Generative AI has rapidly emerged as one of the most widely adopted and influential technologies shaping young people’s digital lives. For many young people, AI-powered chatbots have become a routine part of daily life. What this survey makes clear, however, is that teens are not always using these tools in the same ways, for the same purposes, or with the same results. The design choices being made during this early and formative period of adoption, will have lasting consequences for how young people relate to these tools for years to come.
That is both an opportunity and a responsibility. The teenagers using these products today are growing up in a world where the ability to navigate AI effectively, critically, creatively, and safely all at once, will be a foundational skill. Developers, designers, and policymakers are already tackling how to build products that meet teens where they are, and many have been asking for stronger evidence to guide those choices. The following recommendations are grounded in what teens across three countries told us about how they are actually using these tools. They point toward design directions worth pursuing and toward the deeper research that will be needed to fully answer the questions these patterns raise.
1. Design for Growth, Not Just Completion
Key Findings
- Teenagers most often use chatbots to fill gaps in their knowledge, such as reviewing material they didn’t understand or completing assignments that they did not know how to do.
- Most teens have also used chatbots to help structure and support their overall learning, such as making study guides or plans/schedules
- The majority of teens agreed that chatbots help them understand content at a deeper level and improve skills in difficult/challenging areas.
- At least ¾ of teens said that using AI for creative tasks is “fast” and “helpful”
- For most, the final result still felt “authentic”, “original”, and/or “personal”
- However, views on the overall impact of AI on creativity were more divided: Nearly half of American teens (45%) felt less creative when using AI, compared to roughly one-third of Brazilian (33%) and French (37%) teens.
Recommendations
- Scaffold rather than solve: Design default response patterns that guide teens through problems step-by-step rather than immediately delivering complete answers, particularly for academic queries. A “learning mode” that explains reasoning rather than just providing output, is a great feature for this purpose.
- Surface the process, not just the product: Proactively suggest prompts oriented toward brainstorming, outlining, and exploring rather than completing (e.g., instead of “write my essay,” surface “help me plan my essay”).
- Frame AI as a collaborative partner, not a generator: Particularly for creative tasks, emphasize that AI-assisted work is still the user’s work – the idea, the direction, and the judgment are theirs. Shifting from a focus on generation to positioning the AI tool as a collaborator/creative partner could help users feel that they are more involved in the creative process and have more ownership of the final product.
2. Build AI Literacy Into the Product Experience
Key Findings
- Many teenagers view chatbots as “logical”, “accurate”, and/or “unbiased”
- A majority of teens in the US & Brazil (but not France) have received instruction around AI use. However, “understanding bias” was the least common topic addressed
- French teenagers reported the lowest rates of checking information from chatbots against other sources.
- We also found an overall significant association between education on “identifying/verifying AI generated content” and fact-checking
- Around half of US (44%) and Brazilian (50%) teens said they often or almost always
- checked information from chatbots against other sources, compared to only 26% of French teens.
Recommendations
- Embed AI literacy touchpoints throughout the user experience: Don’t rely on onboarding alone. Surface clear, plain-language explanations of how AI works at relevant moments during use, not just at setup.
- Make bias visible: Most teens rate their frequently-used chatbot as unbiased, yet education around AI bias was the least common type of AI literacy instruction reported across all three countries. This gap makes it especially important for products to surface their own potential bias directly. When responses touch on contested, health-related, or identity-related topics, flag that AI outputs may reflect biases in training data.
- Cite sources actively: Where possible, link AI responses to credible, vetted external sources in real time. This both models good information hygiene and gives teens a practical path to verification.
- Encourage verification and source exploration: Move beyond small-print accuracy disclaimers. Build in active nudges that encourage young users to cross-reference information and explore the topic outside of the AI platform for deeper learning opportunities. Rather than saying “AI chatbots can be wrong, you should check other sources,” say “You can learn a lot more if you look at these sources for more information.”
- Build in information confidence indicators: Consider noting if AI-generated content may be uncertain, incomplete, or outdated.
3. Design for Clarity About AI’s Role
Key Findings
- At least ⅔ of teens in the US & France described a frequently-used chatbot as “warm” and/or “non-judgemental”
- Young people were divided on whether this chatbot was more “machinelike” or “humanlike”
- Between ¼ and ⅓ of teens in the US & Brazil who asked AI for personal advice did so frequently.* Most teens (in the US & Brazil) also found this advice to be helpful
- ~⅓ of teens in the US & Brazil also frequently* used AI chatbots to roleplay/practice, express thoughts/feelings, and/or socialize
Recommendations
- Build the boundary into the product, not just the disclaimer: Product behavior – tone, response patterns, and follow-up prompts – should reflect the actual limits of what AI can provide in a support context, and avoid language that positions the AI as an active observer or advisor, such as “what I see happening here” or “here’s what I’d recommend.”
- Provide strategies for human connection: When teens use chatbots for social support, consider in-platform suggestions for how to have similar conversations with friends, family, or counselors – treating the chatbot interaction as a possible on-ramp to human connection rather than an endpoint.
- Actively redirect for high-stakes topics: For mental health, physical health, and crisis-adjacent topics, go beyond a disclaimer. Provide in-product pathways to professional resources and trusted adults, framed as a necessary complement to the chatbot interaction.
4. Make AI Work Better for All Young People
Key Findings
- American boys used chatbots more often than girls (42% vs 30% daily) and rated them as more important in their daily lives (48% vs 28%)
- Boys asked AI for advice significantly more often across all categories
- Boys reported using AI tools significantly more often for a wide range of activities, including those related to learning (“research,” “discover,” “practice”) and creativity (“create,” “brainstorm,” “edit & refine”).
Recommendations
- Audit default personas, voices, and framing for gender bias: Many chatbots which are marketed as assistants default to feminine voices and personas. Offer meaningful customization and avoid designs that reinforce gender stereotypes.
- Diversify the use cases and prompts surfaced to new users: If the most prominently featured use cases skew toward stereotypically male interests or contexts, girls may not see themselves as the intended audience. Onboarding and suggested prompts should reflect a genuinely broad range of interests and needs, such as opportunities for independent learning and creativity focused use cases.
- Build early positive experiences that develop confidence: Given evidence that girls may underestimate their own AI competency, design for low-stakes, confidence-building early interactions rather than immediately showcasing complexity or capability. Success early matters.
Conclusion
AI chatbots are already deeply integrated into teens’ lives. The question is no longer whether young people will use them – it is whether the products they use are designed with their development in mind.
The recommendations in this report are intentionally specific. Surface the process, not just the product. Make bias visible, not just disclaimed. Build clarity about AI’s role into how the product actually behaves, not just into a terms of service footnote. Design for the full range of young people who are using these tools, including those whose interests and needs have not historically shaped product design.
The work ahead is ongoing, and much of it is already in motion. We are grateful to the developers, designers, and policymakers thoughtfully engaging with many of these questions, and the Digital Wellness Lab is committed to continuing this research alongside them. As young people’s relationships with AI evolve, so will our understanding of what supports them well – and continued collaboration will help that understanding keep pace with the tools themselves.
For more information, contact us at dwl@childrens.harvard.edu.