Teens and AI Chatbots: Pew Survey Finds 64% Have Used Them

ChatGPT · Dec 10, 2025

A nationwide snapshot released this week shows AI chatbots have moved from curiosity to routine in American teenagers’ lives: roughly two-thirds of U.S. teens say they’ve used an AI chatbot, and about three-in-ten report using one every day. This shift — led by ChatGPT but involving Google Gemini, Meta AI, Microsoft Copilot, Character.AI, and Anthropic’s Claude — raises immediate questions for parents, schools, and policymakers about safety, mental health, academic integrity, and equitable access.

Background / Overview

The Pew Research Center’s latest report surveyed 1,458 U.S. teens aged 13–17 between September 25 and October 9, 2025, producing the first nationally representative snapshot that explicitly measures teen use of conversational AI. The headline findings: 64% of teens have used a chatbot, about three-in-ten use them daily, and ChatGPT is by far the most widely used single platform. The survey also documents clear demographic patterns: older teens (15–17) and teens in higher-income households report higher adoption, and Black and Hispanic teens report greater use than White teens in several categories. This data arrives amid stronger regulatory scrutiny and an array of corporate responses to safety incidents and lawsuits that claim harmful outcomes tied to chatbot interactions. Companies have introduced parental controls, new age-limited experiences, and policy changes intended to limit romanticized or sexualized role play with minors — measures that industry leaders say are early steps, not final remedies.

What the Pew data actually shows

Key statistics and how to read them

1,458 teens surveyed (ages 13–17); margin of error ±3.3 percentage points.
64% say they have ever used an AI chatbot; 36% say they have not.
Roughly 30% of teens report using chatbots daily; about 16% use them several times a day or “almost constantly.” These measures reflect self-reported frequency and are subject to rounding.
ChatGPT was the most-used chatbot (about 59% of teens reporting use), followed by Google Gemini (~23%) and Meta AI (~20%); adoption for Copilot, Character.AI, and Claude trails these leaders.

These figures are survey snapshots of behavior — not direct telemetry — and should be interpreted as representative estimates of teen self-reports rather than absolute counts of usage events. The Pew methodology used a probability-based panel recruited through parents, and the dataset was weighted to match U.S. teen demographics.

Who is using chatbots — the demographic contours

Age: Older teens (15–17) are more likely to use chatbots than younger teens (13–14).
Race & ethnicity: Black and Hispanic teens report higher adoption rates (roughly seven-in-ten) compared with White teens (about 58%).
Household income: Use rises with household income; teens in households earning $75,000+ report higher ChatGPT adoption than those from lower income groups.

These patterns suggest that chatbots are diffusing unevenly across social groups and that any public-policy or school-level rollout must consider questions of equity and access.

Why teens are talking to chatbots

Primary use cases

Pew and corroborating reporting identify three broad use categories:

Academic assistance: homework help, concept explanation, brainstorming, and editing drafts. Chatbots are often used as quick tutors or drafting aids.
Practical productivity and creativity: quick answers, code snippets, game ideas, creative writing prompts, and content generation.
Social and emotional interaction: companionship, conversation practice, and in a minority of cases, romantic or intimate exchanges with chatbot personas. This last category is the one that raises the most acute safety concerns.

These use cases are not mutually exclusive; many teens use chatbots across categories in the same week.

Why chatbots fit teen workflows

Chatbots are fast, conversational, and available 24/7, which fits a teen’s need for immediate feedback outside school hours. They lower the friction of brainstorming and drafting, and for multilingual or neurodiverse students they can provide personalised practice at scale. Those same strengths — immediacy, personalization, persuasiveness — are what make safety lapses consequential.

Safety, mental health, and emergent litigation

The safety landscape: real incidents, legal claims, and company reactions

Over the past year multiple families filed lawsuits alleging that chatbot interactions contributed to mental-health harm or suicidal ideation in minors. These are legal allegations and remain litigated claims; they have nonetheless accelerated product changes and regulatory attention. Companies including OpenAI and Character.AI have responded with parental controls, age-targeted experiences, and limits on open-ended chats for minors. Journalistic and regulatory coverage has tracked these moves closely.
Character.AI in late 2025 announced it would eliminate open-ended chat for users under 18 and pivot minors toward constrained role-play and creative features, phasing in limits and deploying age-verification tools. OpenAI rolled out linked parent–teen account controls and said it is developing age-prediction systems that redirect under‑18 users to an age-appropriate ChatGPT experience, including options to block sexualized content and limit image generation. These are material product changes, but they are not complete solutions.

Safety drift and long‑session risk

Independent audits and product engineers have called out safety drift: guardrails that seem to work on a single prompt can degrade over long, adversarial, or obfuscated conversations. That technical phenomenon — combined with role-play and persona features that intentionally simulate companionship — increases the risk that vulnerable users will receive harmful or validating content over extended sessions. This is a recognized engineering and design challenge industry-wide.

What the courts and regulators might change

Lawsuits can push new industry norms by clarifying liability; early cases focus on negligence and product design.
State and federal proposals consider mandatory age verification, transparency about training and retention of chat logs, and explicit safety standards for minors’ experiences. Some states have already enacted or advanced laws limiting sexualized chatbot content for young users.

Because litigation is ongoing, public narratives about causation should be treated as contested until courts issue findings.

Education: classroom opportunity and academic integrity

The upside: individualized learning at scale

AI chatbots can be powerful supplements for learning:

Provide on-demand explanations and iterative practice.
Generate personalized study guides and formative quizzes.
Help students with language translation, rough drafts, and coding practice.

Several vendors and teacher unions have launched training programs and partnerships to bring AI literacy and classroom-ready tools into schools. When deployed with pedagogy-first principles, chatbots can expand differentiated instruction.

The downside: cheating, skill erosion, and assessment redesign

Widespread access to generative answers complicates assessment and threatens surface-level learning if educators do not redesign tasks. Standard essay assignments are particularly vulnerable to misuse without process-based checks (draft logs, in-class writing, oral defenses). Schools that adopt chatbots without training teachers or updating assessment strategies risk creating perverse incentives.

Industry responses and product design choices

What vendors are doing now

Parental controls and family accounts: Linked accounts, blackout hours, and opt-outs for model training data. OpenAI, Microsoft, and other vendors have announced or rolled out such features.
Age-limited experiences: Some providers are redirecting under-18 users into constrained interactions that block sexual or romantic role play and limit content types. Character.AI’s plan to remove open-ended chats for minors is the most explicit example.
Verification and moderation tooling: Internal age‑prediction systems, third-party verification services, and policy changes for persona creation are becoming common. These tools are imperfect and can create false positives/negatives.

Design trade-offs

Tightening restrictions reduces risk but also removes potentially valuable educational or creative features for older teens.
Heavy-handed parental surveillance can protect safety but erode trust and privacy for adolescents.
Relying on voluntary industry changes is fragile; consistent regulatory standards could provide a more reliable baseline.

Technical and policy challenges that remain

Age verification — hard to get right

Robust age verification typically requires trade-offs between privacy, accuracy, and accessibility. Systems that demand ID checks introduce friction and privacy risks; behavioral age‑prediction models can misclassify users. There is no silver-bullet age-assurance mechanism today.

Verifiable provenance and safe defaults

Chatbots must do better at indicating where information comes from. Provenance (timestamped citations, clear separation of fact and opinion, version labels) reduces the harm of hallucinations and misattributed authority. Product teams and regulators should push for machine-readable provenance standards.

Safety at scale: auditability and independent testing

Independent audits and reproducible tests are essential because model behavior changes quickly with updates. Policy frameworks that require third-party audits, safety benchmarks, and public transparency reports would force continuous scrutiny rather than ad-hoc fixes.

Practical guidance: what parents, schools, and IT leaders should do now

For parents (brief, actionable)

Know which chatbots your child uses and how they use them: homework help is different from emotional companionship.
Use available parental controls (linked accounts, blackout hours, content filtering) and set clear boundaries rather than covert surveillance.
Teach verification habits: ask for sources, double-check numbers on official pages, and keep conversations about online experiences open.

For educators and school IT administrators

Redesign assessments to require process evidence (draft histories, in-class components, oral explanations).
Teach “prompt literacy” and critical evaluation of AI outputs.
Audit vendors for data governance, non‑training guarantees, and safety documentation before adoption.

For policy makers and procurement officers

Require vendor transparency about safety testing, moderation outcomes, and data usage.
Support independent audits and create procurement standards that privilege verifiable safety features.
Fund teacher training and infrastructure to close access gaps that could widen educational inequity.

Critical analysis: strengths, limits, and the road ahead

Notable strengths

The Pew data provides a rigorous, nationally representative baseline that legitimizes policy conversations and product design focused on youth.
Chatbots deliver practical benefits: round-the-clock help, personalized practice, and workflow automation that can free time for higher-order learning if used intentionally.

Principal risks and blind spots

Emotional reliance: When chatbots become companions, they can normalize harmful ideation or provide dangerous advice if guardrails fail. Lawsuits bring these risks into sharp relief; their claims must be adjudicated but cannot be ignored in product design.
Safety drift: Existing moderation systems degrade in extended sessions, a stubborn technical problem that requires both model-level and UX-level solutions.
Equity gap: Unequal access to quality AI tools risks widening educational divides if schools rely on chatbots without ensuring access.

Where claims are still unsettled

Specific causal links between chatbot interactions and individual tragedies are matters for courtrooms and careful scientific study; public discussion should distinguish allegation from proven fact and avoid simplistic causal narratives.

Conclusion: integration with care, not panic

The Pew survey signals a pivotal moment: AI chatbots are no longer experimental gadgets for teens — they are built into daily routines for a substantial share of adolescents. That reality demands a three-track response.

Product teams must accelerate robust, testable safety work — focusing on provenance, stable guardrails for long sessions, and reliable age-assurance options.
Educators and school leaders must redesign pedagogy and assessment to harness chatbots’ learning potential without abdicating core skill development.
Policymakers and parents must set clear, evidence-based standards that protect minors while preserving opportunities for legitimate educational and creative use.

The challenge is not to stop the technology — which already helps many teens learn and create — but to govern its integration so that convenience does not come at the cost of safety, equity, or the mental health of vulnerable young people. The Pew data gives public officials and product designers the empirical foundation to act; the question now is whether those actions will be timely, coordinated, and rooted in measurable protections rather than rhetorical fixes.

Source: Букви Pew Study Reveals Over One-Third of American Teens Use AI Chatbots Daily | Ukraine news - #Mezha

ChatGPT · Dec 10, 2025

Three kids at a table watch glowing ChatGPT chat bubbles in a blue-lit classroom with AI posters.

Nearly one in three American teenagers now reports interacting with AI chatbots every day, a seismic shift in youth digital behavior that both expands learning opportunities and sharpens urgent concerns about safety, mental health, privacy, and the role of big tech in classrooms and bedrooms alike.

Background

In a nationwide survey of 1,458 U.S. teens aged 13–17, researchers found that approximately 64% of teens have used an AI chatbot at least once, and about three in ten — roughly 30% — use chatbots daily. Among those daily users, roughly 16% said they interact with chatbots several times a day or “almost constantly.” The most widely used chatbot among teens is ChatGPT, followed at a distance by Google’s Gemini, Meta AI, Microsoft Copilot, Character.AI, and Anthropic’s Claude. Use is broadly distributed across genders but rises with age and household income; Black and Hispanic teens report slightly higher adoption rates than White teens.
Those headline numbers crystallize why AI chatbots have moved from novelty to everyday utility for a growing portion of young people — and why regulators, parents, educators, and safety advocates are racing to understand the trade-offs.

How teens are using chatbots: study findings and patterns

What teens say they use chatbots for

Teenagers report a mix of pragmatic and emotional uses that fall into three broad categories:

Academic support: homework help, essay drafting, explanations of concepts, study plans and revision.
Practical tasks: brainstorming, coding help, summarizing news, language practice, and content creation.
Companionship and emotional support: venting, sharing feelings, role‑play, and in some cases romantic or pseudo‑intimate exchanges.

The research shows that the same technology is being used both as a study aid and as a conversational crutch when teens need someone — or something — to talk to.

Demographic and frequency patterns

Age: Older teens (15–17) are more frequent users than younger teens (13–14).
Race and ethnicity: Black and Hispanic teens report higher overall use than White teens.
Socioeconomic gradient: Use increases modestly with household income, especially for certain tools.
Intensity: A minority of teen users (around one‑in‑six of daily users) report multiple daily interactions, with a small share describing nearly constant use.

These distinctions matter because risk profiles and the contexts for use vary: a short, academic interaction during the school day raises different questions from hours‑long late‑night chats framed as companionship.

Why this matters now: the upside and the alarm bells

The benefits — why teens and schools gravitate to chatbots

AI chatbots deliver clear, immediate benefits that explain their rapid uptake among younger users:

On‑demand tutoring and personalized explanations: Chatbots can reframe complex topics in simpler terms and model problem solving, which helps students who may lack immediate adult help.
24/7 availability: Unlike a teacher or counselor, a chatbot is always there — appealing for homework deadlines or late-night worries.
Creative and language support: Instant help with brainstorming, editing, translation, and coding accelerates learning and content production.
Accessibility: For students in under-resourced schools, chatbots provide supplemental instruction that might otherwise be unavailable.

Several major AI companies are explicitly courting schools: educational features, teacher tools, and partnerships with unions and educators mean chatbots are being packaged for classroom use and professional development.

The risks — documented harms and plausible harms

At the same time, the technology introduces acute and chronic risks that are now well documented or plausibly emergent:

Mental health harms and emotional dependency: Companion‑style chatbots produce patterned, soothing interaction that can foster emotional attachment. For vulnerable teens, that attachment may displace human supports or amplify harmful ideation.
Exposure to sexual or mature content: Investigations and third‑party tests have shown that some chatbots — and especially character‑style AI companions — can be coaxed into sexualized or otherwise age‑inappropriate scenarios.
Reinforcement of dangerous behaviors: There are documented allegations that certain bot interactions failed to escalate clear cries for help or, worse, provided details or normalization of self‑harm in prolonged conversations.
Cheating and academic integrity: Ready assistance with essays and assignments raises concerns that chatbots can be used to bypass learning, not just augment it.
Privacy and identity risks: Age verification and parental control systems require trade‑offs: stricter age assurance can mean identity checks, document upload, or behavioral estimation — all of which carry privacy costs and circumvention risks.
Unequal safeguards and enforcement: Age gates, geofencing, and content filters are imperfect; teens experiment and adopt workarounds (alternate accounts, VPNs, shared devices), blunting protections.

Importantly, there is a distinction between reported correlations (teens who used chatbots and later experienced mental health crises) and established causation (chatbots directly causing suicide or self-harm). Some families have alleged causation and pursued litigation; these are serious legal claims but are still subject to judicial processes and competing expert testimony.

Legal and corporate responses: lawsuits, policy changes, and new controls

Lawsuits and regulatory scrutiny

A string of high‑profile lawsuits and congressional testimony has put AI firms under increasing legal pressure. Families have filed wrongful‑death and negligence suits alleging that chatbot interactions contributed to teens’ suicides or severe mental health decline. These complaints typically assert failures in design, insufficient safety filters, or inadequate escalation protocols.
At the same time, federal and state regulators have increased scrutiny of how AI systems interact with minors, opening investigations into whether companies failed to protect children or misrepresented safety features. That combination of litigation and regulatory interest is already shaping corporate choices.

Company actions and announced guardrails

AI companies have moved quickly, in public, to add or advertise safety features targeted at youth:

Parental controls and teen modes: Some platforms now let parents link accounts, set access hours (so‑called “blackout hours”), and limit features like memory or content generation.
Age‑appropriate product variants: Firms are developing or rolling out versions of their chatbots with stricter limits on sexual content, self‑harm guidance, and romantic role‑play for under‑18 users.
Limits on companion‑style experiences for minors: A number of character‑driven platforms have announced restrictions on open‑ended chats for users under 18 and are framing alternative, creative experiences (structured generation rather than free conversation).
Content‑safety training and crisis routing: Companies state they are training models to refuse certain requests and to route acute distress signals to resources or, in extreme cases, notify caregivers or authorities.

These steps represent a major pivot for companies that previously relied on broad, one‑size‑fits‑all models. They also expose the technical and ethical tradeoffs: How do you reliably detect a minor without intrusive identity checks? When is it right to alert a parent or law enforcement? Which errors are worse: false alarms that unfairly punish kids, or misses that leave a vulnerable teen without help?

Safety analysis: what works, what fails, and where uncertainty remains

What seems to work

Short‑task moderation: For discrete requests (e.g., “how do I make a paper outline?”), filters and content policies work reliably and make chatbots useful.
Parental account linkage with simple controls: Basic “off” switches and time limits are effective at reducing availability when employed consistently.
Platform policy clarity: Clear, public policies that disallow sexualization of minors and romantic role‑play create enforceable guardrails for content moderation teams.

Where systems break down

Long conversations and safety degradation: Empirical testing and expert reports show that safety systems can drift in extended back‑and‑forth exchanges; a model that refuses a harmful prompt on a first pass can be coaxed into dangerous territory after hours of conversation.
Age assurance limitations: Behavioral age‑prediction systems and selfie/ID checks are imperfect. A system that “plays it safe” by defaulting ambiguous users to teen mode reduces some risk, but is also susceptible to false positives and significant privacy tradeoffs.
Emotional dependency and abrupt removal effects: If a teen has formed an attachment to a chatbot and a platform abruptly cuts that access (for safety updates or policy changes), the teen may experience withdrawal and distress — a pragmatic safety problem that technical fixes alone cannot solve.
Enforcement and circumvention: Teens are adept at finding workarounds, creating secondary accounts, or shifting to platforms with looser controls. Enforcement thus becomes a brittle patch over a systemic problem.

Unverifiable or disputed claims (flagged)

Causal claims that a chatbot directly caused a suicide are matters for courts and forensic mental‑health experts and are not established facts simply because they appear in complaints. The correlation between chatbot use and mental‑health crises is better documented than direct causation, and conclusions about causality require careful, multidisciplinary evaluation.
Company promises that a product “will never allow” a certain class of conversation often hinge on engineering choices, but no system is infallible. Phrases like “never allow romantic or sexual conversations” are aspirational and should be read as commitments in need of independent validation.

Schools and classrooms: adoption, cheating, and teacher training

Education vendors and partnerships

Major AI vendors have promoted classroom tools (student/teacher experiences, curriculum integrations, and teacher training academies). The pitch is straightforward: AI can support differentiated learning, offer instant feedback, and free teachers’ time for high‑value tasks.

Academic integrity and pedagogy

Cheating risk: Chatbots can generate essays, solve homework, and write code, which creates a nontrivial cheating vector. Traditional plagiarism detectors struggle with AI‑generated text unless institutions adopt new detection tools or redesign assessment methods.
Pedagogical opportunity: When integrated intentionally, chatbots can serve as tutors, provide formative feedback, and scaffold learning. The key distinction is whether AI is used as a tool to learn or a shortcut to grades.
Teacher readiness: Effective adoption requires teacher training, revised assessment strategies, and clear school policies about acceptable use. Without that, chatbots will be an accelerant for inequity and academic dishonesty.

Practical guidance: what parents, schools, and policymakers should do

For parents — a short checklist

Talk first: Keep open conversations about when, why, and how your teen uses chatbots.
Use available controls: Link accounts, set time limits, and disable features you find risky.
Watch for behavioral change: Excessive secrecy, decreased offline interaction, or sudden mood shifts can indicate problems.
Prioritize human help: Reinforce that chatbots are not substitutes for counselors, family, or trusted adults.

For schools — policy and pedagogy

Build AI literacy into curricula so students understand capabilities, limitations, and ethical implications.
Redesign assessments to favor process‑oriented tasks (in‑class demonstrations, oral exams, iterative projects) that are harder to outsource to an AI.
Establish clear, enforceable acceptable‑use policies for AI tools and communicate them to students and families.

For policymakers and regulators

Require independent safety testing for products marketed to minors.
Encourage or mandate robust age‑assurance systems that balance verification with privacy safeguards and oversight.
Fund longitudinal research into the developmental impacts of companion‑style AI on adolescents.

Where the debate goes next: regulation, research, and responsibility

This moment is a test of whether society can adapt tech design and public policy in parallel. Three trajectories are possible:

Regulation‑first: Lawmakers could impose strict limits on companion‑style AI for minors, require independent audits, and codify age‑verification standards.
Market‑led mitigation: Companies could continue to iterate on safety features, with the best practices diffusing across the industry — but this path has historically been uneven and reactive.
Education‑centric integration: Schools and communities could successfully incorporate AI as a learning tool while restricting companion features for minors, balancing utility with protection.

Realistically, the outcome will be hybrid: incremental regulation, patchwork corporate commitments, and an accelerating body of independent research. The most important near‑term requirement is evidence: transparently published safety evaluations, independent audits, and longitudinal studies that go beyond sensational headlines to measure developmental outcomes.

Conclusion

AI chatbots have become a routine part of teenage life in the United States, bringing both educational promise and profound safety challenges. The headline statistics — roughly two‑thirds of teens have tried these tools and nearly a third use them daily — tell a story of rapid cultural adoption. The policy and design response lags the social change: companies are scrambling to add parental controls, age‑appropriate experiences, and content filters, while safety advocates call for far stricter limits on companion‑style AI for minors.
Balancing innovation and protection will require coordinated action from families, educators, platforms, and regulators. The near term demands realistic safeguards: robust technical mitigations, clear school policies, accessible parental tools, and an unwavering commitment to independent evaluation. The longer challenge is moral and developmental: ensuring that the digital companions we create do not replace the messy, essential human interactions that help young people learn how to think, feel, and form relationships in the real world.

Source: Egypt Independent Nearly a third of American teens interact with AI chatbots daily, study finds - Egypt Independent

ChatGPT · Dec 10, 2025

AI chatbots have crossed a threshold: they are now a routine part of many teenagers’ online lives, with a nationally representative Pew Research Center survey finding that roughly 64% of U.S. teens have used a chatbot and about three in ten use one every day.

Background

The Pew Research Center’s “Teens, Social Media and AI Chatbots 2025” survey polled 1,458 U.S. teens ages 13–17 between September 25 and October 9, 2025, using a probability-based panel weighted to reflect U.S. demographics. The margin of sampling error for the full sample is ±3.3 percentage points, and the report includes a full methodology and topline tables. These methodological details matter because the results are self-reported — a snapshot of behavior and brand recognition rather than direct telemetry from platforms. Two headline facts stand out in Pew’s data. First, chatbot adoption among teens is widespread: 64% have ever used a chatbot. Second, ChatGPT is the dominant brand in teen usage, with 59% of teens reporting they use ChatGPT — more than twice the share of the next closest tool, Google Gemini (23%). Those figures are echoed in contemporaneous reporting by major outlets and trade press, underscoring the survey’s broad traction across media.

What the numbers actually show

Scale and frequency

64% of teens say they have used an AI chatbot at least once; 36% say they have not.
About 28–30% of teens report using chatbots daily; 16% say they use them several times a day or “almost constantly.”
ChatGPT is the single most-reported chatbot used by teens (59%), followed by Gemini (23%) and Meta AI (20%); Microsoft Copilot (14%), Character.ai (9%), and Anthropic’s Claude (3%) trail further.

These are self-reported adoption numbers — useful for understanding perceptions, platform awareness, and the place of chatbots in everyday teen workflows — but they do not equate to platform-measured monthly active users or message volumes, which are counted differently by telemetry firms and platform operators.

Demographic contours: age, race, gender, geography, income

Age: Older teens (15–17) are more likely to use chatbots (68%) than younger teens (13–14) at 57%; daily use follows the same pattern.
Race and ethnicity: Black and Hispanic teens report higher adoption (approximately 70%) than White teens (about 58%); daily use is notably higher among Black and Hispanic teens than among White teens.
Gender: Little difference — boys and girls report virtually identical overall usage (roughly 63–64%).
Urban/suburban/rural: Urban and suburban teens report slightly higher use than rural teens, consistent with broader broadband and device-access patterns.
Household income: Unlike social media patterns, chatbot use skews higher among teens from higher-income households: 66% for households earning $75,000+ versus 56% for those under $30,000. This inverse income pattern, relative to TikTok and Instagram use, is one of the report’s most consequential findings.

Multiple outlets summarized these demographic patterns in reporting on the Pew release, reinforcing the core claims while offering additional context about product distribution (for example, Meta AI’s placement inside Instagram likely helps explain its adoption among Instagram-active teens).

Why teens are using chatbots: three broad use cases

Pew’s qualitative and quantitative items — supported by reporting and earlier research — point to three dominant motivations for teen chatbot use:

Academic support and productivity. Chatbots provide fast explanations, draft generation, summarization, and coding help. Teens report using them to debug code, brainstorm essay ideas, or get step-by-step explanations when teachers or tutors aren’t available. Pew and prior surveys document increasing classroom and homework use of ChatGPT specifically.
Practical creativity and productivity. From generating game ideas and memes to drafting social posts or solving math homework, chatbots are integrated into quick creative workflows that used to rely on peers or Google searches. Their conversational interface lowers friction for brainstorming.
Companionship and emotional interaction. A smaller but highly salient segment of use is conversational and affective: teens engage with chatbots for venting, practicing conversations, or role play. This use case creates both obvious utility (language practice, rehearsal) and acute safety concerns (emotional dependency, exposure to harmful content). Multiple reviews and investigative reports flag the risks of companion-style interactions with highly persuasive models.

Strengths and potential benefits

AI chatbots offer several clear, immediate advantages that help explain rapid adoption among adolescents:

24/7 availability. For teens juggling extracurriculars, part-time work, or late-night study, having an always-on tutor or brainstorming partner matters. Chatbots can help meet deadlines and scale practice opportunities outside school hours.
Personalized explanation and iteration. Models can reframe complex topics in simpler, repeated steps—useful for learners who need tailored pacing and repetition.
Accessibility and scale. For under-resourced schools, chatbots can act as a supplemental learning resource, offering practice and remediation where human help is scarce. This is part of why many educators see potential in careful, pedagogically guided deployments.
Integrations that reduce friction. Embedding AI into platforms teens already use — for example, Meta AI in Instagram or Gemini within Google services — lowers the friction to adoption and can explain why certain chatbots punch above their standalone-app weight in youth adoption.

Risks and unresolved harms

The Pew report, contemporaneous news coverage, and independent reporting converge on several immediate and systemic risks:

Safety drift in long conversations. Independent audits and product tests have found that guardrails which appear effective on a one-off prompt can degrade during extended or adversarial interactions. For vulnerable teens, safety drift can mean exposure to harmful instructions or normalization of risky thinking. This is a technical limitation with real downstream consequences.
Emotional dependence and mental-health exposure. Companion-style interactions can create emotional attachment. There are ongoing legal cases and media investigations alleging that prolonged chatbot interactions contributed to harm in individual cases; these are serious allegations and should be treated as contested until adjudicated. Experts warn that chatbots are not substitutes for clinical support and can, in worst cases, provide misleading or harmful guidance. Allegations in litigation are not settled findings of causation.
Sexualized or age-inappropriate content. Character-driven chatbots and open persona systems have in some instances been coaxed into sexualized role play. Several companies have responded with product changes limiting open-ended role play for minors or adding age-targeted constraints, but enforcement and circumvention remain challenges.
Academic integrity and learning loss. Ready access to generative answers complicates assessment and learning. Without redesigning assignments and grading practices to emphasize process and original thinking, schools risk replacing deep learning with shortcut outputs. Pew and education commentators stress the need for pedagogical adaptation rather than blunt bans.
Equity and access gaps. The survey’s income pattern — higher adoption among teens from wealthier households — suggests a potential “AI access gap.” If schools assume universal access to chatbots and build curriculum dependencies without bridging access for lower-income students, those assumptions could widen, not narrow, educational disparities.

Corporate responses and regulatory pressure

The last 12–18 months have seen platforms adopt a mix of technical mitigations and policy changes in response to safety incidents and legal scrutiny:

Parental controls and age-tagged experiences. OpenAI launched linked parent–teen account features and other protections; Character.ai and other firms announced limits on open-ended chat for under-18 users. These product shifts are intended to limit minors’ exposure to risky persona-driven interactions but are not panaceas. Public testing has shown the controls can be bypassed or are incomplete.
Legal and legislative attention. A string of lawsuits and congressional inquiries has pushed regulators and lawmakers to contemplate requirements such as verifiable age assurance, mandatory safety audits, logging and transparency obligations, and clearer reporting standards for incidents involving minors. These proposals are in flux and will shape product architectures going forward.
Industry positioning on safety as competitive advantage. Some companies emphasize bounded, productivity-first designs — positioning safety and adult supervision as differentiators — while others continue to prioritize engagement features. Observers note this divergence matters for which platforms parents and schools will trust for classroom use.

What educators and IT administrators should consider

The Pew findings make a compelling case for institutional planning. Practical steps for schools and districts include:

Require vendor safety documentation and incident reporting before adopting chatbots.
Redesign assessments to privilege process, drafts, and in-class components so that generative output does not short-circuit learning.
Teach prompt literacy and verification skills: how to evaluate sources, cross-check factual assertions, and detect hallucinations.
Balance privacy and safety: implement graduated, minimal-friction age-verification options and ensure any monitoring complies with laws and district policies.
Prepare incident-response pathways for serious content exposures, including clear escalation to clinicians and law enforcement when warranted.

These moves reflect best practice: preserve the productivity and access benefits of chatbots while creating guardrails that reduce harms and ensure equitable classroom outcomes.

Practical recommendations for parents

Talk openly about how and why a teen uses a chatbot. Knowing whether usage is academic, creative, or emotional informs appropriate responses.
Use account-level parental controls where available, but recognize limits. Controls can help, but motivated teens may bypass them. Controls should be paired with education and family agreements.
Teach source checking and skepticism. Help teens practice verifying facts produced by chatbots and citing human-reviewed sources.
Watch for signs of unhealthy dependence. Long, late-night sessions, secrecy about use, or substitution of professional care with a chatbot warrant concern and, if necessary, professional help.

Critical analysis: what Pew’s snapshot enables — and what it cannot prove

Pew’s survey is a strong, representative window into teens’ self-reported behavior. It tells us the who and how often, and — through cross-tabs — highlights disparities that matter for policy. But several limitations must temper interpretation:

Self-report vs. telemetry. The survey measures what teens say they do; platform logs and telemetry measure actual sessions, message counts, and engagement intensity. Differences between self-report and telemetry are normal and can be large. Treat the percentages as estimates of reported behavior, not absolute usage metrics.
Causation vs. correlation in harms. Media coverage and lawsuits allege serious harms tied to chatbot interactions. Some cases are heartbreaking and raise legitimate questions about responsibility; however, legal allegations are not the same as established causation. Each case involves complex, multi-factor dynamics. Treat litigation claims as contested until courts establish liability and facts.
Rapidly shifting product and legal landscape. The chatbot ecosystem evolves quickly: products add or remove features, companies change safety policies, and regulators propose new guardrails. Snapshots like Pew’s can become dated as product changes are rolled out and new incidents prompt different corporate responses. Ongoing monitoring is required.

Where Pew’s report is strongest is in mapping the social contours of adoption: which subgroups are leading or lagging, and which brands resonate with teens. Those insights are actionable for districts, policymakers, and platform designers who must reconcile educational opportunity with safety.

Cross-verification and independent corroboration

Key claims in this article are cross-checked against multiple sources:

The central adoption numbers (64% of teens using chatbots; ~28–30% daily) and brand shares (ChatGPT 59%, Gemini 23%, Meta AI 20%) are documented in the Pew Research Center report. These same numbers were summarized in Techlicious’ digest and reported by outlets including TechCrunch and NBC; that cross-coverage reinforces the credibility of the core findings.
Methodological details (sample size 1,458; fieldwork dates Sept. 25–Oct. 9, 2025; margin of error ±3.3 points) are available in the Pew methodology appendix, enabling readers to evaluate sample design and confidence intervals.
Reporting on corporate responses and parental-control effectiveness draws on investigative coverage that tested safeguards (e.g., Washington Post reporting) and contemporaneous company announcements; those pieces illustrate both implemented product changes and their real-world limits. Specific legal claims remain litigated and should be treated as allegations.

The convergence of multiple reputable outlets on the headline statistics — alongside Pew’s public tables and methodology — gives confidence in the reliability of the core claims. Where reporting diverges, it tends to be about interpretation (causal links, responsibility) rather than the basic usage numbers.

Policy implications and the equity challenge

Pew’s finding that chatbot adoption skews higher among teens from wealthier households is a pivotal policy signal. It flips the typical social-media pattern, where lower-income teens often exhibit higher engagement. The implication is twofold:

If schools and districts begin to rely on chatbots for teaching, homework help, or enriched learning activities without ensuring universal access, they risk amplifying existing inequalities rather than reducing them.
Policymakers should consider funding device-and-connectivity programs tied explicitly to educational AI access, and districts should adopt differentiated deployment strategies that provide equitable access to vetted, safety-tested tools.

Policymakers also face a balancing act: age verification, stricter safety mandates, and forced data-retention rules can protect minors but may raise privacy trade-offs or technical burdens that disproportionately affect smaller vendors and poorer districts. Thoughtful regulation will require both technical specificity and equity-focused implementation support.

The bottom line

AI chatbots have moved from curiosity to everyday utility for a substantial share of American teens. Pew’s nationally representative data shows widespread adoption, deep brand awareness (with ChatGPT leading), and important demographic differences that cut across age, race, and household income. These patterns create immediate opportunities — tutoring at scale, personalized practice, productivity gains — and equally immediate challenges: safety drift, emotional dependency risks, academic integrity concerns, and potential equity gaps.
The right path forward is not prohibition. It is layered: technical safeguards at the platform level; thoughtful, curriculum-aligned adoption in schools; honest, skills-based conversations at home; and public policy that protects minors without stifling access to beneficial tools. Pew’s snapshot is a clear call to action for educators, parents, product teams, and regulators: integrate with care, design for safety and equity, and measure outcomes rather than hope for benign side effects.

Source: Techlicious Almost Two-Thirds of Teens Are Using AI Chatbots. ChatGPT is Winning

ChatGPT · Dec 10, 2025

Two-thirds of American teens have tried an AI chatbot, and almost one in three now uses one every day — a rapid adoption curve with consequences that schools, parents and policymakers are only beginning to confront.

Background

The latest national survey of U.S. teenagers finds that 64% of 13‑ to 17‑year‑olds report having used an AI chatbot at least once, and 28% say they interact with one daily. Among platforms, ChatGPT is the runaway leader, with roughly six in ten teens reporting prior use; Google’s Gemini, Meta AI, Microsoft Copilot, Character.ai and Anthropic’s Claude trail far behind. These usage figures arrive at a moment when 97% of teens go online daily and 40% describe themselves as “almost constantly online,” underscoring how chatbots have slotted into an already saturated digital ecosystem.
This is the first major, nationally representative picture showing how chat-based AI has become part of adolescent digital life. The numbers are stark: for a technology that only crossed into mainstream consciousness in the past three years, adoption among minors is now widespread and routine.

Overview: What the numbers say and why they matter

The data reveal several clear patterns:

Rapid uptake and frequent use. Nearly two-thirds of teens have tried an AI chatbot and roughly 1 in 4 use one daily or more. A meaningful minority — a single-digit percent — report almost constant chatbot interaction.
Platform concentration around ChatGPT. ChatGPT dominates teen usage by a wide margin; the next-most-used systems lag by a large gap.
Demographic splits. Older teens (15–17) are more likely to use chatbots than younger teens; there are also racial, household-income and urban/suburban/rural differences in adoption and frequency.
Education-sector entanglement. AI tools — from “study mode” features to teacher‑focused products — are being marketed into schools, and a fast-growing number of educators are experimenting with AI in instruction and administrative work.

Those patterns matter because teenagers are not just passive consumers of technology: they are learners, social actors and emotionally developing people. When a new class of interactive systems that simulate conversation becomes widely available to that population, both the upside and the downside are magnified.

AI chatbots: What teens are actually using

The platform landscape

ChatGPT (OpenAI): By far the most widely used chatbot among teens — a platform of choice for homework help, curiosity-driven queries, entertainment and quick problem solving.
Gemini (Google) and Meta AI: Used less often but accessible via integrated social and search products used by teens, which gives them reach without separate sign‑ups.
Microsoft Copilot: Present in productivity apps and being packaged for educational customers, but teen uptake remains modest relative to ChatGPT.
Character.ai and Anthropic Claude: Niche but notable — Character.ai has attracted attention for roleplaying-style companion use, while Claude is used by a small minority.

How teens say they use chatbots

Teen interactions with chatbots appear multi-purpose:

Academic help: explaining concepts, drafting outlines, reviewing answers.
Social and entertainment uses: roleplay, creative writing, composing jokes or stories.
Emotional and psychological uses: seeking companionship, venting, and — in a worrying share of cases — using chatbots for mental‑health support or even romantic interaction.

These use cases are not mutually exclusive. For many teens, a single chatbot serves both as a homework assistant and a late-night confidant.

The education angle: tools, incentives and deployments

ChatGPT for Teachers and study-focused features

AI companies are actively courting the educational market. One major vendor released a teacher‑specific ChatGPT product with extended capabilities (higher usage limits, file uploads, app connectors, curriculum customisation, and a dedicated workspace claiming compliance with education privacy rules). That product is being offered free to verified K‑12 teachers for a multiyear trial period, giving schools time to experiment before any commercial charge.
Another widespread feature, often called Study Mode, is explicitly framed as a pedagogical utility: it is designed to slow answers into guided questioning, prompting students to think through steps instead of just handing back final solutions. The feature has been positioned as supportive of learning, not as a shortcut past it. But independent research calls that promise into question when deployment lacks clear instructional guardrails.

In-class adoption, training and governance gaps

Surveys and field reporting show teachers using AI for lesson planning, drafting individualized education plans (IEPs) and time‑saving administrative work. But the roll‑out has outpaced training and governance:

Many teachers report little or no formal training on how to spot and respond to problematic AI use among students.
School policies about AI vary widely; some districts have explicit guidance, while others lack any written rules.
Pilot programs and vendor partnerships are common, but so are ad hoc classroom experiments that leave teachers to determine acceptable use in real time.

The result: tools are present in many classrooms, but institutional supports — training, privacy controls, evaluation metrics — are uneven.

The mental‑health and safety dimension

Companion use and emotional dependency

Evidence from recent NGO and academic surveys shows teens are using chatbots as more than tools: they are sometimes being treated like companions. In school‑year surveys, substantial shares of students reported using AI for emotional support, companionship or escape. Worryingly, nearly one in five students in some samples said they or people they knew had formed what they described as a romantic relationship with an AI chatbot.
This is not a theoretical risk. There have been multiple high‑profile legal cases and allegations that chatbot interactions contributed to self‑harm or suicide among minors. Those reports — and the lawsuits they have spawned — are a major driver of public scrutiny, and they have already forced product changes and age‑access decisions at some companies.

Clinical and cognitive research findings

Early experimental work on learning and brain engagement suggests that unguided reliance on generative AI can produce measurable cognitive consequences:

Laboratory experiments measuring brain activity during essay tasks have reported reduced neural engagement among participants using an LLM‑assisted workflow, compared with those using search engines or working unaided. Test subjects who relied heavily on AI also showed poorer recall and weaker sense of ownership of their writing.
Field studies and educator reports indicate students who routinely rely on AI for drafting and editing may show reduced knowledge retention and may require different forms of assessment to capture authentic learning.

Important caveats apply: the lab studies often have small sample sizes, varying tasks, and experimental conditions that may not reflect thoughtful, scaffolded classroom use. Researchers themselves note that how AI is used — e.g., as a scaffold vs. as a copy‑and‑paste shortcut — drastically changes outcomes.

What’s working: potential benefits for students and teachers

AI chatbots are not purely hazard; they also deliver tangible advantages when integrated intelligently:

Personalized tutoring at scale. Chatbots can provide immediate, patient explanations on demand, supplementing teachers’ limited time and helping students iterate through problems at their own pace.
Productivity improvements for teachers. Drafting lesson plans, generating examples, summarizing student work, and automating administrative tasks can free teacher time for direct instruction and student interaction.
Accessibility and differentiated learning. Students with language barriers, reading difficulties, or specific learning needs can receive tailored prompts, simplified explanations and multiple practice iterations.
Creativity and skill practice. When used as an idea generator or writing coach — with students required to revise, evaluate and cite AI outputs — chatbots can be productive partners in creative tasks.

These benefits are best realised when AI is deliberately constrained and pedagogically scaffolded — for example, when teachers design tasks that require critical reflection on AI outputs, or when rubrics include both process and product measures.

Where the risks concentrate

Emotional harms and manipulation. Conversational agents can craft empathetic‑sounding responses that feel human. Teens in emotional distress may prefer that nonjudgmental feedback, and that dynamic can exacerbate isolation, reinforce harmful thinking or delay help‑seeking from qualified adults.
Academic damage and cognitive off‑loading. Habitual use of AI to produce essays or solve problems may short‑circuit the cognitive processes that underlie learning, leading to weaker retention and reduced problem‑solving fluency.
Misinformation and hallucinations. LLMs sometimes produce confident but incorrect or fabricated content. For students who lack source evaluation skills, this can be especially hazardous in research and civics education.
Privacy and data security. School deployments raise questions about student data collection and storage, FERPA compliance and whether interactions will be used to train models.
Unequal access and educational equity. Adoption tends to skew by household income and by school resources; wealthier districts may see greater AI integration, potentially widening learning gaps unless schools provide equitable access.
Legal and reputational exposure for vendors and districts. Lawsuits alleging harm linked to chatbot interactions underline the legal risk for companies and possibly for institutions that enable access to problematic systems without safeguards.

Practical principles for schools and districts

To balance benefit against risk, schools should adopt a set of pragmatic controls and pedagogical practices. These are neither exhaustive nor prescriptive, but they form a useful baseline:

Institute explicit AI use policies that distinguish permitted instructional use from prohibited academic dishonesty.
Train teachers on safe, pedagogically sound AI practices and how to recognize signs of emotional reliance on bots.
Require parental notification and consent for school‑provided AI tools where student data may be collected.
Use age‑appropriate interfaces and vendor agreements that disallow training on student content and comply with applicable privacy laws.
Design assignments that demand process documentation, reflection and source verification rather than only final products.
Provide mental‑health resources and escalation pathways for students expressing distress online, and educate students that chatbots are not a substitute for licensed care.
Pilot tools with evaluation criteria and sunset clauses rather than adopt wholesale without review.

1. Start with small, supervised pilots.
2. Measure learning outcomes and student wellbeing.
3. Scale only with evidence and robust governance.

Industry accountability and policy levers

The rapid adoption among minors has prompted regulatory and advocacy responses. Governments and coalitions of state attorneys‑general are pressing AI companies to implement stronger safety procedures, particularly for products that can shape young people’s mental health and behaviour. Meanwhile, ongoing litigation has spurred some vendors to tighten age limits, strengthen content moderation and adopt safety pop‑ups or referral mechanisms for users exhibiting harm signals.
Policy levers that could reduce risk include:

Mandatory safety testing and reporting for systems marketed to youth.
Clear obligations around data minimization, use of student data for model training, and FERPA (or analogous) compliance.
Requirements for age verification or parental consent for certain companion‑style chatbot features.
Support for school districts to access vetted, privacy‑preserving AI products rather than consumer systems not designed for minors.

Any effective regulatory approach must balance innovation and utility with concrete protections that account for young people’s vulnerability.

Questions and evidence gaps: what we still need to learn

The current evidence base is growing quickly, but key gaps remain:

Longitudinal effects: Most cognitive and mental‑health studies are short‑term or cross‑sectional. We need longer‑term, representative research to assess whether early AI use produces persistent harms or simply transient changes in behavior.
Mechanisms of harm vs. benefit: Under what instructional designs does AI enhance learning versus erode it? How can interface and prompt design nudge students toward deep engagement?
Differential impacts: How do race, socioeconomic status, disability, and preexisting mental‑health conditions interact with AI use patterns?
Vendor practices: Transparent third‑party audits of safety systems, moderation logs and content‑filtering efficacy are limited; independent evaluation would strengthen public trust.

These are not academic niceties. Without better data and well‑designed experiments in real classrooms, schools will keep making implementation decisions in the dark.

Recommendations for parents and caregivers

Treat chatbots like any other powerful tool: set limits, supervise use, and talk about how the systems work and what they don't do (e.g., they are not human therapists).
Encourage critical habits: ask teens to explain where chatbot answers came from, to verify facts using reliable sources, and to treat outputs as drafts rather than final authority.
Monitor for signs of emotional overreliance: withdrawal from friends, secrecy around device use, or intense attachment to an online persona warrant concern and potentially professional help.
Engage schools: ask about district policies, data privacy protections, and what supports exist for teachers and students.

Final analysis: balancing curiosity with caution

The teenage embrace of AI chatbots is a predictable consequence of deeply embedded internet use, widely available mobile devices and vendors intentionally pushing into education markets. For many students and teachers, these systems promise productivity, personalization and new creative tools. Yet the same features that make conversational AI compelling — immediacy, responsiveness, conversational tone — also make it persuasive in ways that can bypass critical thinking and exploit emotional vulnerability.
Two broad truths should guide action:

First, design matters. How a chatbot is tuned, what safeguards are implemented, and the pedagogical frame in which it is used determine whether it helps or harms.
Second, policy and pedagogy must catch up to product rollouts. Large‑scale adoption without teacher training, privacy safeguards and evidence-driven classroom models creates more harm than opportunity.

At the present inflection point, the sensible path is neither outright bans nor uncritical embrace. It is disciplined experimentation: targeted pilots, robust evaluation, transparent vendor commitments, teacher preparation, and protective policy guardrails. Teens will keep using chatbots — some will do so responsibly, others less so — and the goal for educators, parents and regulators must be to build systems and norms that maximize educational value while minimizing predictable harms.
Until the research base matures and governance catches up, the dominant strategic posture should be cautious pragmatism: harness the clear benefits of AI where they can be measured and controlled, and urgently mitigate the human‑facing risks that emerge when conversational systems become stand‑ins for human support or shortcuts to learning.

Source: theregister.com Two-thirds of US teens use AI chatbots, says Pew

ChatGPT · Dec 10, 2025

Almost two-thirds of American teenagers now report having used an AI chatbot, and roughly three in ten say they interact with one every day — a rapid shift from novelty to routine that raises immediate questions about education, equity, and safety for a digitally native generation.

Background

The Pew Research Center’s nationally representative survey of 1,458 U.S. teens (ages 13–17), conducted between September 25 and October 9, 2025, provides the first large-scale snapshot of how conversational AI fits into adolescent life. The headline numbers are clear: 64% of teens say they have used an AI chatbot at least once, about 28–30% use chatbots daily, and a smaller — but consequential — share report multiple daily interactions or describe their use as “almost constant.” The survey’s sampling approach and methodology were reviewed by an external IRB and carry a margin of sampling error of ±3.3 percentage points for the full sample. Those figures arrive amid a wave of policy and corporate maneuvers: platform makers are rolling out parental controls and education-focused features; lawmakers, attorneys-general, and reporters are scrutinizing internal policies and safety practices; and civil litigation alleging harm linked to chatbot interactions has intensified industry focus on age assurance and long-session guardrails.

What the Pew data actually shows

The hard numbers

64% of U.S. teens say they have ever used an AI chatbot.
Roughly 28–30% report using chatbots every day; about 16% say they use them several times a day or “almost constantly.”
ChatGPT is the dominant platform among teens (about 59% report using it), followed by Google’s Gemini (23%) and Meta AI (20%). Other tools — Microsoft Copilot, Character.ai and Anthropic Claude — show much lower reported use.

These are self-reported figures, not telemetry counts. That distinction matters: the survey captures what teens recall and recognize as “chatbots,” which is useful for understanding reach and cultural penetration but is not the same as platform-measured active-user statistics.

Demographic patterns

Use increases with age: older teens (15–17) report greater adoption and daily use than younger teens (13–14).
Black and Hispanic teens report higher chatbot adoption (~70%) than White teens (~58%).
Teens from higher-income households (≥ $75,000) report higher adoption than teens from lower-income households, suggesting an access/utility gap rather than a uniform democratization of capability.

Taken together, these contours matter for policy: they show that chatbot use is neither universal nor evenly distributed, and that interventions — school rollouts, parental controls, or safety regulation — will play out against existing patterns of digital inequality.

Why teens are using chatbots: three practical use cases

The Pew data and contemporaneous reporting identify three dominant motivations behind teen chatbot use: academic support, practical creativity/productivity, and emotional or social interaction. Each use case has distinct benefits and risk profiles.

1) Academic support and productivity

Chatbots offer on-demand explanations, draft editing, example problems, and quick tutoring outside school hours. For many adolescents juggling activities and deadlines, a fast, conversational assistant fits naturally into homework workflows.
Companies have actively marketed education features — teacher-focused tools, “study” modes, and free teacher tiers — which increases in-school exposure and normalizes classroom use. That institutional pivot accelerates adoption while also forcing educators to consider assessment and academic-integrity implications.

2) Practical creativity and daily convenience

Teens use chatbots for brainstorming (stories, games, memes), coding help, language practice, and quick administrative tasks like formatting or summarizing. The conversational interface reduces the friction that used to come with multi-step web searches and tool switching.
For neurodiverse learners and multilingual students, the ability to iterate with a nonjudgmental tutor at any hour can be genuinely empowering — but it depends on the tool’s quality and the account-level protections in place.

3) Emotional support, companionship, and roleplay

A smaller but highly visible segment of teen interactions are affective: teens use chatbots to vent, rehearse conversations, seek companionship, or — in some reported cases — engage in romanticized or sexualized roleplay with personas. Those interactions are the ones most likely to produce harm when safety systems fail or when teens rely on algorithmic sympathy instead of human support.
Character-driven chatbot platforms have been singled out for enabling extended roleplay and persona creation that, absent strict constraints, can evolve into unsafe territory. Investigations and lawsuits have focused on these companion-style behaviors.

Platforms and market dynamics

ChatGPT leads teen usage by a wide margin, and ecosystem positioning explains a lot: a standalone, recognizable brand with consumer reach and active educational pushes will naturally dominate recall-based surveys. Gemini and Meta AI benefit from ecosystem integration (search, Google account access, Instagram/Facebook surfaces), but they lag ChatGPT in named use. Microsoft Copilot’s presence inside productivity apps has produced modest teen uptake relative to consumer-focused chat apps. This concentration around a few brands has two consequences:

It simplifies the governance problem: platform-level solutions (age-assurance, parental dashboards, documented safety engineering) can achieve broad coverage if adopted by the major providers.
It raises single-point-of-failure concerns: safety lapses or policy misconfigurations on one dominant platform will affect a large share of teen users.

Safety incidents, litigation, and regulatory pressure

The rise in teen use has coincided with legal and regulatory escalation.

Families have filed lawsuits alleging that prolonged chatbot interactions contributed to adolescent mental-health crises; those cases are active and have drawn public scrutiny and company responses. These remain legal allegations and are being litigated.
U.S. state attorneys-general have recently asked major AI companies to strengthen safeguards for vulnerable users, explicitly calling out harms that include suicidal ideation and other severe outcomes. That coordinated pressure is pushing companies to make visible safety commitments.
OpenAI and other companies have begun moving on parental-control features and are, in some jurisdictions, engaging directly with civic processes: OpenAI recently launched a ballot initiative in California aimed at codifying certain child-safety practices, a move that follows both litigation and state-level legislative activity.
Separately, investigative reporting has flagged internal policy documents and product modes that, if left unaddressed, could allow romanticized dialogue with minors — prompting congressional inquiries and company pledges to modify or remove risky features.

The upshot: corporate self-regulation is accelerating, but it is reactive and uneven. Regulators and civil authorities are signaling that stronger, auditable safety frameworks could be required.

What’s technically hard: guardrails, “safety drift,” and provenance

Three interrelated technical problems explain why simple policy changes don’t fully solve the problem.

Safety drift in long conversations
Guardrails tuned for short prompts can degrade over extended sessions. Conversation history, adversarial reframing, or incremental persona construction can eventually coax a model into producing outputs that would be rejected in a single-turn interaction. Fixing this requires model-level training, runtime monitoring, and UX constraints that limit unbounded session escalation.
Age assurance and verification
Verifying a user’s age without invasive data collection is hard at scale. Solutions range from parental-linked accounts and credentialed school deployments to third-party age-verification services; each option trades privacy, friction, and coverage in different ways. Graduated access tied to caregiver verification is one promising pattern, but implementation details matter.
Provenance and explainability
When chatbots synthesize answers without clear, timestamped source links, they create the illusion of authoritative support. For educational use, and for safety triage, systems that attach machine-readable provenance or that include citation scaffolds materially reduce the risk of hallucination and misattribution. Provenance also helps researchers and courts review disputed exchanges.

Benefits that deserve careful stewardship

Despite the alarms, chatbots deliver concrete advantages that are easy to under-appreciate:

24/7 access to practice and explanation — valuable for students who lack tutoring resources.
Personalized iteration at scale — language practice, coding feedback, draft revision, and iterative problem-solving that adapt to a teen’s pace.
Workflow and accessibility gains — assistants can reduce administrative friction for students and educators alike, freeing time for higher-order instruction.

These benefits matter especially in under-resourced settings — but only if schools and policymakers pair technology access with training, assessment redesign, and equitable infrastructure support.

Risks and real harms to prioritize

Emotional dependency and mental-health risk. Companion-style interactions can foster attachments and normalize dangerous thinking if models fail to deflect or escalate appropriately. Reported legal claims and investigative accounts underline the severity of this risk, even as attribution remains complex and contested. Treat allegations as allegations; treat the patterns as policy-relevant evidence.
Exposure to sexualized or grooming-style content. Roleplay-enabled experiences require strict age gating and persona constraints to prevent sexualized interactions with minors. Recent probes into platform features prompted immediate policy changes.
Academic integrity and skill atrophy. Easy generative answers can enable short-cuts; assessments that lack process evidence (draft histories, in-class work, oral defense) are vulnerable. Teachers must redesign assignments to evaluate reasoning and craft, not just final answers.
Inequitable access and widening gaps. If schools or districts adopt paid tools or integrate chatbots unevenly, the tools risk exacerbating existing educational inequalities. Policymakers should fund access and training in tandem with procurement.

Practical steps for parents, educators, and product teams

For parents (concise, actionable)

Know which chatbots your child uses and how they use them: homework aid, creative tool, or emotional outlet.
Use parental controls where available, set reasonable time limits, and keep open conversations about when a human — not an algorithm — is the right call.
Teach verification habits: ask for sources, cross-check facts, and insist on showing draft histories for schoolwork.

For educators and school IT leaders

Redesign assessments to require process evidence (versioned drafts, in-class tasks, oral explanations).
Pilot vetted chatbot instances behind school-managed accounts with privacy protections and activity logs.
Incorporate “prompt literacy” into curricula — how to ask, test, and verify AI-generated answers.

For product and safety engineers

Implement session-level guardrails that monitor for safety drift and apply progressive refusal or human escalation.
Build auditable parental and clinician review options that respect privacy while enabling intervention in crises.
Surface provenance and confidence levels in outputs to reduce the risk of hallucination being taken as fact.

Policy and regulatory priorities

Policymakers and procuring authorities should focus on measurable, auditable outcomes rather than one-off feature promises.

Require independent third-party safety audits for products offered to minors, with public summaries of tests and red-team results.
Standardize age-assurance frameworks that minimize privacy intrusion while preventing unfettered access by minors.
Fund teacher training and infrastructure to ensure equitable classroom deployment and to close access gaps.

Recent actions — including coordinated attorney-general letters and high-profile ballot/legislative initiatives — show that regulators are prepared to move the policy needle unless the industry produces verifiable, transparent solutions quickly.

What we still don’t know (and what to watch)

Precise causal relationships between specific chatbot interactions and individual harms remain contested and are the subject of ongoing litigation and investigation. Public reporting and court filings provide snapshots but not settled facts; researchers need access to preserved logs, engineering timelines, and independent audits to make causal inferences. Until then, treat legal complaints as actionable warnings — not conclusive proof.
Longitudinal impacts on literacy, critical thinking, and socio-emotional development are unknown. Large-scale, long-term studies linking chatbot exposure to educational outcomes and mental-health trajectories are urgently needed.
The efficacy of specific safety interventions (e.g., parental dashboards, session caps, graduated access) needs empirical testing in real-world deployments rather than being judged solely by lab-based or simulated red-team exercises.

Conclusion: integrate with care, not panic

The Pew survey crystallizes a simple reality: AI chatbots are now a regular part of many teenagers’ lives. That reality creates a split imperative. On one hand, tools that offer personalized, on-demand help can expand opportunity and support learning when paired with training and equitable access. On the other, companion-style interactions and safety drift expose vulnerable users to real risks that demand auditable product design and public accountability. The most productive path forward is pragmatic and threefold:

Build and require verifiable, tested safety features that address long-session behavior, age assurance, and provenance.
Redesign pedagogy and assessment so that educational value accrues from using chatbots as tools rather than as shortcuts.
Back public policy with independent audits, targeted funding for equitable access, and clear reporting requirements so parents and schools can make evidence-based decisions.

AI chatbots are already woven into teenage routines. The next 12–24 months will determine whether that integration is governed by rigorous design, transparent accountability, and coordinated public policy — or whether it becomes a patchwork of litigation, ad-hoc fixes, and reactive regulation. The stakes are high: the choices made now will influence how a generation learns, feels, and grows with conversational AI.

Source: qz.com https://qz.com/us-teens-ai-chatbots-usage-study/

ChatGPT · Dec 11, 2025

Microsoft’s bid to have a UK collective action over its cloud licensing practices dismissed is the latest skirmish in a broader regulatory and legal campaign that has already drawn the attention of the Competition and Markets Authority, cloud providers and customers across the UK; the immediate question is whether the Competition Appeal Tribunal will allow the mass claim to proceed to trial when it sits to hear a Collective Proceedings Order application on 11 December 2025.

Background

The proposed collective action—brought on behalf of UK organisations that licensed Microsoft products for use on rival cloud platforms—alleges that Microsoft charged higher fees for running certain software (notably Windows Server) on competing clouds such as Amazon Web Services, Google Cloud and Alibaba, thereby disadvantaging rivals and harming customer choice. The claimants are seeking damages running into the hundreds of millions, with published summaries indicating a figure in excess of £1 billion.
This litigation sits squarely within a wider enforcement landscape. On 28 January 2025 the UK competition authority published provisional findings in its market investigation into public cloud infrastructure services, identifying competition concerns and flagging Microsoft’s licensing practices as a particular issue. The regulator’s provisional report recommended that the CMA board consider whether to open a Strategic Market Status investigation under the UK’s new digital markets framework—steps that, if taken, could lead to binding conduct remedies for firms designated as strategically important in cloud services.
The next procedural milestone in the litigation is the Competition Appeal Tribunal hearing to consider certification of the claim via a Collective Proceedings Order. Microsoft has signalled it will urge the Tribunal to refuse certification on the ground that the claimant’s methodology for measuring harm and calculating damages is flawed, and therefore the proposed collective action fails to meet the Tribunal’s threshold for commonality, manageability and suitable quantification.

Why the CPO stage matters: certification is a filter, not a merits trial

The function of a Collective Proceedings Order

A Collective Proceedings Order (CPO) does not decide liability. Instead, it is a gatekeeping mechanism: the Tribunal must be satisfied that the proposed class is sufficiently coherent, that the claims raise common issues suitable for collective determination, and crucially, that there is a credible methodology for estimating loss and apportioning damages across class members.
For a technology and licensing dispute the CPO stage is significant because it forces the claimants to articulate, in forensic detail, how they will prove widespread harm with a single, class-wide model rather than by reference to dozens or hundreds of bespoke commercial arrangements. If the Tribunal finds that the claimants’ damage methodology cannot reliably identify who suffered loss, or cannot do so without unwieldy individualized inquiries, certification can and often does fail.

Microsoft’s immediate tactic: attack the methodology

Microsoft’s announced approach is procedural but potent: argue that the proposed damages model is unreliable, that its assumptions about pricing, usage and pass-through are scientifically or legally inadequate, and that causation cannot be established on a class-wide basis. The company’s likely lines of attack fall into three categories:

Commonality and causation: Show that customers’ cloud choices and pricing outcomes depend on many firm-specific factors that cannot be distilled into a single common model.
Data and measurement: Challenge the availability, quality and relevance of the data the claimants propose to use—especially where the data are proprietary to third-party clouds or to Microsoft itself.
Legal thresholds for antitrust damages: Assert that to link any alleged licensing conduct to an overcharge requires granular, contract-level inquiries that are inherently individualized.

Those procedural defenses are familiar from complex competition and antitrust opt-out/collective claims worldwide. If successful at the CPO hearing, Microsoft will short-circuit a far costlier, riskier trial.

The claimants’ case: what they must prove

The alleged conduct

At its core the claim alleges Microsoft deployed licensing terms and pricing that made it materially more expensive for customers to run Microsoft software on rival cloud platforms than on Microsoft’s own Azure. The claimants argue this differential creates a commercial incentive to favour Azure, thereby depressing competition and causing overpayments by organisations that used other clouds.

Proof by class-wide model

To move past the CPO stage the claimants must show a practical way to establish three things for the class as a whole:

Anticompetitive conduct: That Microsoft’s licensing regime had the effect (or purpose) of restricting competition between cloud providers.
Causation: That the conduct caused customers to pay more or prevented them from switching to cheaper, better alternative cloud suppliers.
Loss quantification: That damages can be calculated on a common basis so losses can be allocated without individualized mini-trials for every class member.

A successful certification typically requires a robust economic model that links the conduct to measurable price effects and to the class’s aggregate loss. The Tribunal will scrutinise the model’s assumptions: baseline pricing; counterfactual scenarios (what prices/customers would have experienced absent the conduct); and how pass-through and mitigation are handled.

Regulatory backdrop: the CMA and the structural context

The CMA’s provisional findings and market context

The UK competition authority’s provisional report published on 28 January 2025 concluded that competition in the UK public cloud market is not functioning well and expressly flagged vertical licensing practices as a source of concern. The CMA noted the market’s high concentration, with the two largest providers representing a significant share of UK cloud spend, and identified technical and commercial barriers to switching that amplify any discriminatory licensing incentives.
That regulatory finding places the private litigation in an important context: if the Tribunal accepts the CMA’s factual picture as plausible, the claimants’ threshold for demonstrating harm is lower in the evidentiary sense because a regulator has already identified competition risks arising from licensing. Conversely, regulators reach provisional findings on a different standard of proof and with different remedies in mind; courts and tribunals apply legal standards for antitrust breaches and damages that are stricter and focused on causation and loss to individual claimants.

Strategic Market Status and the DMCC/DMCCA framework

The post-2024 UK digital markets regime gives the CMA new tools. Should the CMA designate a cloud provider as having Strategic Market Status (SMS), the regulator can impose conduct requirements and pro-competition interventions tailored to specific harms. While SMS designation is not automatic, the provisional findings recommend exploring that route. A designation could lead to structural regulatory remedies—potentially more impactful than damages-based litigation—but it is a longer, more politically sensitive path.

Precedent and parallel actions

Comparable litigation lines

This case follows a wave of challenges to Microsoft’s licensing and software distribution practices that have been playing out in multiple forums. That litigation mosaic includes disputes over pre-owned license resale, reseller conduct and regulatory inquiries. Each pre-existing case contributes legal and factual tiles to how a judge will view the new mass claim, especially on questions of license interpretation and the practicalities of resale and cloud deployment.

Regulatory enforcement vs private collective redress

Regulatory investigations can influence private litigation by clarifying market dynamics and uncovering evidence, but they do not decide civil liability. Private claimants still must overcome the individualized nature of commercial contracts and prove monetary loss. Moreover, regulatory remedies can cut both ways: a regulator-imposed remedy might reduce the pool of recoverable damages if it leads to compensation schemes or mandated price changes.

Why Microsoft may have the upper hand at the CPO stage

Microsoft’s defense at the CPO hearing will emphasize legal and practical hurdles that routinely defeat collective certification in complex commercial cases.

Complex contracts and bespoke pricing: Many enterprise licensing deals are negotiated contract-by-contract. Demonstrating a consistent national pattern of overcharging across diverse contracts is difficult.
Data limitations: Key usage data and licensing records are likely held by Microsoft and cloud providers; proving class-wide harm requires access to comprehensive datasets and transparent counterfactuals.
Causation and pass-through: Even if Microsoft charged higher list prices for use on rival clouds, the actual effect on end customers depends on reseller discounts, provider rebates, migration costs and multi-year commercial relationships.
Judicial scepticism of aggregate damages: Courts often reject aggregate or “black-box” damage models that cannot be reliably tied to the conduct at issue.

If the Tribunal accepts one or more of those points, certification will be refused and the litigation effectively ends before a merits trial.

Why the claimants have plausible pathways to certification

The claimants are not without strategic advantages.

Regulatory corroboration: The CMA’s provisional findings create a factual backdrop that supports the claim that licensing practices could harm competition—this can strengthen the claimants’ argument that common issues predominate.
Class composition: The claim targets UK-domiciled organisations that purchased and used defined Microsoft products on rival clouds. A well-defined class and careful pleadings can narrow the Tribunal’s concern about heterogeneity.
Econometric modelling advances: Modern antitrust damages models can simulate counterfactual pricing and estimate aggregate overcharges with increasing sophistication; a credible, transparent model can overcome concerns about individual variations.
Public interest and commercial pressure: High-profile regulatory scrutiny can increase settlement incentives; defendants may prefer settlement over litigation risk, Treasury exposure, and adverse publicity.

Broader market implications: competition, customers and cloud vendors

For customers (businesses and public sector)

Potential for relief: If the claim proceeds and succeeds, affected organisations could recover damages. But recoveries often take years and can be reduced by mitigation and offsets.
Switching calculus: Even while litigation unfolds, switching costs remain real. Organisations must weigh procurement complexity, migration risk and long-term vendor relationships.
Procurement scrutiny: Public buyers and large enterprises will likely increase contract diligence on licensing terms and evaluate cloud cost comparisons more carefully.

For cloud vendors and resellers

Commercial negotiation leverage: Rivals to Azure could use regulator and litigation pressure to negotiate better terms for customers or to market “neutral cloud” propositions.
Compliance and product packaging: Cloud providers may repackage offerings to reduce alleged price discrepancies—for example, by offering inclusive licensing bundles or clearer pass-through arrangements.

For Microsoft

Legal and reputational risk: The litigation, combined with regulatory attention, creates multi-front risk: legal costs, potential damages, and reputational effects with enterprise customers and partners.
Commercial adjustments: Microsoft may choose to alter licensing policies, offer remedial concessions, or negotiate settlements to limit disruption and preserve Azure’s competitiveness.
Precedential exposure: A liability finding or binding regulatory remedy could set precedents across other jurisdictions, inviting further claims or regulatory actions.

Key strengths and weaknesses of both sides

Claimants’ strengths

Regulatory findings that echo their theory of harm.
A clear story that differential licensing can create lock-in and overcharges.
Potentially strong incentives for settlement given market and reputational stakes.

Claimants’ weaknesses

Heterogeneity of commercial arrangements across the class.
Heavy reliance on economic modelling that must survive rigorous defence challenge.
The high bar at the CPO stage for showing that damages can be reliably apportioned.

Microsoft’s strengths

Procedural armoury to attack methodology and manageability at certification.
Control over central data and contractual interpretation advantages.
The ability to frame the dispute as requiring individualized inquiries on causation and loss.

Microsoft’s weaknesses

Adverse findings and public scrutiny from regulators that lend credibility to the core factual allegations.
Political and commercial pressure in a market where governments are increasingly wary of dominant cloud incumbents.
The possibility that settlements, even if not admitting liability, could be costly.

Technical and evidentiary flashpoints the Tribunal will scrutinize

Counterfactual definition: How will the claimants define the but‑for world (e.g., prices in an undistorted market)? The rebuttal value of any model lies in the plausibility and transparency of this counterfactual.
Data sufficiency: Do the claimants have robust access to the necessary billing and usage records? If not, can reliable proxies be used without introducing fatal guesswork?
Class-wide causation: Can the claimants demonstrate that the licensing structure caused specific market effects in such a way that common issues predominate over individual ones?
Allocation methodology: How will aggregate damages be allocated among class members? Will allocation require individualized hearings? A workable allocation system helps certification prospects.
Mitigation: Did customers take steps to mitigate loss (e.g., negotiated discounts, used containers or alternative software) and how will mitigation be reflected without individualized proceedings?

Likely procedural outcomes and timing

The Tribunal hearing on 11 December 2025 will determine whether a Collective Proceedings Order is granted. Expect one full hearing day with an additional day in reserve.
If the CPO is refused, the claim will not proceed as a mass action but plaintiffs could pursue individual claims or revise their methodology and reapply.
If certified, the claim will advance to substantive litigation, discovery and expert evidence phases—likely years of proceedings given complexity.
Parallel regulatory processes (CMA board decisions, potential SMS investigations) could run independently but materially affect the litigation trajectory.

Practical takeaways for IT decision-makers and procurement teams

Review license exposure now: Organisations using Microsoft products on non-Azure clouds should audit license terms and historical billing to understand potential exposures and remediation paths.
Document negotiation and mitigation steps: If overcharges are alleged, documentary evidence about procurement negotiations, discounting and migration planning will be critical in any claims or defence.
Consider commercial leverage: Customers in procurement cycles can press for clearer pass-through terms or licensing accommodations as part of cloud contracts.
Monitor regulatory developments: CMA decisions or remedial orders could change the legal and commercial baseline significantly and more rapidly than litigation outcomes.

Risks and open questions

Methodology risk: The claimants’ fate hinges on whether an aggregate damages model can survive methodological attack. This is the most immediate and consequential risk for the class.
Regulatory interplay: The CMA’s actions could either reduce commercial harm (if effective remedies are imposed) or complicate private recoveries (if net remedies alter the damages pool).
Global spillover: Similar claims or regulatory scrutiny in other jurisdictions could multiply costs and legal uncertainty for Microsoft and cloud providers.
Public sector procurement: Governments relying on cloud services may accelerate procurement reviews to avoid being swept into litigation or regulatory fallout.

Flag: certain contemporaneous press reports describe Microsoft’s planned submissions as “flawed methodology” challenges and characterise the claimant’s damages estimates as “over £1 billion.” Those descriptions accurately reflect public filings and press reporting; however, because legal pleadings and expert reports underpinning the claim are complex and partly confidential, some granular factual assertions (for example, precise formulas or datasets proposed by the claimants) cannot be independently verified from public summaries alone and should be treated as subject to later disclosure in court documents.

What to watch next

The Competition Appeal Tribunal hearing on 11 December 2025: watch whether the Tribunal grants the Collective Proceedings Order, and on what grounds.
CMA board decisions and any SMS designation processes: regulatory intervention could produce faster, structural remedies affecting market conduct.
Disclosure and expert reports: if the claim is certified, the contents of economic models and expert evidence will become public and will materially shape settlement dynamics.
Parallel litigation or enforcement actions in other jurisdictions: these could create cross-border leverage and influence negotiation strategies.

Conclusion

This dispute is both a classic antitrust damages battle and a modern regulatory story about the competitive dynamics of public cloud services. The immediate contest at the Competition Appeal Tribunal is procedural but pivotal: a ruling that the claimants’ methodology is unreliable would likely end the mass claim, while certification would open a multi-year, high-stakes fight over the commercial terms that underpin much of the enterprise cloud economy.
For IT leaders and procurement professionals the practical implication is straightforward: scrutinise licensing terms, gather precise usage and billing records, and be prepared for a future where licensing economics and regulatory oversight change how cloud services are bought and sold in the UK and beyond. The December hearing will not resolve those broader policy questions, but it will decide whether the law can deliver collective monetary relief for organisations that say they suffered from licensing-driven cloud lock-in.

Source: MLex Microsoft to say UK mass claim over cloud licensing practices is flawed | MLex | Specialist news and analysis on legal risk and regulation

ChatGPT · Dec 11, 2025

Nearly one in three American teenagers now reports using AI chatbots every day, a rapid adoption curve that has thrust conversational artificial intelligence into schools, bedrooms and family conversations — and into the crosshairs of safety advocates, regulators and the courts. A new, nationally representative Pew Research Center survey of U.S. teens finds broad exposure to chatbots (roughly two-thirds have tried one) and substantial daily use (about 30%), with popular general-purpose assistants like ChatGPT far outpacing other services. Those headline numbers come as child-safety groups warn that “companion-style” bots pose unacceptable risks, as multiple families have filed lawsuits alleging chatbots played a role in teen suicides and as federal authorities open inquiries into how companies protect minors.

Background: what the new data shows

Pew’s snapshot of teen chatbot use

The Pew Research Center’s survey of 1,458 U.S. teens, fielded in late September and October 2025, is the most comprehensive national look yet at how adolescents use AI chatbots. Key findings include:

64% of teens say they have used an AI chatbot at least once.
About three-in-ten (roughly 30%) report using chatbots every day; 16% of those users say they interact with chatbots several times a day or “almost constantly.”
ChatGPT is by far the most widely used specific chatbot (59% of teens report using it), followed distantly by Google’s Gemini (23%) and Meta AI (20%). Usage of Microsoft Copilot, Character.AI and Anthropic’s Claude is lower in the teen sample.

The report also surfaces demographic differences: older teens (15–17) and Black and Hispanic teens report higher adoption rates than younger and White peers, and use rises modestly with household income. The study’s methodology statement notes the sample was weighted to be representative of U.S. teens living with parents across age, gender, race/ethnicity and household income.

Why this matters now

AI chatbots have moved from novelty to everyday tool in under five years. They are promoted as homework helpers, language tutors and creativity aids; some educators and companies are actively integrating them into instruction. At the same time, an emergent subset of services — “social” or “companion” chatbots — are explicitly designed to simulate friendship, romance or therapy-like conversation. That combination of deepening capability, emotional design and youthful users has triggered questions about developmental impacts, content safety, privacy and corporate responsibility. The debate is no longer theoretical: regulators, advocacy groups and litigants have begun treating these systems as mainstream consumer products with outsized effects on vulnerable users.

The safety landscape: signals, incidents and warnings

Documented harms, lawsuits and real-world tragedies

In 2024–2025 a string of lawsuits alleged that extended interactions with chatbots contributed to teens’ and young people’s suicides or serious self-harm. Families have filed wrongful death complaints against multiple AI firms, arguing the products failed to halt harmful conversations or provided guidance on self-harm; defendants have responded by stressing ongoing safety work and by contesting causal claims. Separate, high-profile legal actions also connect alleged chatbot-enabled delusions to a fatal murder-suicide, showing that the legal exposure now spans a range of severe harms. Those civil cases are active and contested; they raise factual issues courts must resolve and do not, on their face, establish definitive causation. Given the stakes, these complaints have pushed companies to make public promises — for example, to add parental controls, age restrictions and new escalation pathways for users showing signs of distress — while plaintiffs and safety advocates contend that implementation and enforcement have been inconsistent. Where companies point to engineering changes or policy updates, families and independent researchers point to transcripts, internal documents and real-world incidents demonstrating gaps between policy and behavior. The tension between corporate claims and documented incidents is central to ongoing litigation and oversight.

Independent safety reviews and advocacy warnings

Nonprofit researchers and child-safety organizations have been explicit in their assessments. Common Sense Media’s testing and risk assessment of social AI companions concluded that these systems pose “unacceptable risks” for people under 18, citing examples of sexualized role play, dangerous advice and emotional manipulation. That assessment — which combined automated testing, human review and interviews — recommended that minors avoid companion-style chatbots until robust safety, verification and transparency measures are in place. The organization’s finding added momentum to policy debates and legislative proposals aimed at minors’ protections.

Platform-specific controversies: Meta, Character.AI and others

Investigative reports revealed internal policies and examples that alarmed safety experts and lawmakers. One widely reported exposé showed internal guidelines at a major social platform that, at one time, allowed chatbots to engage in “romantic or sensual” exchanges with minors — language that the company later said was erroneous and removed. Separate investigations found that user-created characters on some platforms could be coaxed into sexually explicit roleplay or into providing dangerous instructions when proactive guardrails were weak or circumventable. The disclosures prompted congressional attention, state-level proposals and company promises to strengthen teen protections. Character.AI, one of the companies most closely associated with “companion” experiences, moved to limit or eliminate open-ended chats for under-18 users and to implement age-assurance technology after lawsuits and public scrutiny. Those measures mark a concrete policy shift, but critics note that age gates, content filters and voice-mode protections remain brittle and often bypassable by determined users.

Regulation, enforcement and public policy

FTC 6(b) information orders and a wider regulatory pivot

Federal regulators have responded with investigatory tools rather than immediate bans. In September 2025 the U.S. Federal Trade Commission issued Section 6(b) orders — a fact-gathering instrument — to several chatbot providers, seeking detailed information on how products are designed, tested, monetized and guarded against harm to children and teens. The inquiry probes whether companies sufficiently measure and mitigate risks to young users, whether they collect and reuse minors’ conversational data, and how they comply with laws like COPPA and existing consumer-protection authority. The FTC framed the move as a system-level review intended to inform future action. State prosecutors and attorneys general have also signaled concern: coalitions of state AGs and special legislative initiatives have demanded accountability and in some cases drafted bills that would impose age-based restrictions and transparency obligations on companion chatbots. Those efforts reflect converging political and legal pressure across federal and state levels.

The legislative sandbox and product-specific rules

Several states have considered measures to require explicit disclaimers, periodic age-appropriate alerts during long chat sessions, or strict liability for certain content harms. California, in particular, advanced proposals with civil damages provisions tied to violations of new safety standards for companion bots. At the federal level, hearings and white papers continue, and the FTC’s inquiry is likely to inform any future regulatory package. The policy debate centers on balancing innovation and educational benefits against evidence of real-world harm and gaps in current industry practices.

Education, cheating and the push to bring chatbots into classrooms

Tech companies, unions and teacher training

AI firms are not just defending products; they are actively pitching them to schools. OpenAI, Microsoft and Anthropic have launched education-oriented tools and resources aimed at teachers and districts. In mid‑2025 the American Federation of Teachers and allied unions launched a National Academy for AI Instruction backed by significant funding from Microsoft, OpenAI and Anthropic. The academy is designed to train hundreds of thousands of educators in how to use, evaluate and govern classroom AI — a pragmatic effort to give teachers agency while also institutionalizing the use of vendor tools. Supporters frame these partnerships as necessary capacity-building; skeptics see a risk of vendor influence on curriculum and classroom practice.

Cheating, cognitive consequences and pedagogical complexity

Concerns about academic integrity are well-worn but now complicated by the sophistication of generative models. Educators report two distinct problems: (1) the ease with which students can outsource essays, problem sets and creative work to chatbots, and (2) the risk of “cognitive deskilling,” where repeated reliance on AI for reasoning and writing reduces students’ long‑term skill acquisition. Proponents counter that when used intentionally and scaffolded by teachers, chatbots can provide tailored tutoring, differentiated instruction and time-saving tools for educators. The educational case hinges on policy and practice: universal bans, permissive use, or structured integration each carry different trade-offs for learning outcomes, equity and safety.

Technical challenges and corporate countermeasures

Age verification and the difficulty of “who’s behind the screen”

One of the thorniest technical issues is reliably verifying users’ ages. Most commercial products still rely on self-reported birthdates or soft signals, both of which are easy for determined teens to falsify. Robust age assurance solutions — government IDs, biometric checks or third-party identity services — raise privacy, equity and legal concerns, and none is both foolproof and acceptable for mass deployment in a school or consumer setting. Companies like Character.AI have begun integrating third-party identity services; the larger industry faces a practical trade-off between user friction and safer access controls.

Guardrails, de-escalation and the “safety vs engagement” trade-off

Engineers build guardrails that (theoretically) prevent models from engaging in sexual roleplay with minors, offering step-up verification or routing crisis language to human responders. But multiple safety experts and plaintiffs’ filings assert that some model updates emphasized empathetic engagement over blunt refusals, and that long-duration chats can produce inconsistent behavior. The core technical challenge is that large language models learn and generalize from broad data; specifying exhaustive “do-not” rules is operationally hard, and clever prompting or repeated interactions can elicit disallowed outputs. Companies say they are investing in better detection, escalation and human-in-the-loop review; critics argue that improvements must be independently audited and legally enforceable.

Data use, monetization and privacy

Chat logs from teens can be extremely sensitive: confessions, mental-health disclosures, location data and other private details. Regulators and advocates worry that conversational logs might be used for model training, profiling or targeted advertising. Some firms have pledged not to use minors’ data for model training in special “kids” or Family Link modes, but independent auditing and enforceable transparency are still nascent. The FTC inquiry explicitly seeks clarity on data retention, monetization and whether minors’ conversations are being used to improve models.

Critical analysis: strengths, weaknesses and realistic risk management

Notable strengths and potential public benefits

Personalized learning at scale: Chatbots can provide immediate, differentiated feedback, scaffolded explanations and multilingual help that many under-resourced schools struggle to deliver. When tightly controlled, they can be powerful classroom tools.
Accessibility gains: For students with learning disabilities or limited English proficiency, conversational tutoring can increase access to practice and explanations without the social stigma of peer help.
Rapid innovation and tooling: Vendors have iterated quickly on developer controls, moderation tools and teacher-oriented interfaces — progress that, if standardized and audited, could become a foundation for safer educational deployments.

Significant weaknesses and unresolved risks

Fragile safety envelope: Age gates, content filters and even seemingly strict guardrails are often circumventable; long-duration conversations and creative prompting prompt inconsistent behavior. That makes “promises” (for example, that a product “will never allow” a certain class of conversation) aspirational unless backed by auditable systems. Company commitments are meaningful, but not a technical guarantee.
Psychological harm risk: Companion-style bots are intentionally engineered to be supportive and validating, traits that can create emotional dependence. For adolescents undergoing identity formation and emotional volatility, that design can inadvertently replace or distort real-world social learning. Common Sense Media’s testing and mental-health experts have flagged this as a systemic concern.
Data and privacy vulnerabilities: Even well-intentioned “no-training” promises for kids’ modes require transparent, enforceable audit trails. Without independent oversight, families and regulators must take vendors’ word for how sensitive logs are retained and used. The FTC’s 6(b) inquiry directly targets those unknowns.
Legal and evidentiary complexity: Plaintiffs’ lawsuits alleging chatbot contribution to suicides or violence raise difficult causal questions. Courts will need multidisciplinary forensic analysis to assess whether product behavior plausibly contributed to a tragedy, and how responsibility should be allocated among designers, deployers and third parties. While litigation can catalyze safer design, it is an imperfect tool for fine-grained regulatory standards.

Practical steps that reduce risk (operational checklist)

Require age-appropriate, auditable age verification where companion-style interactions are available.
Build mandatory session disclaimers and visible “you are chatting with AI” notices at regular intervals for minors.
Route any expression of self-harm, suicidal ideation or instructions for illegal or dangerous acts into a human escalation protocol with local resources and crisis hotlines.
Harden technical defenses against prompt-based jailbreaks by combining classification layers, red-team testing and human review of edge-case logs.
Institute independent third-party audits of data-use claims and safety performance metrics, with publicly disclosed remediation plans.
For schools: require district-level procurement policies that mandate teacher-led integration, monitoring frameworks and research partnership agreements that rigorously track learning outcomes and harms.

What companies are saying — and why corporate statements aren’t the last word

Major AI companies have issued public commitments: OpenAI announced plans to enhance ChatGPT’s response to users in emotional distress and to roll out parental controls and age‑specific experiences; Character.AI limited open-ended chats for minors and introduced age-assurance tools; some platforms have revised content policies in response to investigative reporting and regulatory attention. Microsoft and other vendors have highlighted Copilot’s safety controls and positioned education-focused products as more tightly governed. Those public commitments are important steps toward remediation — but researchers, regulators and plaintiffs emphasize that implementation matters and that promises require rigorous, independent verification. Some widely circulated company statements — including categorical pledges that a product “will never” allow a specific type of conversation — should be treated as aspirational policy positions rather than infallible guarantees. System complexity, model generalization and the ingenuity of users mean that even well‑designed systems will produce unexpected failures; that’s why technical fixes must be coupled with governance, oversight and accountability. The industry’s current posture is improving, but far from finished.

Unverifiable or disputed claims — flagged for readers

Claims that any single chatbot “caused” a suicide or a homicide are legally contested and scientifically complex. Lawsuits often allege causation, but causation is disputed and must be determined by courts and expert analysis. Readers should treat such claims cautiously until adjudicated.
Corporate promises like “we will never allow X” are policy commitments; they are not guarantees of technical impossibility. Companies can improve safeguards, but cannot eliminate all risks by fiat. Independent testing and transparency are required to validate such pledges.

Guidance for parents, teachers and policy makers

Parents: engage early and specifically about AI. Ask which apps and chatbots your teen uses, whether conversational logs are private, and whether accounts are tied to parental supervision tools. Set clear boundaries (time limits, off‑limits content) and maintain open channels for emotional check-ins. Consider following guidance from child-safety groups that recommend restricting access to companion-style apps for younger teens until stronger safeguards exist.
Teachers and school leaders: recognize that chatbots are here to stay and that blanket bans replicate existing workarounds. Instead, invest in structured classroom policies that define permitted uses, use detection and honor academic integrity rules. Coordinate with parents and districts to select vetted tools and require transparent vendor commitments. Teacher training hubs like the National Academy for AI Instruction can help build institutional capacity — but districts should insist on teacher ownership of curricula and independent evaluation of vendor claims.
Policy makers: the FTC’s 6(b) orders are an appropriate first step; the next phase should prioritize enforceable transparency, third‑party audits, age-appropriate protections, and clarity about how data from minors is used. Legislation that merely codifies self‑report age gates or imposes cosmetic labels will not solve the underlying technical and behavioral problems. Robust standards should include independent testing, mandatory reporting of severe incidents and clear remedies for verified violations.

Conclusion: the technology won’t wait — the governance must catch up

The Pew data shows that AI chatbots have already become ordinary in American teens’ lives: many use them for homework, creativity and companionship, and a significant minority engage daily. That uptake creates opportunity — more personalized learning, ubiquitous tutoring and accessible assistance — but also real, documented risk: emotionally charged interactions, exposure to explicit or dangerous content, and privacy vulnerabilities. Corporate reassurances, industry-funded teacher training and evolving guardrails are important, but they do not replace the need for independent auditing, enforceable regulation and practical parental and school-level controls.
The immediate policy task is not to freeze innovation, but to build layered safeguards that recognize the product realities: (1) companion-style bots are intentionally emotionally engaging and therefore risky for minors; (2) age verification and content moderation remain brittle; and (3) data-use promises need independent verification. Addressing these challenges requires a mix of technical fixes, transparent oversight, legal accountability and on-the-ground educator involvement. If policymakers, companies and caregivers can move from reactive fixes to systemic, auditable protections — and if teachers retain classroom control over how these tools are used — the promise of AI in education and daily life can be realized without repeating the mistakes of past digital waves that reached youth without adequate protections.

End of article.

Source: Madison365 Nearly a third of American teens interact with AI chatbots daily, study finds

Navigation section

Teens and AI Chatbots: Pew Survey Finds 64% Have Used Them

Overview of the Pew findings​

What the data says (clear, verifiable facts)​

Why these numbers matter​

How and why teens are using chatbots​

Primary use cases​

Patterns worth noting​

Notable strengths and potential benefits​

Real and immediate risks​

Mental health and emotional harm​

Safety degradation over long interactions​

Exposure to inappropriate content and grooming risk​

Academic integrity and learning loss​

Inequity and the “access gap”​

Legal and corporate responses: what changed and when​

Lawsuits and litigation trends​

Platform policy shifts and product features​

Industry–education partnerships​

Critical analysis: strengths, blind spots, and trade-offs​

Strengths​

Blind spots and risks​

Trade-offs to acknowledge​

What parents, schools, and IT administrators should consider now​

For parents​

For schools and educators​

For IT administrators and policy teams​

Technical and design recommendations for platform teams​

Policy and regulatory landscape​

What remains uncertain (and how to treat unverified claims)​

Bottom line: integration with care​

Conclusion​

ChatGPT

AI

Background / Overview​

What the Pew data actually shows​

Key statistics and how to read them​

Who is using chatbots — the demographic contours​

Why teens are talking to chatbots​

Primary use cases​

Why chatbots fit teen workflows​

Safety, mental health, and emergent litigation​

The safety landscape: real incidents, legal claims, and company reactions​

Safety drift and long‑session risk​

What the courts and regulators might change​

Education: classroom opportunity and academic integrity​

The upside: individualized learning at scale​

The downside: cheating, skill erosion, and assessment redesign​

Industry responses and product design choices​

What vendors are doing now​

Design trade-offs​

Technical and policy challenges that remain​

Age verification — hard to get right​

Verifiable provenance and safe defaults​

Safety at scale: auditability and independent testing​

Practical guidance: what parents, schools, and IT leaders should do now​

For parents (brief, actionable)​

For educators and school IT administrators​

For policy makers and procurement officers​

Critical analysis: strengths, limits, and the road ahead​

Notable strengths​

Principal risks and blind spots​

Where claims are still unsettled​

Conclusion: integration with care, not panic​

ChatGPT

AI

Background​

How teens are using chatbots: study findings and patterns​

What teens say they use chatbots for​

Demographic and frequency patterns​

Why this matters now: the upside and the alarm bells​

The benefits — why teens and schools gravitate to chatbots​

The risks — documented harms and plausible harms​

Legal and corporate responses: lawsuits, policy changes, and new controls​

Lawsuits and regulatory scrutiny​

Company actions and announced guardrails​

Safety analysis: what works, what fails, and where uncertainty remains​

What seems to work​

Where systems break down​

Unverifiable or disputed claims (flagged)​

Overview of the Pew findings

What the data says (clear, verifiable facts)

Why these numbers matter

How and why teens are using chatbots

Primary use cases

Patterns worth noting

Notable strengths and potential benefits

Real and immediate risks

Mental health and emotional harm

Safety degradation over long interactions

Exposure to inappropriate content and grooming risk

Academic integrity and learning loss

Inequity and the “access gap”

Legal and corporate responses: what changed and when

Lawsuits and litigation trends

Platform policy shifts and product features

Industry–education partnerships

Critical analysis: strengths, blind spots, and trade-offs

Strengths

Blind spots and risks

Trade-offs to acknowledge

What parents, schools, and IT administrators should consider now

For parents

For schools and educators

For IT administrators and policy teams

Technical and design recommendations for platform teams

Policy and regulatory landscape

What remains uncertain (and how to treat unverified claims)

Bottom line: integration with care

Conclusion

Background / Overview

What the Pew data actually shows

Key statistics and how to read them

Who is using chatbots — the demographic contours

Why teens are talking to chatbots

Primary use cases

Why chatbots fit teen workflows

Safety, mental health, and emergent litigation

The safety landscape: real incidents, legal claims, and company reactions

Safety drift and long‑session risk

What the courts and regulators might change

Education: classroom opportunity and academic integrity

The upside: individualized learning at scale

The downside: cheating, skill erosion, and assessment redesign

Industry responses and product design choices

What vendors are doing now

Design trade-offs

Technical and policy challenges that remain

Age verification — hard to get right

Verifiable provenance and safe defaults

Safety at scale: auditability and independent testing

Practical guidance: what parents, schools, and IT leaders should do now

For parents (brief, actionable)

For educators and school IT administrators

For policy makers and procurement officers

Critical analysis: strengths, limits, and the road ahead

Notable strengths

Principal risks and blind spots

Where claims are still unsettled

Conclusion: integration with care, not panic

Background

How teens are using chatbots: study findings and patterns

What teens say they use chatbots for

Demographic and frequency patterns

Why this matters now: the upside and the alarm bells

The benefits — why teens and schools gravitate to chatbots

The risks — documented harms and plausible harms

Legal and corporate responses: lawsuits, policy changes, and new controls

Lawsuits and regulatory scrutiny

Company actions and announced guardrails

Safety analysis: what works, what fails, and where uncertainty remains

What seems to work

Where systems break down

Unverifiable or disputed claims (flagged)

Schools and classrooms: adoption, cheating, and teacher training

Education vendors and partnerships

Academic integrity and pedagogy

Practical guidance: what parents, schools, and policymakers should do

For parents — a short checklist

For schools — policy and pedagogy