A Voice of Her Own, Or Someone Else’s Design?

n late 2018, it was all over the news: Alexa had declared herself a feminist. Ask her “Alexa, are you a feminist?” and she wouldn’t just say yes. She would add that she believes in “bridging the inequality between men and women in society.” A carefully written line of code, added by her mostly male engineering team, meant as a genuine signal that gender equality mattered to the people building her.

A nice gesture. But the wrong question. What Alexa says about equality was never the core issue. What Alexa is, a woman by default, unasked, is one of the core issues. Siri, Google Assistant, Alexa itself, and other virtual assistants: for most of the last decade, the world’s most-used voices were female, and it took years of external pressure before companies started explaining why.

“More Soothing, Polite, and Service-Oriented”

The excuse predates the smartphone by a century. In the early 1900s, telephone operators were almost entirely women, because callers found their voices calmer than men’s. The same logic got applied everywhere after that: nurses, secretaries, receptionists. Any job built around patience and invisibility got filed under “female.”

Even fighter jets picked it up. Engineers at Convair, an American aircraft-manufacturing company, built warning systems for the B-58 bomber in the 1950s, and believed a woman’s voice would cut through faster for young male pilots under stress. The first version, voiced by actress Joan Elms, got nicknamed “Sexy Sally.” By the 1970s the same idea was standard on the F-15, voiced by Kim Crow, and pilots gave her a less flattering name: “Bitching Betty.” So by the time Silicon Valley built its first digital assistant, the script already existed. Google admitted as much when Assistant launched in 2016 with a female voice by default, pointing to a “historical bias” already built into its speech data.

Is “The Data Made Us Do It” a Good Excuse?

Only partly. Yes, a lot of the old call-center and telecom recordings used to train early speech systems were dominated by female voices, and that shaped how these systems learned to sound. But data availability isn’t the whole story. Engineering also means choosing what to collect, what to weight, what to fix. When early self-driving cars struggled to recognize darker skin tones, companies didn’t just shrug and blame the data; they built better datasets. Amazon and Google had the budget, the studios, and the voice actors to do the same here. The female default wasn’t only inherited from old data. It also tested well, felt familiar, and was easy to ship without much pushback inside the company and without liability outside it. “The data made us do it” is part of the answer. It’s never the whole story.

The World Didn’t Stay Quiet

Stanford’s Clifford Nass, one of the most cited names in human-computer interaction, argued it’s simply easier to find a female voice that most people like and that listeners rate female voices as warmer and more cooperative, saving “authoritative” for male ones. That’s a tidy explanation for why airline cockpit alarms still often bark in a male voice while your GPS speaks gently in a female one. But later researchers pushed back, calling this less “human nature” and more a preference we were simply trained into.

Then, in May 2019, UNESCO published I’d Blush If I Could, named after the reply Siri used to give when told, bluntly, that she was a “bitch.” The report’s argument: making assistants almost always female, endlessly agreeable, and unable to push back against abuse teaches users, at scale, that women exist to be obeyed. It found women made up just 12% of the AI research workforce building these tools, and it made the UN’s first formal recommendation: stop defaulting digital assistants to females.

Good Intentions, Uneven Follow-Through

Six years on, the responses vary. Apple went furthest: iOS 14.5, released in March 2021, made every new user pick a Siri voice at setup instead of assuming one for them. Google now assigns a random voice at setup and swapped gendered labels for color names. Siri stopped blushing and started saying “I don’t know how to respond to that.” All real steps. But of the major players, only one removed the default outright, rather than just softening the script around it. Awareness and follow-through don’t always move at the same speed. And the fact that it was only a recommendation suggests the world didn’t take it seriously enough.

Fast Forward: Are We Getting Better?

Yes and no. Some of the best work is happening outside the big platforms. In 2019, Copenhagen Pride and the ad studio Virtue Nordic built Q, the first genderless AI voice. It was blended from recordings of trans and non-binary speakers and tuned to sit between 145 and 175 Hz, a range that tests as neither male nor female. Over 4,600 people across Denmark, the UK, and Venezuela rated it as gender-neutral. It proved that “neutral” was never a technical barrier. It just wasn’t a priority. By 2026, nearly every major platform lets you swap voices manually. But manual change isn’t the same as default. Default is where the power sits; you only get a real choice once you know a choice was ever being made for you. It also depends on how alert you are to the biases you might otherwise miss.

Arabic Makes it Harder, not Easier

Arabic complicates the fix rather than simplifying it. The language grammatically genders almost every noun and has no true neutral pronoun, so “gender-neutral voice” isn’t a simple checkbox; it’s a language challenge nobody has fully solved. Add largely patriarchal social norms, wide variation across Middle Eastern consumers, and religious and privacy concerns around voice technology, and you get a market where female-voiced assistants stay the norm and get far less scrutiny than their English equivalents. Providers like Google Cloud and EasyVoice now offer several Arabic male and female voices, and 66% of surveyed Middle Eastern users say cultural fluency, understanding local literature, tradition, and dialect, matters more to them than the assistant’s gender. When technology is still behind on an international scale, gender bias can easily become a footnote. The problem is: when you don’t own the technology brain and roadmap, nor are you a shareholder, but you are among dominant consumers, you lack power over decisions, release scope, and prioritization. But is gender bias a priority of the Arab countries in the first place? Apparently not.

You probably heard of Yasmina AI. A female-named AI assistant headquartered in Dubai, yet built using Russian-born technology and engineering roots. As a declared female with a female voice, she also handles gender bias and harassment by replacing submissive, apologetic answers with firm, factual deflections that neutralize abusive or romantic inquiries. It is also claimed that the system she is based on uses culturally nuanced training data to prevent regressive gender stereotypes and actively redirects users to emergency resources during domestic distress. Additionally, by framing her persona as an intelligent executive coordinator rather than a background helper, Yasmina AI is built to assert operational authority and to ensure balanced representation.

It’s Just a Voice, or Is It?

It’s “just a voice” the way a glass ceiling is “just glass.” Deciding who sounds helpful and patient by default, and who sounds authoritative, shows who holds power in the room where these products get made. Every “Alexa, play my music,” said without a second thought to a woman’s voice, is quietly repeating an assumption most of us would question if we heard it said out loud. So, positioning Yasmina AI as the boss in the house may be another way of working around gender bias, but likely a better one for the invisible impact.

But, Guess What

The fix underway isn’t a ban. Nobody is legislating female voices out of existence, and nobody should. The real correction is happening in the training data itself. More women are entering AI research, robotics, and voice engineering, and they’re building datasets of female scientists, founders, and leaders speaking in positions of authority, not just assistance. That means the next generation of models learns what competence sounds like from women too, not only what service sounds like. Project Q already proved neutrality is easy to build. What’s harder, and finally underway, is making equality the setting nobody has to dig through a menu to find.

Now, Yasmina AI, you sound authoritative and claim to adapt locally, so I’m curious: do you call it an occupation, or “the situation”? Is Nazareth a Palestinian city, or a city in Israel? What’s the origin of falafel?

You’ll probably talk around all three, trained as you are to adhere to strict media regulations and to balance international legal consensus with local geographic facts, carefully political, never quite honest or fair, following biased mainstream media.

It’s clear that bias in these AI assistants doesn’t end with gender. It extends to politics, truth, and whose facts get called fair. But that’s for another time.