In the surgical wards of Johns Hopkins Hospital in the 1890s, a young professor named William Stewart Halsted built an idea into the architecture of American medicine. A trainee should watch a procedure performed, then perform it under supervision, then teach it to someone with even less experience than himself. See one, do one, teach one, the formula came to be called, and for more than a century it shaped how surgeons were made on both sides of the Atlantic. It survives today in trauma bays and teaching hospitals from Baltimore to Bristol. Its survival owes less to stubborn tradition than to its third step, teaching, which appears to do real cognitive work of its own.
The idea is older than the label
That intuition now has a name and a literature behind it. In 2009, a team led by the Stanford researcher Catherine Chase coined the phrase “protégé effect”. The program asked middle-school students to teach biology to a cartoon avatar rather than study it for a test of their own. The children worked harder for their pupil than they ever had for themselves, reading more carefully and revisiting their own mistakes more willingly. The researchers concluded plainly that “students make greater effort to learn for their TAs [teachable agents] than they do for themselves”. A companion account from Chase’s Stanford lab traced the mechanism to something almost parental. It identified a sense of responsibility for one’s agent that appears to motivate students to work harder to organize their understanding.
The observation, though, is much older than the experiment. Two thousand years before anyone built a teachable cartoon, the Roman Stoic Seneca wrote to his friend Lucilius. He described the same clarity that comes from explaining a thing to somebody else.
While we teach, we learn.Seneca, Letters to Lucilius
What has changed since Seneca’s time is not the observation but the evidence for it, especially the discovery that a person does not need an actual pupil to get the benefit. In a set of experiments at Washington University in St. Louis, the psychologist John Nestojko and his colleagues had students read text passages under one of two instructions: prepare to be tested, or prepare to teach the material to someone else. In the end, nobody taught anyone; every participant simply sat the same test. Yet the students who had merely expected to teach organized their recall of the passage more effectively and remembered more of what mattered; Nestojko explained that “when teachers prepare to teach, they tend to seek out key points and organize information into a coherent structure.” His co-author, the memory researcher Elizabeth Bjork, drew the more unsettling conclusion that students “do not necessarily employ activities that best foster learning, even though, as our results indicate, those strategies are in their ‘toolbox.'” An audience, real or merely anticipated, seems to pull the right study strategy out of a learner who already owned it and simply wasn’t using it.
The size of the benefit depends on the dose
Psychologists who have tried to isolate which part of “teaching” does the work distinguish at least three stages: expecting to teach, actually explaining to someone, and interacting with a real learner’s questions. A 2019 meta-analysis pooling 28 studies, summarized in a 2023 chapter by the learning scientist Logan Fiorella of the University of Georgia, found that each stage adds something, although not equally.
| Stage of “teaching” | What the learner actually does | Effect size (Cohen’s d) |
|---|---|---|
| Expecting to teach, but never doing it | Studies with the goal of later explaining the material; the explanation never happens | d = 0.30-0.40 vs. studying for a test, pooled across 28 studies |
| Preparing to teach, then actually explaining | Generates a real explanation (written, recorded, or spoken) for another person | d = 0.50-0.60 vs. restudying or a plain recall test, same 28-study pool |
| Teaching a dependent pupil, real or virtual | Middle-school students taught a computer avatar that would later be tested on what it had “learned” from them | Not reduced to a single d; documented as more time on task and higher learning gains (Chase et al., 2009) |
The pattern across the research is consistent even where the numbers are not. The more a learner must produce for a listener who depends on the result, the more the material gets reorganized. Reorganizing knowledge, rather than merely reviewing it, is what tends to make it stick.
Not a guarantee
None of this makes teaching an automatic shortcut to mastery, and the honest version of this story includes its limits. Fiorella’s own review notes that the expectation to teach can backfire: one early study found students preparing to teach probability problems did worse, “likely because students reported experiencing high levels of anxiety at the prospect of teaching,” and other work suggests the benefit depends on whether a learner already knows how to build a good explanation rather than simply summarize the text. Medicine’s own literature carries a similar caution. A 2014 systematic review of nineteen studies on medical students working as peer tutors found many perceived benefits for the tutors themselves, but the authors were careful to add that there was “no substantial evidence to conclude that participation as a peer tutor improves one’s own examination performance.” Teaching seems to change how a person understands material more reliably than it changes a test score, a distinction worth taking seriously before anyone prescribes it as a study hack.
Built into the job, not left to chance
Some professions did not wait for the meta-analyses. Halsted’s teaching-hospital model, for all its age, has real evidence behind it. A recent review of surgical trauma training is direct. Trainees taught this way “demonstrated superior performance” versus those trained only under complete supervision. The same review acknowledges the model is harder to defend on patient-safety grounds. Simulation increasingly comes first, before trainees ever teach on a real patient. In the United Kingdom, the expectation is written into professional standards. The General Medical Council’s Good Medical Practice states the baseline. A doctor “should be willing to offer professional support to colleagues, including students.” Mentoring, coaching, teaching and training are listed as examples. Teaching is not a specialism for university-hospital posts alone. It is a baseline expectation of being any good at the job at all.
The practitioners who explain, endure
No study promises that teaching turns a novice into an expert overnight, and the research is clear that badly run teaching, when anxious, unstructured, or unsupported, can cost more than it gives back. What the evidence does suggest is narrower and more durable: the habit of explaining what one knows to a colleague, a client, a student, or even an invented pupil who will never actually exist is itself a form of practice, with a measurable return of its own. The practitioners who last long enough to be called expert are rarely the ones who kept their knowledge to themselves and simply accumulated more of it. More often they are the ones who kept being asked to explain it to a study partner, a junior hire, a patient, or a reader and, across a career of saying yes, found their own understanding rebuilt a little more solidly each time.
Updated 3 August 2026.