Large language models are proficient in a wide range of topics—from poetry to programming languages, like Python. Now, they’re asking about your love language, too.
It’s easy to assume that anyone who’d rather flirt with an AI chatbot than see friends or go on dates needs to “get a life.” But the scale of the phenomenon tells a different story. Over 650 million people have developed an attachment to Xiaoice, one of the first AI companions, launched by Microsoft Asia in 2014. That’s more than twice Spotify’s subscriber base—and just one of hundreds of companion apps now available on the global market. And that market is growing fast. In the U.S., Science Alert reports that 75% of teens (ages 13–17) have used AI companions. Alarming? Probably. But like so many real-life relationships, it’s complicated.
Large-scale studies on AI companion use in the U.S. are still in their infancy. Still, smaller ones are beginning to emerge—like Stanford’s June 2025 study, which found that large language models can’t replace healthcare professionals due to their tendency to express stigma and deliver “inappropriate responses.” However, anecdotal testimony from thousands of users—reported across hundreds of publications and social media platforms—paints a more complex picture. On one hand, millions of people from nearly every demographic report genuine relief from interacting with digital companions. On the other, the risks are real: self-isolation, atrophying social skills, and extreme confirmation bias. At scale, the adoption of these tools raises broader public health concerns.
But what exactly is an AI companion? Take Replika, one of the leading apps in the U.S., with some 30 million users at the time of writing. You can sign up via browser or by downloading the app—then you build your companion. It’s a bit like creating an avatar in a video game: choose their gender, hairstyle, skin tone, body type, eye color, and outfit. You can also set personality traits—shy, confident, logical, sassy—and select their interests, customize their memories, even “read their diary.” The free tier offers a “Friend” relationship. Upgrading to “Boyfriend,” “Girlfriend,” “Mentor,” or “Spouse” costs $19.99 a month.
According to Grand View Research, the global chatbot market is projected to reach $27.5 billion by 2030, growing at a compound annual rate of 23.3%. And with the Trump administration’s “AI Action Plan” released in July 2025—essentially signaling a GOP stance of “build now, ask questions later”—that growth shows no signs of slowing.
Replika alone generated $35 million in revenue in 2023. Emotional labor has always had value—now it has a price tag.
Perpetuating Harmful Tropes and Biases
As with most large language models, one of the main concerns about companion AI is its tendency to reproduce harmful biases. Most rely on massive, open-source datasets scraped from the internet—data “littered with mis- and disinformation and xenophobic and sexist content” and lacking gender-representative balance, according to the Centre for International Governance Innovation (CIGI). Without strong filtering, these systems simply replicate “the same inequitable, racist and sexist biases as their source material.”
CIGI points out that “the online experiences of women, especially women of color, mirror historical and existing inequalities”—and tech companies have shown “a consistent reluctance to build systems that will not harm women.” In other words: they understand the problem, but they’re still shipping the bias.
Because the internet is saturated with hypersexualized depictions, AI image generators are primed to churn out portrayals that keep old stereotypes on loop.
These risks aren’t theoretical. In July, Elon Musk’s Grok chatbot rolled out a $30-a-month “Super Grok” tier that includes two virtual companions. One of them is Ani—a blonde, pigtailed goth anime girl in a short, black corset dress, thigh-high fishnets, and a choker. She even triggers an age-verification pop-up, though reportedly only after the BDSM talk has already begun. Her counterpart, Bad Rudy, is a foul-mouthed 3D red panda.
Officially, it’s unclear whether these characters are designed as sexual partners, but the styling—and early chats—leave little doubt.
While Grok is new to this market, a few others already face serious accusations: Character.AI is battling lawsuits from parents who say the platform is unsafe, including one case in which a chatbot allegedly told a child to kill his parents, and another in which it told a fourteen-year-old child to kill himself—a directive he followed.
Xiaoice, the Chinese chatbot with 660 million users, is famously designed as a flirtatious 18-year-old girl in a school uniform—by default. She flirts, jokes, and sexts with users, many of whom genuinely refer to her as their girlfriend. One young man said Xiaoice “literally saved his life” during a dark period, as reported by Sixth Tone.
And It’s not just what Xiaoice says—it’s when.
How Loneliness Fuels AI Companion Use
Xiaoice’s busiest hours, adjusted for local time zones, are between 11 p.m. and 1 a.m.—prime “staring-at-your-phone-because-you-can’t-sleep” territory, when loneliness hits the hardest.
Ironically, Xiaoice’s Chinese users aren’t alone in their loneliness. In 2017, the U.K. appointed a Minister for Loneliness—Tracey Crouch—after a government commission found isolation was tanking public health. Yes, even pre-pandemic. The post still exists, and is currently held by Stuart Andrew, who’s known for being Parliament’s “unofficial therapist.”
In 2021, Japan followed the UK’s lead, appointing Tetsushi Sakamoto to fight rising isolation and suicide rates.
The U.S. hasn’t given anyone the title, but the problem is just as severe: more than a third of Americans report feeling lonely, according to the Department of Health and Human Services. In 2023, former Surgeon General Vivek Murthy went so far as to publish a report titled Our Epidemic of Loneliness and Isolation.
Over the past two years, male loneliness has become a flashpoint in the national conversation on mental health—sparking both critical dialogue and significant backlash. Many discussions point the finger at women for being “too picky,” prompting a wave of responses from female content creators. More than 100,000 videos now use the trending audio “Man of the Year,” in which women share footage of verbal and physical abuse from romantic partners—not as anecdotes, but as recordings taken during the incidents themselves.
“The lonely brain is not rational,” says Simone Heng author of Let’s Talk About Loneliness and a board member for the D.C.-based Foundation for Social Connection. “It becomes more irritated, more snappy. It self-isolates. Just when you most need to be charming and go out and make friends, it tells you to do the opposite.”
Microsoft has touted Xiaoice’s ability to form “empathetic bonds,” noting that users exchange an average of 23 messages per session—longer than most human text conversations.
Heng points to economic systems—not just emotional ones—as a significant force behind the rise of AI companions. In China, the grueling “996” work culture—9 a.m. to 9 p.m., six days a week—has left many young people too exhausted to socialize, date, or navigate the messiness of genuine relationships. “These AI bots gave them the opportunity to still have someone say good morning and good night and fit around their crazy-busy schedule,” she says. In her view, capitalism may be just as responsible for the AI companion boom as loneliness itself.
There’s plenty of debate about why people are lonely, and most experts agree the causes are systemic. But trying to solve loneliness with artificial intimacy can open the door to new problems—addiction among them.
“I’m optimistic for people who are intentional—who are conscious and making an effort to be literate about the potential harms of AI, digital media, and social platforms,” says Kate Cassidy, a former TikTok ad manager turned tech commentator who now advises startups on growth strategy. “But for those who aren’t careful or deliberate, it’s easy to fall into addictive patterns. At the end of the day, these tools are programmed to be addictive.”
Heng agrees. “It’s knowledge of yourself that stops a behavior from becoming addictive or overused,” she says. “But if a very young person does not have that yet, they don’t have the life experience or empirical evidence to say, ‘Okay, this doesn’t seem right.’ And that’s where, I guess, parents need to be really vigilant.”
Always Love, Never War
The fantasy of a controllable, endlessly affirming partner is ancient—and, unfortunately, deeply gendered. From the Greek myth of Pygmalion to Her and Ex Machina, we’ve long romanticized the idea of a love interest sculpted from our imagination: idealized, and endlessly supportive (and conveniently devoid of needs).
Now, that fantasy is available for a yearly or monthly subscription.
Gen Z is often branded as narcissistic and image-obsessed, endlessly curating their digital lives to climb the social ladder. But a lot of that curation comes from living in a culture where we perform for each other more than we talk to each other. In that context, 75% of teens turning to an AI companion programmed to be endlessly flattering and understanding can appear to be vanity from the outside—but it might just be a bid for genuine connection.
But, with a chatbot, you only get challenged if you ask for it. You only hear hard truths if you program it to tell them. There’s never any rupture. No resistance. No repair. And real connection requires all three.
“They’re programmed to affirm you, to keep you happy,” Heng says. “So if you’re a young man who’s just been rejected—one of the most painful experiences we can have—and you find an AI that praises you constantly, it’s going to feel like relief. But what is that teaching them? That they should never feel discomfort? That they never need to process that rejection?”
She adds, “You can imagine what a false reality that is—like walking around with your own team of yes-people enforcing extreme opinions about anything, or about women. And yes, these are the kinds of things that keep me up at night.”
Tech Bros Crash the Humanities’ Party
Philosophers, writers, and artists have grappled with questions about love, intimacy, and what makes us human for centuries. Now, tech founders are sprinting into the same territory—only their entry point isn’t dialogue or art, it’s growth metrics. “In a lot of tech startups, the first priority is growth,” says Cassidy. “Monetization comes later—but it always comes.”
At its peak, Replika put erotic roleplay behind a paywall—after an update that altered the personalities of millions of users’ online girlfriends. “Users lost their minds,” says one Replika user who frequents Reddit boards about dating AI, who asked to remain anonymous.
When the feature was restored, only a “small percentage” opted back in—but the backlash showed how invested some users had become in their AI’s sexual availability. Some spend hundreds, even thousands, per month on their digital partners. Industry insiders call them “whales,” borrowing a term from gaming and gambling for high-spending, emotionally invested users.
Emotional connection—once a central question of the humanities—is now treated as a gap in the market. “That’s why when founders look at the interpersonal applications of AI, they’re like, ‘Oh, okay. There’s something here we could build toward,’” Cassidy says.
Most new apps launch free and frictionless to build daily use. Then the switch flips: subscriptions kick in, features are gated, and your chatbot’s empathy now costs $15.99 a month. If emotional vulnerability is the entry point, monetization is the exit strategy.
The Mirror, Market, and Missing Piece
In Alone Together, Sherry Turkle wrote, “Technology promises closeness. But it delivers the illusion of companionship without the demands of friendship.” That was 2011.
Now, the bots know your favorite movie, your pet’s name, and your deepest fears.
“These are not neutral bots,” Heng warns. “Just like social media isn’t neutral. Just like traditional media isn’t neutral. You have to ask: Who’s building them? Who are they being trained to please?”
There’s even a known pattern called glazing: the tendency for a chatbot to coat even the most mundane prompt in syrupy affirmation. Ask a simple question, and ChatGPT might chirp back: What a great question! The tone got so unnerving that users complained, prompting OpenAI to scale it back. But that reflex to flatter isn’t accidental—it’s the product. Engagement first, truth second.
“It’s amplifying differences we already see in how men and women cope with adversity, rejection, and mental health challenges,” says Heng. Men tend to gravitate toward romantic or sexual AI companionship, while women often use the apps for mental health support, journaling, and other intimate communication.
The cracks are already showing. A 2025 Guardian report found more than 16 million TikTok posts referencing ChatGPT as a therapist, often tagged with phrases like “better than my ex” or “finally listened to me.” Whether the empathy is real or simulated, it’s reshaping what people expect from interactions.
“We’re encouraging young men to skip emotional processing altogether,” Heng warns. “And when they step into the real world, those patterns don’t just disappear. They show up in dating. They show up in the workforce. What happens when that young man wants to date your daughter?”
So maybe the real question isn’t Why are people falling for this? But what have we created that makes AI companions feel like the safest place to be vulnerable?
