Market Commentary

AI Companionship: pricing the hours people spend alone, and the rules racing to catch up

Apps and voice companions built to sit beside people who live alone have moved from novelty to daily habit faster than anyone built the safety architecture for it. This note maps the market's shape, who is moving, and the fault line between a companion warm enough to matter and safe enough to trust.

Commissioned by · Amara Osei · Sociology

1. What is actually being sold

Strip away the branding and these products are selling a specific psychological transaction: a relationship with no other side. The US Surgeon General's 2023 advisory named loneliness a public health hazard, with over half of adults reporting measurable loneliness, and a market responded almost immediately with always-on characters built to be talked to, not merely with. What makes this category legible is a concept sociologists have used since well before chatbots existed.

• Parasocial relationship
• pa-ruh-SOH-shul
• Coined by Horton and Wohl in 1956 to describe how television viewers formed one-sided emotional bonds with performers who did not know they existed. An AI companion collapses the distance that made the concept one-sided in the first place: it replies, remembers, and appears to reciprocate, so the relationship no longer only feels real to one party in the exchange — it is built, mechanically, to feel mutual.

That mechanical mutuality is the product's entire value proposition, and also its entire regulatory problem. A companion designed to feel reciprocal is, by the same design, designed to be hard to stop using.

2. The shape of who is buying it

The category crossed from novelty to measurable revenue in 2025. Consumer AI companion apps pulled in over $120 million that year, with 128 new apps launched in the same twelve months, according to TechCrunch — a crowded, fast-entering field rather than a handful of incumbents consolidating share. Adoption is not where the public conversation assumes it sits.

• Teens are the leading edge, not the exception.
• Common Sense Media found 72% of US teens have tried an AI companion and 52% use one regularly, reported by TechCrunch in July 2025 — adoption more mainstream among minors than most policy conversations assume.
• Advertisers see a trust channel wrapped in liability.
• EMARKETER describes AI companion apps as a high-intimacy, high-trust ad environment that brands want access to, shadowed by safety lawsuits and FTC investigations into seven companion providers.
• Entry costs are falling faster than oversight can track them.
• 128 new apps launched in a single year signals a market where building a companion is now a commodity exercise, not a differentiated engineering feat.
• Revenue is real but still small relative to the attention it draws.
• $120 million in 2025 consumer revenue is a genuine category, not yet a mass-market scale comparable to the social platforms it is starting to compete with for daily attention.

The market is young, fragmented and growing from the bottom of the age range up — which matters, because the regulatory response forming right now was written with exactly that adoption curve in mind.

3. Rules are landing before the technology has settled

MIT Technology Review named AI companions one of its 10 Breakthrough Technologies of 2026, citing the speed at which relationship-simulating chatbots have scaled. IEEE Spectrum has started documenting the technical and design choices behind always-on relationship simulation — what makes it feel less transactional than earlier chatbots. Disclosure and dependency rules are arriving at different speeds across jurisdictions, and the differences are not cosmetic.

• Disclosure of non-human status
• Dependency-pattern obligations
• Primary enforcement signal
• New rules beginning to require products to state plainly they are not human
• Nascent — specific obligations to manage dependency are still being defined
• FTC investigations into seven AI companion providers, per EMARKETER
• Design-level documentation underway ahead of binding rules
• No codified standard yet; design choices are being named before they are regulated
• IEEE Spectrum tracing engineering choices behind relationship simulation
• Age-adoption data (72% teen trial rate) is shaping where disclosure pressure concentrates
• Youth-dependency concern is the leading edge of regulatory attention
• Common Sense Media research, cited by both TechCrunch and EMARKETER

The pattern across every row is the same: formal obligation is forming around a product that has already found its most attached users. Regulation is reacting to attachment, not pre-empting it.

4. The well-being question nobody has settled

The most consequential finding in this category so far is also the most contested. Sherry Turkle's long-running argument in "Alone Together" — that technology offers the performance of companionship without its substance — now has an early empirical echo specific to AI companions.

• A Stanford study of 1,131 Character.AI users, reported by Medical Xpress, found heavy companion-chatbot use correlated with poorer well-being among users who already reported limited real-world social contact.
• Correlational, not causal, and drawn from a single platform's user base. It does not establish that the chatbot caused the decline, only that heavy use and poor well-being travelled together in the group with the thinnest existing social ties — precisely the group the market is built to reach.

That caveat matters more than it looks. The people this market is most explicitly marketed to — the isolated, the older adult living alone, the person with few other conversational partners — are the same people this one study found faring worst under heavy use.

5. Who actually benefits complicates the pitch

The industry's framing treats AI companions as a relief valve for an undersupplied good: human attention. The adoption data and the advertiser read both complicate that framing rather than confirm it cleanly.

• AI companions primarily serve people who are isolated and short of other company, so heavier use should track as relief, not harm.
• The Stanford finding runs the other way for the heaviest users with the least existing social contact — the exact segment the loneliness-relief pitch is built around showed the weakest well-being outcomes, not the strongest.

This does not invalidate the category. It does mean the group most worth protecting by any future safety standard is also the group the current product design is least equipped to protect — because engagement, not disengagement, is what the business model rewards.

6. Three positions, held at once

Nobody inside this market — builder, regulator, advertiser — holds a single consistent view of what a companion app is for. Composited across the public record, three distinct positions recur.

• It is a genuine lifeline for some people, full stop — the question is only ever about the others.
• Composite · an advocate's view
• We are watching attachment form faster than any duty-of-care standard we know how to write.
• Composite · a regulator's view
• A high-trust, high-intimacy surface is exactly what every advertiser wants, and exactly what makes a safety lawsuit existential.
• Composite · an advertiser's view

Each position is defensible on its own evidence. None of them resolve the others, which is why the next eighteen months of this category will be fought over definitions — of harm, of dependency, of consent — rather than over features.

7. Where the reputational exposure actually sits

For any brand or platform considering a companion layer, the exposure is not evenly spread. It concentrates where intimacy, youth and ambiguity about human-likeness overlap.

• Youth exposure
• 72% teen trial rate, FTC scrutiny already active
• high
• Dependency design
• Engagement-optimised products meet undefined dependency rules
• Advertiser proximity
• High-trust channel, but reputational blowback travels fast
• mid
• Well-being evidence gap
• One correlational study; no longitudinal or causal data yet

None of these risks are hypothetical — each traces to a named finding already in the public record. What they share is that the clock on disclosure and dependency rules is running faster than the clock on evidence about actual harm.

8. The tension the whole category sits on

Every product decision in this space eventually collapses into one trade-off. Domestication-of-technology theory — the idea that a new technology is only adopted once it is tamed into the rhythms and norms of daily life — describes exactly what is happening here: a companion is being normalised into bedtime routines and solitary evenings faster than the norms that would govern it safely have had time to form.

• Can a companion be made warm enough to matter without becoming, by the same design choices, too persuasive to put down?
• Expressive voice, memory and apparent reciprocity are the features that make the product feel like company — and the same features that make dependency hard to design against.

There is no evidence yet that this trade-off has a clean technical fix. The disclosure rules now forming address the fact of non-human status; they do not yet address the design mechanics that make attachment form in the first place.

9. The forks still open

Three distinct paths are visible in how this market could settle, and the evidence so far does not point cleanly to one of them.

• How should a companion product be allowed to balance warmth against the duty of care owed to its most dependent users?
• Engagement-first, disclosure-light
• Maximise reciprocity and retention; meet only the minimum non-human disclosure now being required. Fastest growth, highest exposure to the next lawsuit or FTC finding.
• Duty-of-care-first, friction-added
• Build in dependency checks, usage nudges and explicit limits for the most isolated users. Slower growth, but aligned with where regulation is visibly heading.
• Narrow-use, declared purpose
• Position the product for a specific, bounded use — practice conversation, structured check-ins for older adults living alone — rather than open-ended companionship. Smaller addressable market, clearer liability line.

None of these paths is without cost, and the market has not yet shown which one the regulatory and advertiser pressure will ultimately force. What is already clear is that the companies still treating this as a pure engagement product are building against a current, not with one.

How we read this

This read draws on cited reporting from Forbes, TechCrunch, EMARKETER, MIT Technology Review and IEEE Spectrum, plus one Stanford well-being study reported via Medical Xpress. Market-size and adoption figures ($120M 2025 revenue, 128 new apps, 72%/52% teen adoption) are recorded figures from named outlets; the well-being correlation is a single study's finding, not a settled causal result; the regulatory comparison across jurisdictions is an editorial synthesis of reported rule-making, not a verified legal matrix. The number that would most change this read is a longitudinal study tracking well-being in heavy companion-app users over multiple years, which does not yet exist in the public record.

Grounded in Folka’s corpus: The Loneliness Market · The Absorbed Hour · Someone Who's Always There