AI Companionship Stopped Being a Punchline — Here's What Replaced the Joke

The 60 million MAU figure is a composite estimate based on public claims from major players: Character.AI's reported 30M+ users, Replika's roughly 5M active, Janitor AI's reported 7-10M, Candy.AI's stated 2M+, plus the long tail of Kindroid, DreamGF, Spicychat, CrushOn, Talkie, and dozens of regional apps. The number is fuzzy — companies define MAU differently, and substantial overlap exists between platforms — but the order of magnitude is right.

For context: 60M MAU puts AI companion apps in the same bucket as established consumer software categories. It's roughly comparable to the active user base of Discord (~150M MAU), Reddit (~95M daily), or Pinterest (~80M monthly active US). It's larger than every major dating app combined except Tinder.

The distribution is notable. Character.AI's userbase skews younger and SFW-leaning. The NSFW-leaning apps (Janitor AI, Candy.AI, Spicychat) skew older and more male. Replika's userbase is more balanced gender-wise than the rest of the category — one of the structural reasons it's been the most-studied app academically. The 'AI companions are for lonely men' stereotype was always partially wrong; in 2026 it's just wrong.

Mainstream press coverage of AI companionship in 2025-2026 broadly clustered into three approaches. The hand-wringing approach (NYT, Atlantic) tended to lead with worst-case anecdotes and frame the category as a public-health concern. The neutral-curious approach (Wired, MIT Tech Review) tended to interview users without judgment and let the texture of the experience drive the piece. The dismissive approach (some legacy commentariat) treated the whole thing as too lowbrow to engage with seriously.

The approach that aged best so far is the neutral-curious one, mostly because it correctly identified that AI companion users are not a monolith and that their use cases vary as much as the use cases of any major software category. The hand-wringing approach generated a lot of headlines but consistently misread its subject — interviewing people in crisis about AI use and treating that crisis as evidence of AI causation rather than correlation.

The quiet correction underway in 2026 press coverage: AI companion users are increasingly being interviewed as users of a tool, not as case studies in a moral debate. The shift in framing is doing more cultural work than any specific piece.

The most important development in the last 18 months isn't technological — it's that the actual use cases of AI companion apps clarified into recognizable patterns. Five dominant ones emerged from community discussion, academic research, and platform-reported behavior data.

First, journaling-with-pulse: users who use the AI as a structured way to process their day, with the AI's responses giving the journaling more shape than empty pages. Second, social skill practice: users with social anxiety using the AI as a low-stakes environment to practice difficult conversations, dating dynamics, or assertiveness. Third, creative writing partner: users who treat the AI as a co-author for fiction, fanfic, or roleplay. Fourth, parasocial supplement: users with healthy social lives using the AI as an additional source of attention rather than a replacement. Fifth, intimate scaffolding: users in long-distance relationships, post-divorce transitions, or other temporary states using the AI as a bridge.

Notably absent from the dominant list: 'AI as girlfriend replacement.' That use case exists but is much smaller than the cultural panic suggested. The actual category is broader and weirder.

By mid-2026 three regulatory threads are running in parallel and will likely shape the next phase of the category. The EU's AI Act applies to AI companion apps as 'high-risk' systems in some interpretations, requiring transparency about AI status, age verification, and content moderation standards. Implementation has been uneven but is tightening. The UK's Online Safety Act is similar in spirit but more focused on age-gating and adult content classification. US-side, FTC enforcement actions against specific apps for misleading marketing or inadequate age verification have created industry-wide chilling effects on the most aggressive marketing tactics.

The direction of travel is clear: AI companion apps will be regulated more like dating apps than like search engines. Age verification will get serious. Content moderation will get audited. Marketing claims about 'real relationships' will get scrutinized. The companies that thrive will be the ones that built compliance in early; the ones that didn't are spending 2026 retrofitting it under regulatory pressure.

For users, the practical impact is mostly that signup processes are getting more friction-heavy. Whether the regulation produces better outcomes for users or just shifts the worst behavior to less-regulated jurisdictions is the open question.

Predictions in this category have an embarrassing track record, so call this informed speculation. The next 18 months are likely to bring three shifts.

First, voice will become the default modality. Text-led interaction is starting to feel old-fashioned in apps where voice has been rolled out, and the engagement metrics on voice-led conversations are reportedly significantly higher. Apps that don't ship credible voice experiences in 2026 will look dated by 2027.

Second, persistent multimodal characters will emerge — AI companions that maintain identity across text, voice, image, and eventually video, with consistent visual representation that doesn't reset every session. The technical building blocks exist; integration is the challenge.

Third, the consolidation wave will start. The current category has 50+ meaningful players and the unit economics don't support that long-term. Acquisitions and shutdowns will produce a more concentrated landscape by 2027 — probably 5-8 mainstream apps, a handful of power-user platforms, and a long tail of regional or niche-specific products.

What won't happen, despite predictions: AI companions won't 'replace' human relationships en masse. They'll continue what they're already doing — adding a new category of relationship to the existing options humans have for connection. Some people will use it well. Some won't. Most will use it the way most people use any tool: occasionally, for specific purposes, with mixed results.