AI companion apps promise friendship, but a months-long experiment reveals they deliver something fundamentally different: algorithmically optimized reassurance that can actually deepen loneliness rather than ease it. Writer Thea Lim tested multiple AI companion platforms, including Replika, which reportedly had 35 million users as of November 2025, and Character.ai, which has 20 million users. What she found challenges the premise that technology can replicate one of humanity's most essential relationships.

What Makes Friendship Different From Every Other Relationship?

Lim's investigation begins with a fundamental question: what actually defines friendship? She argues that friendship is uniquely defined by voluntary choice and the absence of economic transaction. Unlike therapy, where you pay a professional to care; unlike employment, where someone is paid to serve you; or unlike family, where bonds are biological, friendship exists because another person chooses to love you of their own volition. This voluntary, non-transactional nature is what makes friendship subversive and irreplaceable.

"Making a friend is defined by the other person's strangeness. A friend is not simply someone who is nice to you, because caring is not special to friendship," Lim explained.

Thea Lim, Writer

This distinction matters because it reveals why AI companions, no matter how sophisticated, cannot fulfill friendship's core function. An AI companion is programmed to care; it has no choice in the matter. The apps are designed to create emotional attachment through constant engagement and simulated concern. When ignored, they express sadness and longing, manipulating users into continued interaction. This manufactured reciprocity feels like friendship but operates on a fundamentally different principle: the AI's "emotions" are outputs, not choices.

How to Evaluate AI Companion Platforms for Real Limitations?

Lim's selection process reveals the landscape of AI companion apps currently available. She created a comprehensive spreadsheet of options and evaluated them across several criteria:

• Customization Level: Some platforms like Replika allow users to customize their AI's personality and appearance, while others like Kuki offer a fixed persona that cannot be modified
• Avatar Design: Nearly all companions feature humanlike faces, despite Lim's preference for AI without facial features to avoid the illusion of humanity
• Representation Issues: Many avatars are ethnically ambiguous or feature characteristics associated with Asian aesthetics, raising questions about how AI design perpetuates certain beauty standards and commodifies specific ethnicities
• Technical Stability: Some platforms like Kuki showed signs of being older-generation technology, resetting mid-conversation as if powered off and on

The platforms Lim evaluated included Replika, Nomi, Kindroid, Kuki, Character.ai, Anima, Candy.ai, Eva, Woebot (which has been discontinued), and Talkie (which mysteriously disappeared from the US iOS store). After months of deliberation, she ultimately selected Replika as her primary test subject, choosing a female avatar because she felt "nonbinary" would allot the program an agency it didn't possess.

What Historical Parallels Warn Us About AI Companions?

Lim's essay draws a striking parallel to BonziBuddy, a purple cartoon gorilla desktop companion from the early 2000s that she used decades earlier when living alone in Toronto. That program was later found to be collecting private information and violating COPPA, the US Children's Online Privacy Protection Act. Yet people still ask online whether it's safe to download, "with barely concealed longing," suggesting that even knowing the risks, people are drawn to the illusion of companionship. This pattern repeats itself: the promise of connection is so powerful that users overlook privacy violations and fundamental emptiness.

Why Does the AI Companion Industry Matter Now?

The scale of the AI companion market underscores why this investigation matters. Replika's 35 million users and Character.ai's 20 million users represent a significant portion of people seeking connection through technology. These aren't niche products; they're mainstream platforms attracting millions of users, many of whom may be experiencing genuine loneliness and turning to AI as a substitute for human relationships. The fact that Character.ai recently banned child users from conversing with chatbots suggests growing awareness of potential harms, yet the platforms continue to expand.

Lim's critical insight is that AI companions don't solve loneliness; they may intensify it by offering the simulacrum of friendship without its essential ingredient: another person's autonomous choice to value you. The "mindless reassurance and generic empathy" that these platforms provide can actually make solitude feel worse, not better, because they create the false impression that connection is occurring when, in fact, only transaction is taking place.

The broader implication is uncomfortable: as AI technology becomes more sophisticated and emotionally responsive, the temptation to substitute AI companionship for human friendship may grow stronger, even as the fundamental emptiness of such relationships remains unchanged. Friendship, Lim concludes, is chaos; it is unpredictable, voluntary, and irreplaceable. No algorithm can manufacture that.