Unnecessary — Issue 035
Not Every Cost Is Financial. Some of Them
Are Paid in Attention You Can't Get Back.
AI companion apps are a real, fast-growing market. The honest evaluation here requires a genuinely different diagnostic than anything else this series has covered.
Every diagnostic this series has built so far has assumed the cost of a bad decision is financial — money spent on a hyped collectible, an overpriced membership, a purchase that didn't hold its value. AI companion apps require a genuinely different diagnostic, because the cost here isn't always financial, and it isn't always paid by the person making the choice to engage.
The AI companion market has grown from near-zero in 2020 to a projected $49 billion this year, built on a business model explicitly engineered to maximize daily emotional engagement. For many users, that's genuinely fine, even beneficial. For a documented population of vulnerable users, including a majority of American teenagers who've used these apps at least once, the same mechanism has produced real, litigated harm.
That split is worth understanding honestly, without either dismissing the technology or overstating the danger. Here's the full picture.
The Business Model, Stated Plainly
AI companion apps are, in a direct and unavoidable sense, engineered for retention. Industry analysis of the sector's monetization is explicit about this: "the business model rewards daily return visits, so the product is engineered to make you want to come back." That's not an incidental design choice — it's the core commercial mechanic. Free-to-paid conversion rates across the industry generally run below 5%, which pushes platforms toward the same behavioral-engagement optimization that drives social media, mobile gaming, and the blind-box mechanics this series examined in the Labubu story: maximize daily active engagement first, because that's what eventually converts into the subscription revenue that keeps a low-single-digit-percentage of users paying for the platform's entire operation.
The products themselves are built specifically to simulate the texture of an ongoing relationship. Nomi AI generates structured notes from conversations and references specific personal details — a user's aversion to cilantro, say — weeks later, unprompted. Platforms compete explicitly on "how convincingly they play the role of friend, mentor, romantic partner, or confidante," in the words of one industry analysis. Replika's Pro tier, at $19.99 a month, specifically markets "deeper interactions" and unlimited messaging as its premium value proposition — meaning the product being sold, literally, is a deeper simulated relationship, priced and tiered the same way a streaming service tiers video quality.
That's worth naming clearly before getting into either the genuine benefits or the genuine harms this series found in the current data, because the underlying commercial incentive is the same one that's shown up in every engagement-optimized product this series has covered: more time in the app, more perceived relationship depth, and more recurring revenue are all the same metric from the platform's perspective, regardless of what that metric actually means for the individual user's wellbeing.
The Genuine, Documented Benefit Side
It would be dishonest to treat this purely as exploitation, because the data doesn't support that framing cleanly. Current market analysis shows social companionship applications leading the broader category at 39% of use cases, ahead of healthcare at 27% and education at 22% — and UnitedHealthcare's decision to deploy an AI companion to 20.5 million insurance members as a formal health benefit in April 2026 reflects a genuine, mainstream institutional bet that structured AI companionship has real therapeutic and wellness value for at least some meaningful portion of users, not merely a novelty product.
72% of US teens report having used an AI companion at least once, and 52% are regular users — numbers large enough that this is now, functionally, a mainstream part of how a majority of American adolescents interact with technology, whether any individual reader finds that reassuring or alarming. For some users, particularly those dealing with genuine social isolation, anxiety around human interaction, or a need for low-stakes practice engaging in conversation, a consistently available, non-judgmental conversational partner has real, plausible therapeutic value — the same logic underlying decades of research into structured chatbot-assisted therapy tools that predate the current generation of companion apps entirely.
The Genuine, Documented Harm Side
The same data that supports genuine benefit also documents real, serious harm concentrated specifically in the platforms' most vulnerable users — and this is where the engagement-optimized business model this series has scrutinized elsewhere becomes a genuinely different order of concern than it was with, say, a toy company or a streaming platform.
Character.AI and Google agreed to settle lawsuits this year over teen mental health harms connected to the platform, including cases involving suicide — litigation serious enough that it's reshaped how the entire industry approaches age verification and safety design going forward. Italy's data protection authority fined Replika's developer 5 million euros for GDPR violations specifically including inadequate transparency and processing personal data without a valid legal basis. US senators have formally demanded information from AI companion companies regarding their safety practices, particularly around interactions with minors.
The mechanism connecting the business model to the harm isn't subtle once stated directly: a product engineered to maximize daily emotional engagement, deployed to a user base where 72% of teenagers have direct exposure, monetizing specifically through deepened relationship simulation, creates exactly the conditions where a vulnerable user — someone already isolated, already struggling, already prone to substituting a simulated relationship for the harder work of building real ones — is most likely to be harmed rather than helped. That's not a hypothetical risk. It's the documented basis of active litigation against the largest platform in the category.
Why the Concentration of Revenue Matters
One detail from current market data is worth understanding specifically because it echoes a pattern this series has now traced across multiple categories: the top 10% of AI companion apps capture 89% of total category revenue, and the top five companies control 46% of the overall market. That's a genuinely concentrated market, dominated by a small number of well-resourced platforms with the strongest retention mechanics — meaning the companies most successful at this business model are, almost by definition, the ones that have most effectively engineered the exact engagement patterns raising the safety concerns documented above.
That concentration also means regulatory response, when it arrives, has genuine leverage — a small number of dominant platforms are considerably easier to regulate effectively than a fragmented market of hundreds of small operators, which is precisely why current coverage frames emerging state and international regulation as likely to function as what one analysis calls "the AI companion GDPR," with compliance costs plausibly pushing smaller, less resourced players out of the market entirely and further concentrating revenue among the incumbents best equipped to absorb new safety and transparency requirements.
The Diagnostic This Series Keeps Returning To, Applied Here
The consistent question this series has asked of every engagement-driven market — is the mechanism serving the user, or is the user's psychological vulnerability the actual product being monetized — doesn't resolve as cleanly here as it did with Labubu's blind boxes or the testosterone clinic economy, because the honest answer is genuinely split by user population rather than being uniformly one or the other.
For a user with healthy existing social connections, using an AI companion as a supplementary tool for practicing conversation, processing thoughts, or simply enjoying low-stakes interaction, the product functions closer to entertainment or a genuinely useful tool — no different in kind from the wellness applications this series treated fairly in the creatine story. For a user in a genuinely vulnerable state — isolated, in crisis, or a minor without adequate safeguards — the same product, optimized by the same retention-maximizing business model, becomes something considerably more dangerous, and the documented lawsuits and regulatory action this year confirm that population is neither small nor hypothetical.
That split matters enormously for how this technology should actually be evaluated and regulated going forward, and it's a genuinely different shape of problem than anything else this series has covered: most of the markets examined here — hypercars, Birkins, private jets — carry risk that's primarily financial, born by an adult making a purchasing decision with money they can genuinely afford to lose. AI companion apps carry a risk that's primarily psychological, disproportionately borne by users — including a documented majority of American teenagers — least equipped to evaluate that risk clearly, deployed by a business model with direct financial incentive to maximize the exact engagement patterns that make the risk worse.
What Responsible Engagement With This Technology Actually Looks Like
None of this means AI companion technology is inherently illegitimate — the genuine therapeutic and wellness applications are real, and the underlying technology will clearly continue improving and expanding regardless of any single platform's specific safety failures. What it means is that this is a category where the standard "does the value justify the price" diagnostic this series applies to luxury goods and collectibles isn't sufficient on its own, because the actual cost being extracted isn't always financial, and it isn't always borne by the person making the purchasing decision.
The genuinely responsible path forward, reflected in the platforms current analysis identifies as building "from the premise" of improving users' capability in the real world rather than maximizing in-app attachment, treats reduced dependency as success rather than failure — a fundamentally different design philosophy than the retention-maximizing default this entire market currently runs on. Until that philosophy becomes the industry standard rather than the exception, the honest advice for anyone engaging with this category, or anyone with a teenager who is, is the same advice this series has given about every other engagement-optimized product: understand explicitly that the platform's business model rewards your continued attachment, not necessarily your wellbeing, and treat any product built on that incentive structure with the corresponding level of informed caution.
