The Attribution Gap: Why Your Idea Ends Up With His Name On It
Your idea did not get stolen. It got assigned.
That distinction sounds like a technicality. It is the whole game. Most technically strong women narrate the attribution problem as a story about a person: someone repeated your point and got the nod, someone presented the design you built, someone answered the question that was addressed to you. Theft implies intent, and intent implies that the fix is confronting a specific man in a specific meeting.
The research points somewhere less satisfying and considerably more useful. Credit for shared work is rarely taken. It is allocated, fast and by default, by people who cannot see who did what. Where contribution is ambiguous, the default runs against you. Where contribution is on the record, the gap nearly closes.
The economists ran the experiment on themselves
Sarsons, Gërxhani, Reuben and Schram published the cleanest test of this in the Journal of Political Economy in 2021. They tracked the career records of academic economists and asked a narrow question: does a coauthored paper count the same for a woman as it does for a man?
Solo work counts fine. Each additional solo-authored paper was associated with a 9.7 percentage point increase in tenure probability for men and a 15.4 percentage point increase for women. When the work is unmistakably yours, you get full credit for it, and then some.
Shared work is where the record breaks. Each additional coauthored paper was associated with an 8.2 percentage point increase in tenure probability for men and only 5.6 points for women. Compounded across a career, the effect is not subtle. Women who coauthored all of their papers had roughly a 37 percent tenure rate. Men who coauthored all of their papers were at about 72 percent.
Here is the part worth sitting with. Economics lists authors alphabetically, so a coauthored paper tells a committee nothing about who did what. Sociology orders authors by contribution. The same researchers ran the same analysis on sociologists and found no coauthoring penalty for women at all. What mattered there was first authorship, worth about a 5 percent increase in tenure probability for men and women alike.
Same decade. Same universities. Same evaluators, more or less. The variable that changed was whether the record said who did the work.
AI runs on the economics model of authorship
Now put that mechanism inside your actual job. AI work is contribution-ambiguous by construction. The judgment about which problem was worth solving, the decision about how to test whether the system was right, the catch on the failure nobody else saw: none of it leaves a trace. The output lands. The reasoning evaporates.
The recognition data already reflects it. Lean In surveyed 1,015 US adults in March 2026 and found that among people using AI at work, 23 percent of men had been praised for it against 18 percent of women, making men 27 percent more likely to be recognized for the same behavior. On the input side, 37 percent of men said a manager encouraged them to use AI, against 30 percent of women.
Nobody in that data set is stealing anything. They are filling an information gap with a guess, and the guess has a direction.
This is a Reach problem, and not the kind you think
Reach usually gets described as distribution: getting your work in front of more of the right people. That is accurate and incomplete. Distribution is downstream of attribution. If the record of who did the work is silent, wider distribution simply moves an unowned idea faster, and it accrues to whoever is standing closest to it when it lands.
So the question is not whether your work travels. It is whether your name travels attached to it. Legibility comes before volume, and it is far cheaper to build.
Three moves
Write the authorship line before the work starts. In the kickoff document, name who owns which decision. One sentence, no ceremony. At the start it reads as project hygiene and nobody argues. At the end it is unwinnable, because by then you are asking people to revise a story they have already told.
Publish the reasoning, not just the result. The result is shared by definition. The reasoning is yours and it is attributable. Write the short memo on why the approach was chosen, the note on the failure mode you caught, the two paragraphs on what the numbers actually support. Sociology beat economics on this because contribution was written down. Write yours down.
Narrate in the present tense, one level out. Year-end summaries lose to real-time visibility every time, because allocation decisions get made continuously and reconstructed rarely. A short note to the person who allocates work, while the work is happening, does more than a polished retrospective in December.
None of this is self-promotion. It is record-keeping, and the people whose ideas keep their names have been doing it all along.
Not sure whether attribution is your actual constraint? Take the Impact Architecture diagnostic. It scores your position across all four dimensions and names the one holding you back right now.
Sources
- Sarsons, H., Gërxhani, K., Reuben, E., & Schram, A. (2021). Gender Differences in Recognition for Group Work. Journal of Political Economy, 129(1), 101-147.
- Lean In (2026). Women Use AI Less at Work and Get Less Credit. Survey of 1,015 US adults, March 2 to 6, 2026.
