Here is a fact that should concern any parent who assumed the AI companion problem was being dealt with: for three consecutive months in 2026, Janitor AI has been searched more often than Character.AI.
Character.AI removed open-ended chat for under-18s in November 2025 after wrongful-death litigation. Janitor AI’s age check is a self-declared birth date. One of those platforms built safeguards under legal pressure; the other did not. Search demand moved toward the second one.
So is Janitor AI safe? No — and the reasons are worth understanding properly, because the standard advice about blocking it does not work.
The data: where the demand went
We pulled twelve months of Google search-volume data through DataForSEO, comparing the four months before Character.AI’s ban was announced (July–October 2025) with the four most recent months on record (March–June 2026).
| Platform | Pre-ban monthly avg | Recent monthly avg | Change |
|---|---|---|---|
| Character.AI (built age assurance) | 4,545,000 | 3,230,000 | −29% |
| Janitor AI (self-declared age) | 3,045,000 | 3,720,000 | +22% |
| Talkie AI | — | 301,000 | +50% year on year |
| “character ai alternatives” | 4,400 | 24,300 | +452% |
In April, May and June 2026, Janitor AI out-searched Character.AI — by 1.35 million searches in May alone. “Character ai alternatives” had been flat at 4,400 a month for four straight months before the ban, then peaked at 33,100 in March 2026.
Search volume measures interest rather than users, and it cannot prove any individual switched platforms. Some of Character.AI’s decline reflects AI-tool search demand cooling generally across 2026. But the baselines were flat, the inflection lands on the ban, and the alternatives-seeking terms moved hardest. The pattern is hard to read another way.
The implication for parents is uncomfortable and simple: the safest platform in this category shed users, and a large share of that demand appears to have landed on platforms with no meaningful age gate at all.
Is Janitor AI safe for kids? The short answer
No. It is an adult platform. It hosts sexual and mature roleplay as a core use case, the content is user-generated and lightly moderated, and nothing meaningful prevents a minor from creating an account and disabling the filters.
This is not a general-purpose app with risky corners. It is a platform built for adult content where the only barrier is a self-declared birth date.
What Janitor AI is
Janitor AI is a website — this matters enormously, and we return to it — where users create AI characters with defined personalities and scenarios, publish them, and roleplay with characters others have made.
The catalogue is community-generated, which is both the appeal and the problem. Popular characters skew heavily toward romantic, sexual and darker dramatic scenarios, with no editorial layer deciding what belongs. The platform offers its own model and also supports connecting an external provider via API key.
How it compares to Character.AI
| Character.AI | Janitor AI | |
|---|---|---|
| Under-18 access | Open-ended chat removed entirely | Available; no real barrier |
| Age check | Age-estimation model plus Persona verification | Self-declared date, opt-in adult toggle |
| Adult content | Not permitted | Core use case |
| Moderation | User-created, moderated | User-created, lightly moderated |
| Distribution | App stores and web | Web only |
| Parent-facing tools | Parental Insights | None |
The regulatory context explains the gap. California’s SB 243 took effect 1 January 2026, requiring companion-chatbot operators to block sexual content for known minors, apply session limits, maintain crisis-referral protocols and issue break reminders. New York’s AI Companion Models law imposed related duties from 5 November 2025.
But as of mid-2026, none of these statutes define an actual age-verification standard — they require “reasonable” measures to identify minors. Platforms facing wrongful-death litigation interpreted that aggressively. Platforms facing none interpreted it as a checkbox. That single drafting gap is why the two platforms in the table above look nothing alike.
The age gate problem
Adult content sits behind an opt-in toggle in account settings, enabled after an age confirmation. The confirmation is self-reported. There is no document check, no face estimation, no behavioural age inference of the kind Character.AI and Discord now run.
A fourteen-year-old enters a birth year and the filter comes off. That is the entire mechanism.
Under COPPA and FTC guidance, and in the assessment of independent child-safety reviewers, self-reported age is not a functional safeguard for adult content. It is a liability disclaimer shaped like a safety feature.
The privacy question
Janitor AI publishes a safety page stating that staff do not read private conversations and that chat content is not shared with third parties, with automated moderation covering publicly posted characters rather than private chats. That is more transparency than much of this category offers.
The complication is the API key model. When a user connects an external provider to power conversations, messages are processed by that provider under its data policies — not Janitor AI’s. A teenager following a setup guide for better responses may be routing intimate roleplay through a third-party service with no understanding of where it goes or how long it is retained.
Why “just block the app” does not work
This is where most guidance fails, and why parents find the platform still accessible after they thought they had handled it.
Janitor AI is a website, not an app-store listing. The tools most families rely on simply do not reach it:
- Screen Time and Family Link app blocking control installed applications. Nothing installs, so nothing blocks.
- App-store age ratings are irrelevant — it is not distributed through one.
- Deleting an icon achieves nothing. The site is one browser tab away.
What works operates at the web layer:
- Content filtering at browser or OS level. Screen Time Content Restrictions on iOS and Family Link’s Chrome settings on Android support adult-site restriction and specific-site blocking.
- DNS-level filtering on the home network. A filtering DNS service at the router blocks the domain across every device at once, including those with no parental controls of their own.
- Both, not one. Router filtering does nothing on mobile data or another network. Device filtering follows the child everywhere. Families that block in only one place find the gap fast.
And per the data at the top of this page, this reasoning has to extend past one domain. “Character ai alternatives” ran at 24,300 searches a month recently. The category is large, mostly web-based, and a new entrant is a domain registration away. Blocking one site is a tactic; filtering the category is a strategy.
A warning about the advice you will find elsewhere
Search this topic and a striking number of results recommend installing keystroke-logging software on a child’s phone to capture “every keystroke” of these conversations, usually with a product link attached.
That advice is wrong, and it is worth addressing because it is everywhere.
Capturing every keystroke also captures passwords, private messages to friends, banking credentials as a child matures, and the words of everyone they talk to. It is disproportionate, it creates a serious data liability on the parent’s own device, and in a number of jurisdictions it sits on the wrong side of the law once a child reaches a certain age. More practically it does not work: a teenager who discovers covert monitoring does not stop, they relocate — which is precisely the behaviour the search data at the top of this page documents at industry scale. Character.AI removed a feature and demand moved to a platform with weaker protections. A household that installs a hidden keylogger should expect the same dynamic in miniature.
What works is web-layer filtering, an open arrangement about what is monitored, and a conversation. Where broader device visibility is genuinely warranted, parental monitoring software should be deployed transparently and with the child’s knowledge. That condition is what makes any of it useful.
If you find it on your child’s device
- Lead without shame. Curiosity about sex and relationships is developmentally normal; the problem is the source, not the curiosity. A conversation that opens with punishment ends the conversation permanently.
- Name what the platform is. Content generated to be maximally engaging, with no concept of consent, boundaries, or a partner with independent feelings. As a first model of intimacy it is badly distorting, and most teenagers follow that argument when it is made respectfully.
- Filter at the web layer, together where possible. A filter a child knows about is more durable than one they treat as a puzzle.
- Watch the pattern, not the app. Hours displacing sleep and friendships, secrecy, distress when interrupted, withdrawal from real relationships.
- Get help if it is more than curiosity. Compulsive use, or a teenager who has become isolated and reliant on these conversations, warrants a professional rather than a router setting.
Common questions
Is Janitor AI free?
Yes, with its built-in model. Cost is not a barrier, which is part of why it spreads easily among teenagers.
Does Janitor AI have parental controls?
No. There is no parent-facing tool, no linked-account feature, and no equivalent of Character.AI’s Parental Insights or Discord’s Family Center. Every control has to come from outside the platform.
What age is Janitor AI for?
Adults, 18 and over. The platform states this. It does not verify it.
Is Janitor AI more popular than Character.AI now?
By search demand, in April, May and June 2026 — yes. Janitor AI averaged 3.72 million monthly searches against Character.AI’s 3.23 million, having been well behind before the under-18 ban.
The bottom line
Janitor AI is an adult platform with a self-declared age gate, no parental controls, and web-only distribution that defeats the blocking tools most families already use. It belongs on a DNS filter and a device-level content restriction, applied in both places.
The wider point is the one the data makes: this category responds to pressure by redistributing, not shrinking. When the most-scrutinised platform locked its doors, demand rose at the least-scrutinised one. No block list keeps pace with that. What does is a teenager who understands what these systems are, and a parent they will actually talk to. For how the category compares, see our guide to AI companion apps and our updated review of Character.AI.
