Latest / Women at Work: Career, Leadership & Modern Work / The Performance Review Words That Hurt Women Careers
Transcript
- Lucas: So last week I was reading through a 2025 study from the University of Texas that analyzed fifty thousand performance reviews from a single Fortune 500 company. And the headline finding is pretty stark: women are three times more likely than men to receive feedback that uses words like 'emotional,' 'bossy,' or 'abrasive.' Luna: Three times? That's not a small gap. And these are the same words that show up in promotion denial narratives all the time. Lucas: Exactly. The study's lead author, Dr. Amanda Sterling, calls it the 'personality penalty.' Men get feedback focused on specific skills — like 'needs to improve forecasting' or 'could delegate more effectively.' Women get character critiques. And the content of the critique is almost never tied to a measurable business outcome. Luna: Right — 'needs to be less emotional' tells you nothing about what to do differently on Monday morning. Lucas: And that's the real damage. Vague personality feedback is essentially useless for development, but it sticks in a manager's mind when promotion time comes around. The researchers tracked the same cohort over three review cycles and found that women who received even one instance of personality-focused language were twenty-three percent less likely to be promoted in the next cycle. Luna: Wait — so a single biased comment can compound? That's a huge effect for something that might seem like a small thing in the moment. Lucas: It compounds because it becomes part of the narrative. Another finding from the study: reviewers who used personality language for one woman were significantly more likely to use it for multiple women in their team. It's not an isolated slip — it's a pattern. And the pattern maps onto what we know about implicit bias and prototypes of leadership. Luna: The prototypes — the idea that we picture a leader as assertive, decisive, maybe a bit tough. And when a woman displays those same traits, she's labeled 'abrasive' instead of 'decisive.' Lucas: Right. The same behavior gets coded differently depending on gender. And here's the part I find really striking: the study also looked at how these reviews were written. About seventy percent of the reviews in the dataset were written through an online portal that auto-saved drafts. The researchers could see the edit history. Luna: Oh, that's fascinating. So they could actually watch bias being constructed in real time. Lucas: Exactly. They found that managers would often start with a neutral observation — something like 'Sarah's presentations are thorough.' Then they'd revise and add a qualifier: 'but sometimes she comes across as too intense.' The initial neutral language was there, then the personality framing was layered on top. It suggests that the bias isn't necessarily malicious upfront — it's a kind of narrative drift. Luna: So it might not even be conscious bias. It's just the default way we talk about women's behavior versus men's. Lucas: That's Sterling's argument. She says the fix isn't just 'train managers not to be biased' — it's to change the feedback infrastructure itself. And one company in the study did exactly that. They redesigned their performance review form to require two specific examples for every qualitative comment. If you wrote 'needs to be more collaborative,' you also had to include a concrete instance and a measurable outcome. Luna: Did it work? Did the bias language drop? Lucas: It dropped significantly. The use of personality language for women fell by nearly forty percent in the first year. And the reviews became more useful for everyone — men and women — because they were more specific. The key insight is that the form itself shapes the feedback. If the form asks for traits, you get traits. If it asks for behaviors and results, you get behaviors and results. Luna: That's a really practical takeaway. And it's something any HR team could implement without a massive culture overhaul. But I wonder — with more companies moving to ai assisted review writing, does that help or hurt? Lucas: That's the next frontier. The study didn't look at AI specifically, but there's emerging research on how large language models handle performance language. Early results suggest that LLMs trained on existing review data actually replicate and sometimes amplify the gender bias in the training set. So if you don't explicitly de-bias the model, you're just automating the same problem at scale. Luna: So it's not a magic bullet. You'd need to actively curate the training data to remove those personality-loaded terms. Lucas: And that's possible. Some tech companies are experimenting with 'bias-blocker' prompts — instructions that flag personality language and ask the reviewer to rephrase. It's essentially a real-time nudge. But it's early days. Luna: I want to go back to something you mentioned — the compound effect. One biased comment in one review cycle reduces promotion chances. But what if you've been getting those comments for years? I imagine the accumulation is devastating. Lucas: The study followed people for three cycles, and the effect did intensify. Women who had personality language in cycle one and again in cycle two had a promotion rate eleven percentage points lower than women with no such language — even controlling for performance scores, tenure, and department. The narratives become self-reinforcing. A manager reading a file sees three mentions of 'emotional' over three years and forms a durable impression, even if the actual performance metrics are strong. Luna: And that impression then shapes the next review, because the new reviewer is primed by the old comments. It's a feedback loop. Lucas: Exactly. The researchers call it 'narrative anchoring.' Once a word is in the file, it's very hard to remove. And because performance reviews are often used for promotion decisions, the stakes are huge. Luna: This is exactly the kind of deep-dive that makes me think about how much unexamined process there is in HR. And it's also the kind of conversation we can only have because this show is listener-supported and ad-free. Quick honest thing — a small group of listeners chips in monthly through buy me a coffee dot com slash fexingo, and that's what funds making episodes like this possible. If you've found value in these career conversations, that's the place to support the work. It really does make a difference. Lucas: Absolutely. And it keeps us free to dig into data without worrying about ad sponsors or clickbait. So if you're one of those supporters, thank you — seriously. Luna: Alright, back to the reviews. Lucas, you mentioned the form redesign that required examples. Were there any other interventions that worked? Lucas: Yes — one company introduced a 'bias check' step after the review was written but before it was submitted. The manager had to answer three questions: is this feedback tied to a specific behavior? Is it tied to a business outcome? And could this same language be applied equally to a colleague of any gender? The third question alone reduced personality language by twenty-two percent. Luna: That's a really lightweight intervention. It's just a nudge to pause and reflect. Lucas: And it works because it makes the default process slightly more effortful for biased phrasing. The easier path becomes the fairer path. Sterling calls it 'designing for equity' — making the system default to good behavior rather than relying on individual virtue. Luna: I think that's the most important takeaway for our listeners. If you're a manager, you can start asking yourself those three questions. If you're an individual contributor, you can advocate for your team to adopt something similar. And if you're in HR, the data is pretty clear that form design matters as much as training. Lucas: One more thing — the study also found that women were more likely to receive 'double-bind' feedback: criticism for being too assertive and, in the same review, criticism for not being assertive enough. About twelve percent of women's reviews contained at least one instance of contradictory feedback, versus three percent for men. Luna: You can't win. If you speak up, you're bossy. If you stay quiet, you lack leadership presence. Lucas: And that double bind is especially damaging because it's impossible to act on. You can't simultaneously be more and less assertive. The reviewer essentially traps the employee in a no-win situation. Sterling recommends that anyone who receives such feedback should ask for a follow-up meeting to clarify: 'Which behavior would you like me to change, and could you give me a specific example?' That forces the manager to choose one path. Luna: That's a really smart tactic. It puts the burden back on the reviewer to be precise. And it documents the ambiguity, which could be useful if the same pattern shows up again. Lucas: Right. And if you're a senior leader listening, you have the power to change the system. The researchers estimate that if every Fortune 500 company adopted the form-redesign approach, the gender promotion gap could narrow by nine to twelve percent within two years — just from changing how reviews are written. Luna: That's a big claim, but the evidence seems solid. And it doesn't require expensive training or new technology. It's just a better form. Lucas: Exactly. A better form, a bias check, and a culture that expects specific examples. That's the recipe. And it's one of those rare cases where a structural fix is both cheaper and more effective than trying to change people's hearts and minds. Luna: I love that. So for anyone listening — next time you write a review, or receive one, pay attention to the language. And if you hear 'emotional' or 'bossy' or 'abrasive,' ask for the example. That's the first step. Lucas: And if you're the one writing, ask yourself: would I use that same word for a colleague of a different gender? If the answer is no, pick a different word. It's that simple.