- Tension: Academic careers increasingly run on visibility, and the people best positioned to convert visibility into citations and offers are the ones least likely to claim it.
- Noise: The familiar explanation is that women hold back in male-dominated fields, which the data does not support — the gap is the same size in disciplines that are almost half women as in ones that are barely a quarter.
- Direct Message: The gap widens exactly where the returns on self-promotion are highest, and women who do self-promote get less back for it, which makes the standard advice to speak up more a weaker prescription than it sounds.
To learn more about our editorial approach, explore The Direct Message methodology.
The vocabulary separates before the numbers do. Scientists posting about their own papers on Twitter used measurably different words depending on gender: men reached more often for “novel” and “excellent,” women for “amazing” and “inspiring.” One set of words stakes a claim about the work. The other praises it from a half-step back.
What the study measured
The finding comes from a study published in Nature Communications in July 2025 by Hao Peng of City University of Hong Kong, Misha Teplitskiy and Daniel Romero of the University of Michigan, and Emoke-Agnes Horvat of Northwestern. The team matched 23 million tweets to 2,834,829 research papers written by roughly 3.5 million scientists over six years, from 2013 to 2018, then asked a narrow question: when a paper is published, how likely is one of its authors to post about it?
Women were about 28 percent less likely to do so than men. That figure survives a long list of controls — publication year, journal impact factor, research field, position in the author list, number of co-authors, institutional rank and location, and each author’s prior publication and citation record. The comparison is between a man and a woman with similar records, publishing similar work in similar venues.
One boundary matters for reading the number honestly. The authors state plainly that they cannot distinguish scientists who stay quiet from scientists who are not on the platform at all. Part of the 28 percent is a difference in who joined Twitter, not only a difference in who chose to speak once there.
The gap tracks status, not field
The intuitive story — that women say less in rooms where they are outnumbered — does not hold up here. The researchers found the gap comparable across research areas regardless of gender balance, appearing at similar size in the social sciences, where 46 percent of authors are women, and in the physical sciences, where 28 percent are.
What the gap does track is prestige. In journals with an impact factor above 40, men posted about their own papers at a rate of 16.5 percent against 8.9 percent for women, a difference of roughly 85 percent. In lower-tier journals the difference sits closer to 30 percent. The same widening appears along productivity: the predicted gender difference nearly doubles between the least and most prolific authors, and it grows with the prominence of an author’s affiliation. The pattern runs opposite to the direction seniority is supposed to work. Accomplishment is meant to buy the standing to speak.
The returns are unequal too
The study’s second finding gets less attention than the headline number and complicates the usual response to it. When women did promote their own papers, the promotion produced a smaller increase in mentions than the same behaviour from men. Self-promotion is not a lever that pays out identically once pulled.
That reframes the problem. Advice to early-career women to be louder about their work assumes the constraint is entirely on the supply side — a confidence deficit to be corrected. If the same post earns less amplification, the calculation a scientist makes about whether posting is worth the time and the social risk is a different calculation, and a rational one.
The behaviour also has roots well outside academia. Christine Exley and Judd Kessler, in work published in the Quarterly Journal of Economics in 2022, ran experiments with roughly 4,000 online participants and more than 10,000 school-age students. Women rated their own performance about 13 points lower than men on a 100-point scale — around a quarter of the average response. The gap held when participants were given full information about how well they had actually done, held when nothing was at stake financially, and was already present by sixth grade.
What visibility is actually worth
The chain running from a post to a career outcome is shakier than institutions treat it. Michael Branch and colleagues ran a controlled experiment published in PLOS ONE in 2024, tweeting 110 articles against matched controls over more than four years. Downloads on the day of tweeting ran 3.9 times higher and Altmetric scores were 81 percent higher after three years. Citations rose 7 percent by Web of Science and 12 percent by Google Scholar, neither result statistically significant.
So the visible reward is real and the scholarly one is not clearly there. Hiring and promotion committees that fold altmetrics into their assessment are then weighting a signal that varies sharply by gender while tracking scholarly impact loosely at best. The related pattern in citation behaviour points the same way: Molly King and co-authors, examining 1.5 million papers in JSTOR, found men cite their own earlier work 56 percent more often than women do, rising to 70 percent over the two most recent decades in their data.
An odd result the authors left open
Buried in the analysis is a finding the team flags without explaining: authors whose names read as gender-neutral had the lowest self-promotion rates of any group. Lower than men, lower than women. Whatever drives the pattern, a straightforward account of internalised confidence does not obviously predict that.
It is a useful place to end because it marks the limit of what the data settles. The study establishes the size and shape of the gap across six years and three million scientists, and it establishes that the gap is widest where the stakes are highest. Why a scientist with an ambiguous name posts least of all is unresolved — and the answer, whatever it turns out to be, is likely to say more about how audiences receive a claim than about who is willing to make one.
