AI chatbot roleplay is going mainstream, and here's what's driving it

From novelty to habit: how ai chatbot roleplay went mainstream

For most of the 2010s, talking to a chatbot for company was something you did on a niche app, usually with limited, scripted responses. What changed was the underlying model. Once large language models became good enough to hold a consistent voice across a long conversation, and cheap enough to run at consumer scale, the format stopped being a toy and started being something people returned to daily.

Character.AI is the clearest example of that shift reaching a general audience rather than a niche one: a product built around user-created characters, with roleplay as a core use case rather than an add-on. Its scale, and the seriousness with which lawmakers have since treated it, is itself a signal that this stopped being a fringe activity some time ago.

A market splitting in two

What's happened since is a fairly normal pattern for a maturing category: it's split. On one side sit general-audience platforms, Character.AI chief among them, that serve a broad demographic and have been narrowing their content rules under regulatory and legal pressure. On the other sit openly adult-oriented products, Candy AI and SpicyChat among the more visible names, built explicitly around romantic and intimate roleplay for verified adults.

That split isn't cosmetic. It reflects two different answers to the same underlying demand: some users want a broad, filtered, all-ages product, and others want something narrower but more honest about being for grown-ups. Both halves of the market are growing, and the dividing line between them is exactly the sort of thing SB 243 and the GUARD Act are trying to legislate around.

Why ai roleplay chat is growing so fast

Three things are doing most of the work. The first is simply model quality and cost: a believable, in-character conversation partner used to require expensive, purpose-built infrastructure, and now runs on the same underlying models powering general-purpose assistants, at a fraction of the earlier cost.

The second is memory. Early chatbots forgot everything between sessions, which capped how attached anyone could reasonably get. Persistent memory, where a companion actually recalls what you told it last week, is a genuinely different product experience, and it's the single biggest engineering change behind the format feeling less like a toy and more like a relationship.

The third is social. The US Surgeon General's 2023 advisory on loneliness and isolation documented a real and measurable decline in the kind of everyday social connection people used to take for granted. That doesn't mean AI companions are a solution to loneliness, and nobody serious claims that. It does mean the format has an obvious, non-novelty use case that didn't exist in the same way a decade ago, and that's part of why usage has kept climbing rather than plateauing as a fad would.

Regulation as a lagging indicator of scale

If you want a single, checkable proxy for how big this category has become, look at the law rather than the marketing. Nobody writes dedicated state legislation, and advances a federal bill unanimously out of committee, for a product that a handful of hobbyists use. California's SB 243, in force since 1 January 2026, and the GUARD Act, advanced by the Senate Judiciary Committee on 30 April 2026, both exist because AI companion and roleplay products reached a scale where lawmakers felt they had to respond specifically to this category rather than to AI products in general.

Where it's heading

Expect the split to sharpen rather than blur. General platforms will keep narrowing content rules under legal pressure, and adult-oriented products will keep having to prove they take age verification and crisis handling seriously to stay viable at all. The novelty phase is over. What's left is a genuine, competitive product category, and the companies in it are increasingly being judged, correctly, on the same safety questions as any other consumer product handling something this personal.

What mainstream adoption actually changes for a user

A category becoming mainstream isn't just a business story, it changes what you should expect as a user browsing this space for the first time. Three years ago, an unpolished, buggy AI roleplay app with no safety features was most of what was on offer, because the whole category was still experimental. That excuse doesn't hold any more.

A mature market means genuine competition on the things that used to be afterthoughts: how well a character remembers you, how honestly the pricing is laid out, whether the age gate is a real check or a rubber stamp, whether a crisis conversation gets handled properly. Those used to be nice-to-haves in a young category. In a mainstream one, they're the actual battleground, and a product that skips them is choosing to compete on novelty alone in a market that's moved past rewarding that.

The part that hasn't caught up yet

Regulation, as noted above, is a lagging indicator, and it's still lagging. SB 243 covers disclosure and crisis referral; it doesn't mandate real age verification, and the GUARD Act that would has not passed. That gap between a mainstream, fast-growing product category and a still-developing regulatory floor is exactly the space where the difference between a responsible operator and a reckless one matters most right now, because the law hasn't yet forced every company in the category to close it. Until it does, that judgement call is left to individual products, and to the people deciding which one to trust with an evening's worth of their attention.