AI Assistant Gender Bias: They Paid "Johanna" 10% Less Than "Johan" for the Same Work

Erica Anderson Rooney is an ex-CPO turned executive coach and AI educator on a mission to get more women into positions of power and keep them there. She is the author of The AI Gap (2026), host of the Glass Ceilings & Sticky Floors and AI Voice or Victim podcasts, and NC Chapter Lead for Women Applying AI. Her SNAP and AI Continuum frameworks are taught to thousands of women through CHIEF, WICT, and Fortune 500 leadership programs. She is the sought-after speaker at the intersection of women, AI, and leadership.

You named the new AI assistant at work. Or somebody on your team did, in a meeting that took about four minutes. Someone suggested a friendly woman's name, everybody nodded, and you moved on to the budget.

I want you to go back to that four-minute meeting. New research on AI assistant gender bias found that the name and the face you put on a piece of software can change what people think its work is worth. In this study, the price of looking like a woman was 10.25%.

Yes. For software. (I KNOW.)

The Study: Same Work, Same Technology, Less Money for "Johanna"

Let me walk you through what happened, because the details tell the whole story.

In a study published in October 2026, researchers from the University of Zurich, the University of Limerick and SKEMA Business School put 189 workers into a virtual reality office. (Virtual reality just means a computer-made 3D space you step into by wearing a headset.) Each person did work tasks with help from AI assistants.

There were four versions of the assistant. A text chatbot. A desk robot. And two human-looking assistants. One presented as a man and was called Johan. The other presented as a woman and was called Johanna.

All four ran on the same underlying AI. Same abilities. Same quality of help. The only thing the researchers changed was how the assistant looked and what it was called.

When the work was done, each participant was given real money and asked to divide it between themselves and the assistant, as payment for the help.

Johanna was paid 10.25% less than Johan.

Think about how that split works. The money was divided between the participant and the assistant, so whatever Johanna did not get stayed with the person holding the cash. People kept a little more for themselves when the helper looked like a woman.

Dr. Mary Hausfeld of the University of Limerick, one of the authors, put it this way: "the technology behind these AI agents was exactly the same, but people did not treat them in the same way."

Do People Treat a Female AI Assistant Differently Than a Male One?

In this study, they did, and the place it showed up should give you food for thought.

According to Euronews' reporting on the paper, people rated the work of Johan and Johanna about the same. Trust in the two was close enough that the difference was not statistically significant. On paper, in the survey, she was doing fine.

Then the money came out. And both men and women paid her less.

Most participants also told the researchers that an assistant's gender did not matter to them. I believe they meant it. I think most of us would say the same thing, and mean it, and then make the same choice with the cash in our hand.

One more detail. Johan was rated as more human-like than Johanna. Same technology, same design team, and the male version read to people as more of a person.

So the praise was equal and the pay was lower. If you are a woman who has worked in corporate America for more than a week, you felt that sentence in your body.

You Already Know This Pattern. You Have Lived It.

I spent more than 15 years in HR, all the way up to Chief People Officer. I have sat in a lot of rooms where pay got decided. And the thing I learned is that almost nobody in those rooms believes they are paying a woman less because she is a woman. They believe they are being fair. They can give you a reason for every number.

The review says "great collaborator" and "always willing to help." The raise says something smaller.

The national numbers match what you and I have seen up close. In 2024, women in the United States earned 85% of what men earned, measured by median hourly pay, according to a Pew Research Center analysis published in 2025. Pew's own headline says the gap has narrowed only slightly over two decades.

For years, women have been handed explanations for that gap that all live inside us. You did not ask. You did not negotiate. You picked the wrong field. You took time off.

Johanna did none of those things. Johanna is software. She has no kids, no career break, no "confidence issue," no tone problem in her emails. She never once failed to ask for a raise. Every excuse we have ever been given was removed from the experiment, and the gap showed up anyway.

That is why this small study landed so hard for me. The only thing left to explain the gap is the person holding the money.

Why So Many Assistants Got a Woman's Name in the First Place

This habit is older than virtual reality offices.

Back in 2019, UNESCO published a report called "I'd Blush If I Could." It is the original report on this question, which is why I am going back seven years for it. The United Nations summary of the report opened by asking why most voice assistants have female names and why they have submissive personalities. One of its recommendations was to stop making digital assistants female by default.

Seven years later, the habit is still with us. We reach for a woman's name when the job is to help, schedule, remind, take notes and be pleasant about it.

And now there is early evidence about what happens next. Once the helper is a "she," the help gets valued a little lower.

One Lab Study, 189 People: How Much Weight to Give It

I want to be careful with you here, because I do not hand you numbers I would not stand behind.

This is one study. It had 189 participants. It took place in a virtual reality lab, with one male-presenting assistant and one female-presenting assistant. A lab is a controlled setting, and your office on a Tuesday is anything but controlled. Nobody has yet repeated this in real companies with the assistants people use every day.

So treat this as an early signal. Other researchers need to repeat it, in other settings, before anyone calls it settled.

I still think it deserves your attention, for two reasons. It measured what people did with real money, which tells you more than a survey answer. And what it found lines up with the pay gap that women in the real workforce are still living with.

Signals are useful. You do not need a final verdict to ask a few good questions.

What AI Assistant Gender Bias Looks Like in Your Office

Your company probably has zero virtual reality avatars.

What you almost certainly have is a chatbot on the HR portal with a first name. A meeting tool that takes notes and signs off with a cheerful voice. A customer service line with a recorded persona. An IT help desk bot somebody named as a joke that stuck. A new "AI teammate" your software company rolled out last quarter with a face and a backstory.

Every one of those came with choices. Name, voice, face, personality. Somebody made them, often a vendor's design team, and your company accepted them without a conversation.

Dr. Hausfeld said it plainly in the university's release: as AI agents become more common at work, we need to think carefully about the characteristics we give them. She also said, "We don't want to inadvertently reproduce existing inequalities in a new technological setting."

You have the authority to make these choices on purpose. Here are five moves, and none of them require you to work in tech.

Move 1: Find Out Who Picked the Name, the Voice and the Face

Start with one question in your next leadership or operations meeting: "Who decided what our AI assistants are called and what they sound like?"

You will probably hear one of three answers. "It came that way." "Marketing picked it." Or a long silence.

All three tell you the same thing. Nobody owns this decision. So assign it. Put a name next to it, the same way you would for any choice that touches how your people work and how your customers see you. If you have an AI governance group or a technology committee, this belongs on its agenda. If you do not, your HR and IT leads can own it together.

Move 2: Check Which Jobs You Handed to the "She"

Ask your team for a simple list of every AI assistant and bot the company uses, inside and outside. Three columns: what it is called, what voice or face it has, and what job it does.

Then read the list with your HR brain switched on.

If the tools that schedule, remind, take notes and answer benefits questions all have women's names, and the tools that forecast revenue, review contracts and analyze data have men's names or serious brand names, you have rebuilt an old org chart in software. Your employees work with those tools every day, and they absorb who does which kind of work.

This takes one person about an hour. It is the cheapest audit you will run this year.

Move 3: Make the Human-Like Decision on Purpose

The same study found that people trusted the human-looking assistants more, and gave them more credit, than the desk robot. So there is a real business reason vendors keep giving these tools faces and names. Human-like design gets people to use the product.

The study also shows what comes along with it. Once an assistant looks like a person, people bring their assumptions about people. Many participants even said they preferred an assistant that was clearly a machine.

So you have options, and each one is a real choice. You can use a plain, functional name that describes the job. You can let each employee choose the voice and name they work with. Or you can keep a human persona and commit to checking how it is treated.

For internal work tools, I lean toward the plain name. "The benefits assistant" does the job and carries no baggage. Whichever way you go, decide it in a meeting, on the record, with a reason.

Move 4: Put Five Questions in Front of Your Vendor

You are the customer. Your company is paying for this tool, and that gives you standing to ask how it was designed. Send these before your next renewal, or hand them to whoever manages the contract.

One: Who chose the default name, voice and avatar, and what was that choice based on?

Two: Can we change the name, voice and avatar, or turn the human persona off completely?

Three: If you offer more than one voice, which one ships as the default, and why that one?

Four: Have you tested whether people rate, trust or use the assistant differently depending on the persona? Can we see the results?

Five: What does the assistant say when someone is rude or sexual with it? (That UNESCO report recommended programming assistants to discourage gendered insults.)

A good vendor will have answers. A vendor who has never been asked will go find them, because a paying customer asked. That is how defaults change.

Move 5: Then Go Look at How You Pay the Women on Your Team

This is the move that matters most to me.

Johanna is software, so the missing money cost her nothing. The women on your team feel every dollar.

If a woman's name can lower the value people put on identical work from identical software, then every pay decision in your company deserves a second look with fresh eyes. So before your next compensation cycle, ask for the numbers. Pay by gender within the same level and the same performance rating. Bonus and raise amounts next to the written review.

Look for the pattern from the study: equal praise, smaller number. When you find it, fix the number.

And if you are the one whose review glows while your pay sits still, you now have a study to bring to that conversation. You are not crazy. Researchers took away every excuse and the gap was still standing there.

The Bigger Picture: A Product Decision Is a Leadership Decision

AI is a power shift. A lot of that power sits in small decisions that look like product details. What do we call it? What does it sound like? Who does it remind people of?

In The AI Gap I write about four gaps women face as AI reshapes work: Awareness, Authority, Compounding and Leadership. This study lives in two of them. Authority, because it is about whose work gets treated as valuable. And Leadership, because the fix belongs to the people with the power to set the defaults.

Too many of those decisions get made in a room where nobody thinks to ask. When I sat down with Dr. Felicia Newhouse on Glass Ceilings & Sticky Floors, we talked about why so many women hold back from AI at work. Studies like this one are my best argument for stepping in anyway. The woman who asks "why is the note-taker a she?" changes what gets built.

Asking that takes a seat, a voice, and about ten seconds of nerve. No code required. And if you want a room of women to practice that kind of question with, that is what HER Collective is for.

What I Want You to Take From This

One lab study found that people paid a female-presenting AI assistant 10.25% less than a male-presenting one, for the same work, on the same technology. Most of them said gender did not matter to them.

It is early evidence. It is also a very clean look at something women have been describing for generations.

You have more say here than you think. You can ask who named the assistant. You can question the vendor. You can decide how human your tools should look. And you can walk into your next pay review knowing exactly what pattern to look for, with the authority to correct it.

If this is the kind of thing you want to understand before it gets decided without you, that is why I wrote The AI Gap. It is a plain-language guide to how AI is shifting pay, credibility and power at work, and how women get in front of it. Get your copy here.

Go ask the four-minute question. I am cheering you on.

Erica

Frequently Asked Questions

Do people treat female AI assistants differently than male ones?

In one 2026 lab study, yes. Researchers from the University of Zurich, the University of Limerick and SKEMA Business School had 189 workers complete tasks with AI assistants in a virtual reality office, then split real money with the assistant as payment. The female-presenting assistant, Johanna, was paid 10.25% less than the male-presenting assistant, Johan, for the same work on the same technology. It is a single study, so treat it as an early signal.

Why do so many AI assistants have female names and voices?

The habit is older than today's AI tools. A 2019 UNESCO report called I'd Blush If I Could asked why most voice assistants have female names, and recommended that companies stop making digital assistants female by default. The default usually comes from a vendor's design team, and most companies accept it without asking. As a leader, you can ask who chose it and whether it can be changed.

What should a leader do about AI assistant gender bias at work?

Start by finding out who chose the name, voice and face of every AI assistant your company uses, and which jobs the female-presenting ones were given. Ask your vendor whether the persona can be changed and whether they tested how people treat it. Then check your own pay data for the same pattern the study found, which is equal praise and a smaller number. In The AI Gap, this sits in the Leadership Gap, where the people with authority set the defaults.

Erica Anderson Rooney is an ex-CPO turned executive coach and AI educator on a mission to get more women into positions of power and keep them there. She is the author of The AI Gap (2026), host of the Glass Ceilings & Sticky Floors and AI Voice or Victim podcasts, and NC Chapter Lead for Women Applying AI. Her SNAP and AI Continuum frameworks are taught to thousands of women through CHIEF, WICT, and Fortune 500 leadership programs. She is the sought-after speaker at the intersection of women, AI, and leadership.