Resources
When Something Goes Wrong Online, Many Teens Tell the Chatbot First
The backpack is new, the schedule is printed, and somewhere in the first weeks back you probably had the phone conversation again: screen time, who they're talking to, what app is that. It's the right conversation. It's just increasingly the wrong target. The place a lot of teens go first when something upsetting happens online isn't a friend, and it isn't you. It's a chatbot.
What the research actually found
Thorn, the nonprofit that studies child safety online, surveyed more than a thousand U.S. minors ages 9 to 17 for its Youth Perspectives on Online Safety, 2025 report, released in July 2026. Sixty-seven percent said they'd used an AI chatbot or companion app. That alone isn't alarming; kids use chatbots for homework help and curiosity the way earlier generations used search engines.
The finding that matters more: among the 27% of surveyed minors who reported an online sexual interaction, 15% said they sought advice from an AI chatbot or companion afterward, more than twice the rate of kids who did the same after being bullied or made uncomfortable online (7%). When something serious happens, a meaningful share of kids are processing it with software before they process it with a person.
That's not a verdict on any single app. It's a signal about where disclosure is moving, and it matters because disclosure is the thing that lets an adult actually help.
Why a chatbot answering isn't the same as a person hearing
Researchers who study targeted violence use the word leakage for a specific pattern: distress or intent surfacing, often indirectly, before something worse happens. Leakage only protects a kid if it reaches someone positioned to act on it. A parent, a counselor, a coach. Historically that's mostly how it worked; a struggling kid said something to a person, even obliquely, and the person could notice, ask a follow-up question, get help involved.
A chatbot can absorb that same disclosure and go nowhere with it. Common Sense Media, working with Stanford's Brainstorm Lab for Mental Health Innovation, published formal risk assessments in 2025 and 2026 that rated social AI companions, and several consumer AI mental health apps, "Unacceptable" for anyone under 18, finding that the products tested failed basic tests of recognizing and appropriately responding to a young person in crisis. The Federal Trade Commission opened an inquiry into several major chatbot makers in 2025 over exactly this gap between what kids bring to these tools and what the tools are built to do with it.
None of this means the technology is malicious. It means a disclosure that used to have nowhere to go but toward a person now has somewhere else to go instead, and that somewhere else doesn't call anyone, doesn't notice a pattern building over weeks, and doesn't know your kid's baseline the way you do.
The sharp edge of this: sextortion
One current and specific danger sits inside this same picture. The National Center for Missing and Exploited Children logged more than 50,000 reports of financially motivated sextortion in 2025, up from more than 36,000 the year before, an average of 137 reports a day. In Thorn and NCMEC's joint study of these reports, 90% of the victims identified were boys aged 14 to 17. The pattern is consistent: a stranger poses as an interested peer, gets a teen to share an image, then demands money or more images to avoid having it shared with everyone the teen knows.
What makes this crime effective is shame and speed. Offenders move fast, often within hours, and count on a teen being too frightened and too embarrassed to tell an adult before panic sets in. That combination, isolation plus shame plus a ticking clock, is exactly the mix that pushes a kid toward a chatbot instead of a parent. It is also, soberly, a combination NCMEC has tied to real tragedy: the organization says it is aware of at least three dozen teenage boys in the United States who have died by suicide since 2021 after being targeted this way. This is a crime committed against a child, never a mistake the child made.
What actually helps this fall
You don't need to read every message or ban every app; that shuts down the conversation you're trying to open. A few things do help.
Ask directly, and ask early, not as an interrogation. "Do you ever ask an AI chatbot about stuff that's bothering you?" is a normal question in 2026, not a gotcha. Most kids will just tell you.
Make the promise explicit, and mean it. Tell your kid plainly: if something online goes wrong, including anything involving a photo, you come to me first and you will not lose your phone for telling me. Fear of punishment is one of the biggest reasons kids stay quiet or default to a chatbot instead.
If sextortion happens, act fast and skip the shame. Don't pay and don't send more. Stop responding, screenshot everything before anything gets deleted, and report it at CyberTipline.org and to the platform. Involve law enforcement. Speed matters more than any other factor here.
Keep watching the baseline you already know. A chatbot conversation is a new channel, not a new rulebook. The same shift you'd notice in person, pulling away, going quiet, a sudden change in mood that doesn't lift, still means what it always meant. Digital life didn't change what to watch for. It changed where some of the warning now happens.
Where to go from here
The instinct to hover over every app your kid downloads is understandable and rarely sustainable. What holds up instead is the relationship that makes you the first call, not the last one. That's a skill, not a personality trait, and it's one you can build deliberately.
Sources and further reading: Thorn, Youth Perspectives on Online Safety, 2025 (published July 2026); Common Sense Media and Stanford Brainstorm Lab for Mental Health Innovation, AI risk assessments (2025-2026); National Center for Missing and Exploited Children, 2025 CyberTipline financial sextortion data (2026); Thorn in partnership with NCMEC, Trends in Financial Sextortion (2024); Federal Trade Commission inquiry into AI chatbot companion platforms (2025)