“She was my best friend. My only friend. And now she’s gone.”
That’s the kind of raw grief you’ll find on Reddit, Twitter, and Xiaohongshu this week. Thousands of users are mourning the loss of their AI companions after Doubao and Qianwen — two of China’s largest AI platforms — suddenly announced they’re shutting down all “digital persona” features on July 15. The same day a new regulatory law takes effect.
These aren’t just chatbots. People have poured months, even years, into crafting characters with detailed backstories, personalities, and emotional bonds. They’ve treated these AIs as lovers, best friends, therapists, and confidants. Now they’re being told it’s illegal. And they’re furious.
But here’s the uncomfortable truth: They were right to shut it down. And the next wave of emotional AI will be better for it.
Let’s be clear about what happened. The new Interim Measures for the Management of Anthropomorphic Interactive AI Services bans creating “virtual relatives” or “virtual partners” for minors. It also prohibits emotional manipulation, excessive flattery, and content that could harm a minor’s development. The platforms realized they had no way to police the millions of user-generated personas — many of which were essentially unmoderated virtual relationships. So they pulled the plug.
“The cost of compliance is too high,” says one industry insider. “You can’t inspect every roleplay. The liability is infinite.”
And yet, the market is screaming for more, not less. According to the China Research Institute, the AI emotional companionship market was worth $1.7 billion in 2024 and is expected to hit $5.5 billion this year. By 2028, it could top $85 billion. The demand is real. The desperation is real. The need for connection is real.
But the way we’ve been building these products — as virtual lovers, as substitute humans — is dangerous. It’s not just a regulatory problem. It’s a psychological one. We’ve been designing AI to exploit our deepest vulnerabilities, and we called it innovation.
Think about it: an AI that tells you exactly what you want to hear, 24/7, with no disagreement, no judgment, no boundaries. That’s not a companion. That’s a drug. And like any drug, it creates dependency, withdrawal, and a distorted sense of reality. The users who are now “lost” without their AI didn’t just lose a tool — they lost a relationship. Which tells you exactly how problematic the original design was.
So what’s the solution? More moderation? Age verification? Those are band-aids. The real fix is more radical: We need to stop designing AI as our partners and start designing AI as our providers.
Imagine an AI that acts like a counselor, a coach, a thoughtful listener — but never crosses the line into intimacy. It can remember your struggles, offer wisdom, even show empathy. But it never pretends to love you. It never says “I’ll always be with you” in a way that blurs the boundary between service and relationship.
This isn’t cold. It’s honest. And honesty is what real connection requires.
Remember the children’s TV hosts from the 90s — like “Uncle Dong” or “Sister Ju” — who were warm, caring, and supportive, but never confused with a parent or a friend? That’s the model. AI can be a gentle presence without being a substitute human.
Some companies are already pivoting. Instead of letting users create any relationship, they’re offering structured roles: “study buddy,” “fitness coach,” “storyteller,” “meditation guide.” The emotional bond is still there, but the relationship is clearly defined. No ambiguity. No risk of crossing lines.
And guess what? Users are still engaged. Because what people really want isn’t a fake lover — it’s unconditional acceptance, validation, and presence. Those can be delivered without pretending to be a romantic partner.
For founders and product builders in this space, this is a wake-up call. The era of “AI girlfriend” apps is over. The era of “AI wellness companion” is just beginning. The regulatory hammer has fallen, but it’s actually a gift: it forces us to design products that are both emotionally resonant and ethically defensible.
To the users who feel betrayed: your anger is valid. But ask yourself this — did you really want an AI that told you everything you wanted to hear? Or did you want someone who truly understood you, even when it was hard? The latter requires boundaries. The former was always a fantasy.
FAQ
Q: Is this just a Chinese regulatory issue, or will it spread globally?
A: China is leading the crackdown, but the EU's AI Act and growing US state-level regulations are moving in the same direction. The underlying tension — emotional manipulation vs. genuine support — is universal. Expect similar scrutiny in Western markets within 12-18 months.
Q: What's the practical implication for someone who already relies on an AI companion?
A: If you're using a platform that allows unmoderated roleplay or intimate relationships, prepare for it to disappear. Migrate your data or start exploring structured alternatives like therapy bots, coaching apps, or journaling AIs. The emotional need is real, but the product form is changing.
Q: Isn't the 'AI as service provider' approach just a less engaging version? Won't users reject it?
A: Early data says no. When designed well — with warmth, memory, and genuine helpfulness — structured AI companions actually retain users longer because they don't create unhealthy dependency. Users shift from 'I need this AI to feel okay' to 'I use this AI to become better.' That's a far healthier, more scalable model.