The word secretly in that headline needs an audit before we go anywhere.
When I went looking for hard numbers, I expected to find a wave of parents installing hidden AI spyware on their kids’ phones. What the research actually shows is messier, and more interesting. Monitoring is common. AI-powered alerting is a fast-growing product category. But the best data we have suggests that most tracking is known to the teenager, and that the real fights are about something else: trust, data security, and whether the alerts are even right.
So this is a case study of what’s documented, what’s marketing, and what nobody has measured yet. Where a number comes from a company selling the product, I say so. Where I couldn’t verify something, it’s left out.
How to read the confidence tags. [High] = peer-reviewed study, major survey organization, or official regulator. [Medium] = credible journalism or a vendor’s own published data, correctly attributed. [Low] = single secondary source; treat with care.
The short version
- Monitoring is mainstream. In a 2016 Pew survey, about 6 in 10 U.S. parents of 13–17-year-olds had checked which websites their teen visited, and nearly half had looked at their teen’s calls or messages. [High]
- Location tracking is now roughly half of families in U.S. studies, and 52% of parents of 18–25-year-olds in a 2026 University of Michigan poll. [High]
- Covert use exists, but it isn’t the norm. One peer-reviewed study found 14% of parents said they tracked while their teen said they weren’t being tracked. [Medium: reported secondhand via Scientific American]
- Teens dislike surveillance-style apps. Reviewing 736 app-store reviews, researchers found most children’s reviews were one star. [High]
- The strongest predictor of what parents know isn’t snooping. It’s whether the kid tells them. [High]
- The documented harms are real: spyware-vendor breaches, and school AI monitoring that outed LGBTQ+ students. [High]
Case file 1: The baseline nobody remembers (Pew, 2016)
Before AI alerts, there was just worry. In Pew’s national survey of parents of 13–17-year-olds (fieldwork 2014–2015), a majority reported hands-on checking, while technical tools lagged behind (Pew Research Center, 2016):
| What parents reported doing | Share of parents |
|---|---|
| Talked with teen about online behavior | 92% |
| Checked which websites the teen visited | ~61% |
| Looked at teen’s call records or messages | 48% |
| Used parental controls to block, filter or monitor | 39% |
| Used controls to restrict teen’s cellphone use | 16% |
| Used tools to track teen’s location | 16% |
Two things worth noticing. First, parents preferred the old-school methods (asking, checking logins) over software. Second, this was a decade ago, before AI-driven alerting products matured. It’s a baseline, not a current measure. [High]
Case file 2: Location tracking goes mainstream
Roll forward. A 2023 Journal of Family Psychology study led by Kaitlyn Burnell (Duke) examined 729 adolescents, average age about 15, and found roughly half of parents and adolescents reported digital location tracking. Girls and younger teens were tracked more (Duke Scholars record). [High]
The authors found tracking linked to more externalizing problems and alcohol use, but those links didn’t hold up across multiple informants and statistical corrections. They call it a preliminary investigation and say the direction of cause is unknown. In other words: nobody can yet claim tracking causes worse behavior. It may simply be that parents track the kids they’re already worried about.
The same research, as reported by Scientific American, noted that 14% of parents said they used a tracking app while their child reported not being tracked. That’s the closest thing to a “secret monitoring” statistic I could find in peer-reviewed work. [Medium]
A 2024 Dutch survey of 525 parent–teen pairs reported that school tracking and location tracking had both increased over five years, and that most adolescents accepted tracking, though they accepted location tracking less than school-performance tracking (PubMed Central). [High]
And the habit doesn’t stop at 18. The University of Michigan’s Mott Poll (June 2026) found 52% of parents track a young adult child’s location; 96% of those said the child knows; 71% keep it always on; only about half gave the child a chance to opt out; and 1 in 4 said tracking sometimes makes them more anxious than reassured (Mott Poll report; NPR coverage). [High]
That last number is the one I keep returning to. Tracking is sold as anxiety relief. For a quarter of the parents who do it, it isn’t.
Case file 3: Where “AI” actually enters
This is the part people mean when they say “AI monitoring.” Products such as Bark, Aura, and Qustodio scan messages, searches, or social content with machine-learning classifiers and push an alert to the parent when something looks risky (bullying, self-harm language, predatory contact, drugs). The pitch, repeated across review sites, is that parents see flagged items rather than every message.
Here’s the honest state of the evidence:
- Independent testing of these tools’ accuracy: I couldn’t find any peer-reviewed evaluation. One blog post I came across cited a 2024 JAMA Pediatrics study of Bark’s models. I could not verify that it exists, so I left it out. If you see that claim elsewhere, ask for the DOI.
- The numbers that do exist come from the vendor. Bark’s own annual reports show the share of monitored teens who triggered at least one self-harm or suicide alert:
Share of monitored teens with ≥1 self-harm/suicide alert (Bark, vendor-reported)2020 ██████████████████████████████████░░░░░░░░░░░░░░░░ 66.6%2021 ██████████████████████████████████████░░░░░░░░░░░░ 74.6%2022 ████████████████████████████████░░░░░░░░░░░░░░░░░░ 64.3%2024 ██████████████████████████████░░░░░░░░░░░░░░░░░░░░ 60%(2023 not retrieved)Sources: Bark annual reports for 2020, 2021, 2022, 2024. [Medium: vendor data, not independently audited]
How to read that chart: it does not mean 60–75% of teens are in crisis. Bark defines the category broadly, from a text about cutting to a drafted suicide note, and counts a teen once any alert fires. The 2022 report itself notes depression alerts fell 27% in tweens and 22% in teens from the prior year, which tells you these rates move with the product’s own detection settings as much as with kids’ lives. Treat it as a measure of what the tool flags, not of teen mental health.
Chatbots are the new front. OpenAI launched parental controls for ChatGPT on September 29, 2025, including notifications to parents when the system detects signs of acute distress. Setup requires the parent to send an invitation that the teen accepts (Bitdefender summary). That’s the opposite of covert: it only works if both sides agree. [Medium]
Case file 4: What teens say back
In 2018, Pamela Wisniewski’s team at the University of Central Florida analyzed 736 public Google Play reviews of 37 parental-control apps written by children and teens. About 76% of the child reviews gave the apps a single star (abstract; CHI 2018 paper). [High]
The same research program found that authoritarian, low-autonomy parents were the most likely to use the apps, and that app use was associated with more online risks for teens, not fewer (UCF news release). Wisniewski cautioned the direction is unclear: parents of teens already having problems may be the ones installing apps. Lead author Arup Kumar Ghosh summarized what the reviews showed: kids would rather their parents talk to them. [High]
Caveat: app-store reviews are self-selected, and this predates today’s alert-only AI products. It’s the best evidence of teen sentiment, not the final word.
Case file 5: The finding that reframes the whole debate
Psychologists Margaret Kerr and Håkan Stattin studied over 1,000 Swedish 14-year-olds in 2000 and found something that still stings: what parents know about their kids’ lives tracks most strongly with what kids choose to tell them, not with how hard parents ask or control. A U.S. replication in a more ethnically diverse sample supported it (PubMed Central replication). [High]
Why that matters for AI monitoring: a tool that catches a message but erodes the child’s willingness to talk may be trading away the thing that actually protects them. That is a hypothesis consistent with this research, not something any study has tested on alert-style AI tools.
Case file 6: When surveillance software leaks
Not every monitoring product is Bark. Some are plain stalkerware sold with a parental-control label, and several have been hacked:
| Incident | What was exposed | Source |
|---|---|---|
| mSpy, 2024 | About 2.4 million unique email addresses in leaked customer-support data, plus attachments including payment-card photos | The Register, CSO Online |
| mSpy, 2018 | Call logs, texts, contacts, location data from monitored phones | SC Media |
| SpyX, 2024 (reported 2025) | Nearly 2 million accounts, plus ~300,000 linked to two clones | The Register |
| LetMeSpy (2023) and pcTattletale | Breaches reported; both shut down | The Register |
[High / Medium] Lesson: a parent who installs an all-seeing app creates a single database of a child’s private life. If the vendor is breached, the child’s data is in someone else’s hands. This is a documented pattern, not a hypothetical.
Case file 7: School AI monitoring and the people it hurt
Parents don’t control what schools install, but this is the best-documented case of AI monitoring backfiring on teens. The Center for Democracy & Technology’s Hidden Harms report (August 2022) surveyed students, parents, and teachers. Among its findings (CDT report):
- 89% of teachers said their school used student activity monitoring.
- 44% said a student had been contacted by police as a result.
- 13% of students at monitored schools said they or someone they know had been outed (sexual orientation or gender identity disclosed without consent) due to monitoring; among LGBTQ+ students the figure was 29%.
- About half of students said they weren’t comfortable expressing their true thoughts online when monitored.
[High] Senators Edward Markey and Elizabeth Warren opened an inquiry into four vendors (Gaggle, GoGuardian, Securly, Bark) in 2021. Later, The 74 reported that Gaggle dropped LGBTQ-related keywords from its tool after bias concerns. [Medium]
Different setting (schools, not parents), same technical lesson: keyword and AI flagging can’t tell a frightened teen from a dangerous one, and the consequences of a false flag land on the child.
Case file 8: What European regulators actually say
If you’re in France or the EU, you’re covered by a clearer framework than most Americans. In 2021, the CNIL, France’s data-protection authority, urged parents and vendors to favor privacy-respecting parental controls, warning against features that give a child the feeling of permanent surveillance. Its principles include proportionality (considering age and maturity, and avoiding things like permanent geolocation), transparency toward the child, and data security (CNIL, Recommendation 5). In a 2023 opinion on parental-control decrees, the CNIL also noted that local-only configurations, with no account or server upload, are better for privacy and reduce leak risk (CNIL). [High]
Note: this is regulator guidance, not a ban on parents tracking minors. French reporting suggests nothing prohibits a parent from following a minor’s location, with the CNIL advising to discuss it with the child and get their agreement once mature enough, but I only found that detail in a secondary source, so confirm it with the CNIL directly. [Low]
So: help or harm? What the evidence can and can’t say
| Question | What we know | Confidence |
|---|---|---|
| Is monitoring common? | Yes, across several large surveys | High |
| Is it mostly secret? | Mostly known to teens; a minority (14% in one study) covert | Medium |
| Does AI alerting catch real danger? | No independent accuracy study found; vendor claims only | Low |
| Does tracking cause worse teen outcomes? | Associations exist, causality unproven | High (that it’s unproven) |
| Do teens dislike surveillance-style apps? | Yes, consistently | High |
| Can surveillance vendors leak data? | Yes, repeatedly | High |
| Do open conversations beat covert snooping for parental knowledge? | Strong evidence for disclosure as the main source | High |
What nobody has measured: a randomized trial of AI-alert tools on teen safety; the long-term effect of alert-based monitoring on trust; and the false-positive rate of consumer tools on real teen messages.
A practical playbook
This is my synthesis of the research above, not a clinical protocol. Think of it as a ladder: start at the bottom and climb only if you need to.
| Rung | What it looks like | Why |
|---|---|---|
| 1. Talk first | Ask what apps they use and what worries them. Share what worries you. | Disclosure is what predicts parental knowledge |
| 2. Agree on rules in writing | Screen time, what you’ll check, what you won’t | Predictable beats surprising |
| 3. Built-in, transparent tools | Phone-maker family settings, shared location the teen can see | No third-party vendor holds the data |
| 4. Alert-only AI tools, with the teen’s knowledge | Flagged items only, not full message access | Narrower collection, less trust damage |
| 5. Full-content monitoring | Only for a specific, serious safety concern, ideally time-limited | Highest breach and trust cost |
Rules I’d follow:
- Tell your teen. Nearly every expert and regulator quoted here points to transparency. Covert use is where the research shows trust damage and a surprise discovery risk.
- Skip anything marketed as hidden or “stealth.” Several of those vendors were breached.
- Check the vendor. Search the company name plus “breach,” and look up your own email at Have I Been Pwned.
- Prefer local or minimal data where possible, as the CNIL suggests.
- Plan the exit. Decide now at what age or milestone each control comes off. Dutch and U.S. data both show teens accepting oversight they understand and that eases over time.
- When an alert fires, don’t ambush. A flag is a prompt to ask, not proof. The school cases show how often flags misread kids.
- Don’t let a tool replace the relationship. If you only have one rung available, make it the first.
If an alert is about self-harm, treat it as a reason to reach out and get support, not a reason to punish. In the U.S., the 988 Suicide & Crisis Lifeline offers call or text support; in France, the national number is 3114. If someone is in immediate danger, call local emergency services.
Sources and sourcing notes
- Pew Research Center, Parents, Teens and Digital Monitoring (2016)
- Burnell et al., Journal of Family Psychology (2023), digital location tracking
- Dutch parent–adolescent tracking study (2024)
- University of Michigan Mott Poll, Safety or Surveillance: Tracking of Young Adults (June 2026)
- Ghosh, Wisniewski et al., CHI 2018, Safety vs. Surveillance
- Kerr & Stattin (2000); U.S. replication
- Center for Democracy & Technology, Hidden Harms (2022)
- CNIL, Recommendation 5 (2021)
- Bark annual reports (vendor-published)
- Breach reporting: The Register, CSO Online, TechCrunch / SC Media coverage
What I excluded: an unverifiable “JAMA Pediatrics” claim about Bark’s accuracy; any stat I couldn’t trace to a primary or clearly attributed source.
Limits: Most peer-reviewed work predates today’s AI-alert products. Survey data is self-reported. Direction of cause is mostly unresolved. If you spot an error, tell us, and we’ll fix it publicly.
Last updated: November 6, 2026.