information, may breach data privacy rights. Your unpublished manuscript, in other words, is only as confidential as the habits of the reviewer holding it.

It is worth keeping the scale honest. A large international survey of 1,260 medical journal peer reviewers, posted as a preprint this year, found that while 86.2 percent were familiar with AI chatbots and 87.7 percent had used tools like ChatGPT generally, 70.3 percent had not used them for peer review at all. Nearly 70 percent said their institutions offer no training on the question, and 60.7 percent said they would want it. Most reviewers are not doing this. Enough are that it changes what you can assume.

Part Three · Cardiothoracic Surgery News

What Published This Season

Whether you are building a capture system for interview season, stress testing a manuscript before it reaches a reviewer, or working through what a new guideline changes in your own practice, that is the kind of thinking AskDimitri is built for.

Work through it with AskDimitri →

Editor's note: in October 2026 the figures in this letter were checked against the sources listed below, and the text was corrected where the letter as sent differed from them.