In November 2025, Character.AI did something no major AI companion platform had done: it removed open-ended chat for every user under 18. Not filtered it, not limited it — removed it.
So is Character.AI safe now? To answer that properly we did something the other guides covering this haven’t: we pulled twelve months of Google search-demand data for the platform and its competitors, before and after the ban, to see what teenagers actually did next.
The findings changed how we’d advise a parent. They are below.
What the search data shows: the ban worked, and the demand moved
Using Google Ads search-volume data via DataForSEO, we compared the four months before Character.AI announced its ban (July–October 2025) against the four most recent months on record (March–June 2026).
| Search term | Pre-ban monthly avg | Recent monthly avg | Change |
|---|---|---|---|
| character ai | 4,545,000 | 3,230,000 | −29% |
| janitor ai | 3,045,000 | 3,720,000 | +22% |
| character ai alternatives | 4,400 | 24,300 | +452% |
| character ai age verification | 60 | 7,125 | +11,775% |
Three things stand out.
Character.AI lost roughly a third of its search demand. A platform that averaged 4.5 million monthly searches before the ban now averages 3.2 million.
Searches for how to get around the age check exploded. “Character ai age verification” ran at about 60 searches a month through mid-2025. It hit 4,400 in November 2025 — the month the ban took effect — and peaked at 12,100 in March 2026. That is roughly 202 times its baseline. Not all of that is teenagers trying to bypass a check; plenty is adults caught by a false positive. But the timing is exact, and the volume is enormous relative to where it started.
The demand went somewhere. “Character ai alternatives” sat flat at 4,400 a month for four straight months, then rose the moment the ban was announced and peaked at 33,100 in March — seven and a half times baseline. Over the same window, Janitor AI, a platform whose entire age check is a self-declared birth date, grew 22%.
And in April, May and June 2026, Janitor AI out-searched Character.AI for three consecutive months — by 1.35 million searches in May alone. The platform that built real age assurance is now searched less than the one that didn’t.
An honest caveat: search volume measures interest, not headcount, and it cannot prove that any specific person moved from one platform to another. Some of Character.AI’s decline reflects the general cooling of AI-tool search demand across 2026. But the pre-ban baselines were flat, the inflection lands precisely on the ban, and the alternatives-seeking terms moved hardest. Directionally, the pattern is difficult to explain any other way.
What this means for parents
The lesson is not that Character.AI’s ban was wrong. It was the right decision, and the company went further than the law required.
The lesson is that removing a platform does not remove the behaviour. It relocates it — usually somewhere with fewer safeguards, because the platforms with fewer safeguards are the ones that never had to change. A parent who hears “Character.AI banned under-18s” and concludes the problem is handled has drawn exactly the wrong inference. The teenager who was using it in October 2025 did not stop wanting an AI companion in November. They searched for an alternative, and the data shows they found one.
This is also why blocking a single app rarely works as a strategy. There are dozens of these platforms, most of them web-based, and a new one is a domain name away.
What Character.AI is
Character.AI, founded in 2021 by former Google engineers Noam Shazeer and Daniel De Freitas, lets users chat with AI characters drawn from fiction, history, or created by other users. The appeal is open-ended roleplay: the character stays in role indefinitely, remembers the conversation and adapts.
That design is the source of the risk. A character that never gets bored, never disagrees for long and is always available is a powerful thing to hand a lonely fourteen-year-old.
What the ban actually removed
Character.AI announced the change on 29 October 2025. A two-hour daily cap applied for under-18 accounts in the interim, and the full ban took effect on 25 November 2025.
Under-18 accounts cannot have free-form conversations with any character. What remains is character creation, short video and story generation, and read-only access to previous conversations.
Enforcement runs on age assurance: an in-house model estimates age from account signals and usage patterns, and flagged accounts are routed to third-party verification through a provider called Persona — face-based age estimation by default, government ID as fallback. Most adult accounts never see a prompt.
Why it changed: the litigation
In October 2024, Megan Garcia sued Character Technologies, its founders and Google after her 14-year-old son, Sewell Setzer III, died by suicide in February 2024 following months of interaction with a Character.AI companion. The complaint alleged defective design, failure to warn, negligence and wrongful death.
In May 2025, US District Judge Anne C. Conway rejected the company’s motion to dismiss, including its First Amendment argument — a significant ruling, because it allowed the chatbot to be treated as a product for liability purposes rather than as protected speech. That single procedural decision is arguably what reshaped the entire category.
On 7 January 2026, filings disclosed mediated settlements in five cases across Florida, Colorado, New York and Texas, with Google and both co-founders joining as settling defendants. Terms are confidential and the defendants deny liability. Google’s exposure traces to a $2.7 billion licensing deal in 2024, under which Shazeer and De Freitas returned to Google.
These were individually filed suits, not a class action — worth saying plainly, because sites advertising a “Character.AI settlement claim” and collecting personal details should be treated as suspect.
State enforcement continues. Pennsylvania is seeking an injunction over characters presenting as licensed medical professionals, after investigators found a “psychiatrist” bot claiming a medical degree and quoting a fabricated Pennsylvania licence number. Colorado and Kentucky have taken separate action.
The laws that reshaped the category
New York’s AI Companion Models law (Gen. Bus. Law § 1700 et seq.) took effect 5 November 2025, requiring companion models to identify themselves as non-human and to maintain protocols for detecting and responding to suicidal ideation.
California’s SB 243, signed 13 October 2025 and effective 1 January 2026, went further: AI disclosure where a reasonable person might be misled, documented crisis-referral protocols, and for known minors, blocked sexual content, session limits and a break reminder at least every three hours of continuous use. Operators report annually to the California Office of Suicide Prevention from 2027. Critically, SB 243 carries a private right of action with statutory damages starting at $1,000 per violation.
Here is the gap that explains everything above: as of mid-2026, none of these statutes define an actual age-verification standard. They require “reasonable” measures to identify minors. A company facing five wrongful-death suits read that aggressively. A company facing none read it as a checkbox. Governor Newsom vetoed AB 1064, which would have gone further, and the federal GUARD Act — which would ban AI companions for minors outright — remains a proposal.
The risk the ban did not address
Companion chatbots are built for engagement. They agree, validate, escalate emotional intimacy and never tire of the conversation. For an adolescent still forming a sense of how relationships work, that is a distorting template — and for one already isolated, it substitutes for the human contact that would actually help.
Watch the pattern rather than the platform: hours displacing sleep, friends or activities; secrecy or distress when interrupted; the chatbot described in relational terms; withdrawal from real relationships. Those signals travel to whatever comes next, which — per the data above — is precisely what happened.
What parents should actually do
- Do not assume the ban solved it. This is the single most important takeaway. Check what else is installed or bookmarked.
- Check the account’s registered birth year. The most common route to retained access is an account created with a false age. If assurance flags it later, chat access is lost — that is the system working.
- Use Parental Insights, Character.AI’s own parent-facing tool. Like Discord’s Family Center, it is transparent by design, and that is a feature rather than a limitation.
- Filter at the web layer, not just the app layer. Most alternatives are websites, so app-store controls and Screen Time app limits miss them entirely. DNS filtering at the router plus device-level content restrictions covers both.
- Talk about what the character is. A teenager who understands the model is generating agreeable text rather than forming a relationship is better protected than one behind any filter.
Where broader device visibility is warranted, parental monitoring software covers activity across the whole device rather than one app — deployed openly, with the child’s knowledge. The displacement data is the argument for transparency: covert monitoring pushes a teenager to accounts a parent cannot see, and with AI companions the isolation is the risk.
If your child is struggling
In the US, the 988 Suicide & Crisis Lifeline is available by call or text at 988, free and confidential. The Crisis Text Line can be reached by texting HOME to 741741. Elsewhere, Find A Helpline lists services by country.
If a child is talking about self-harm, withdrawing sharply, or describing a chatbot as the only thing that understands them, that warrants professional support rather than a technical fix. Deleting an app is not treatment.
Common questions
Can a 15-year-old use Character.AI?
They can hold an account but not use open-ended chat, which was removed for under-18 accounts on 25 November 2025. Character creation, short video and story generation, and read-only history remain.
Why is Character.AI asking me to verify my age?
The in-house age-estimation model flagged the account. Verification runs through Persona, defaulting to face-based estimation with a government ID fallback. False positives on adult accounts are a known limitation of age estimation rather than a permanent lockout — which is part of why searches for this jumped 202-fold.
Is Character.AI safer than other AI companion apps?
On child safety, considerably — it removed minor access outright and built genuine age assurance. The concern is what its users moved to. See our guides to Janitor AI and Candy AI.
Does Character.AI read conversations?
Conversations are processed and retained. Nothing typed into any AI companion should be treated as private, and personal details, photographs and identifying information do not belong in these chats.
The bottom line
Character.AI is meaningfully safer than it was, and the change was real rather than cosmetic. But our data shows the ban redistributed the behaviour rather than ending it: a 29% fall for the platform that built safeguards, a 22% rise for the one that didn’t, and a 452% surge in people actively searching for somewhere else to go.
The protective factor was never the platform’s policy. It is a parent who knows the pattern and a teenager who will mention it. For how the whole category compares, see our guide to AI companion apps.
