Researchers Watched What 3,363 Kids Actually Typed Into AI Chatbots — 15% Was Violence and 15% Was Sexual Role-Play
Back to blog
Family Safety·OCTOBER 17, 2026

Researchers Watched What 3,363 Kids Actually Typed Into AI Chatbots — 15% Was Violence and 15% Was Sexual Role-Play

The Blaick Team 5 min read

Most research on kids and AI asks children or parents what they think happens. This study, published in July 2026, did something different: it looked at what 3,363 US young people aged 1 to 17 actually typed, using keystroke data from a parental monitoring keyboard across more than 42,000 app-days of chatbot use.

The number to hold: 62%. That is the share of chatbot use that was ordinary problem-solving and looking things up. It is reassuring, and it is also the number that makes the rest easy to miss.


The everyday use
62% tool use
Homework help, questions and information seeking. Among 16 and 17 year olds it rose to 80%.
Violence
15% of interactions
Roughly 6,460 of 42,355 logged interactions involved violent content.
Sexual role-play
15% of interactions
Another 8% was romantic role-play and 9% general role-play.
The younger the child
More of the risky stuff
Younger users leaned relatively more toward violent, sexual and romantic role-play than older teens.

Source: "How US Youth Use AI Chatbots: Conversation Patterns From Naturalistic Keystroke Observations" (PubMed Central, July 27, 2026). Data covers January to September 2025 from the Aura parental monitoring app; categories are the researchers' classifications of logged interactions, not a count of children.


Why This Study Is Different

Surveys capture what families remember and are willing to say. A keystroke record captures what was typed. That is why the finding matters: the gap between "we mostly use it for homework" and what a child does in the other third of their time is invisible to a conversation at the dinner table, and it is visible to a tool that actually looks.

What stands out

Role-play is a big category. Combined, violent, sexual, romantic and general role-play made up nearly half of all logged interactions.

Emotional support is small but real. Only 3% of interactions were about feelings, and 6% were friend-like chats, yet those are the moments that matter most.

Young kids are in there too. The sample ran from age 1 to 17, so these are not just teens.

A child who uses AI for homework 62% of the time is also a child who uses it for something else the rest of the time.

The Caveat Worth Stating

Read it fairly

These families chose monitoring. The children in the sample had a parental monitoring app installed, so they may not match every household.

Percentages are of interactions, not children. It does not mean 15% of kids are doing this. It means that across the logged activity, that share fell into those categories.

Why it still matters

Even with those caveats, the pattern is clear: AI chatbots are not only a study aid for children. They are also a place to act out scenes no adult is watching.

You do not need to assume the worst. You do need to stop assuming it is all homework.

What to Do This Week

Ask about the stories Ask whether they ever use AI for characters, games or storytelling, and what happens in them.
Mind the younger ones The study found more risky role-play among younger users. Keep AI use for under 13s in shared spaces.
Agree a no-secrets rule Make it safe to say "something weird happened in a chat" without losing the device.
Look at the real picture Use visibility into which AI apps your child opens and how, instead of relying on "it's for school".

The data says most of it is homework. Your job is to find out about the rest, calmly, before someone else does.


See the whole picture, not just the homework.

Start a free 14-day Blaick trial — no credit card required.

Blaick gives you clear visibility into the AI tools your child uses, so you can have the right conversation at the right time.

Start free trial →
Cancel anytime · Trusted by security-conscious families · Built on enterprise AI security