These recommendations translate findings from the Digital Wellness Lab’s recent Pulse Survey on adolescent AI chatbot use into design and policy considerations for product, trust and safety, and policy teams. Findings draw on responses from nearly 3,000 adolescents ages 13–17 in the United States, Brazil, and France.
1. Learning and creativity
Learning is the dominant use case for AI chatbots across all three countries, and a majority of teens report that these tools help them understand content more deeply. At the same time, a substantial share of teen respondents—particularly in the US—report feeling less creative when using AI. How products are designed to support learning and creative work shapes whether teens engage these tools as a shortcut or as a way to build skills.
Finding
Teens in all three countries most often use chatbots to fill knowledge gaps, with a majority agreeing that chatbots help them understand content at a deeper level and improve skills in difficult areas.
Recommendation
Default to step-through responses for academic queries. A “learning mode” that walks through reasoning rather than delivering a final answer supports teens’ ability to form a deeper understanding of the topic or skill.
Finding
Most teens in the US and Brazil have used chatbots to structure their learning, including making study guides and plans or schedules, while only one-third of French teens reported the same.
Recommendation
Surface process-oriented prompts alongside completion-oriented ones. Recommended prompts should include brainstorming, outlining, and planning options (e.g., “help me research and plan my essay”).
Finding
At least three-quarters of all teens described AI for creative tasks as “fast” and “helpful,” and most described the final result as “authentic,” “original,” or “personal.”
However, 45% of US, 33% of Brazilian, and 37% of French teens reported feeling less creative when using AI.
Recommendation
Position the tool as a collaborator on creative tasks. Given the share of teens—particularly in the US—who report feeling less creative when using AI, creative context-framing should emphasize user ownership of the idea, direction, and judgment.
2. Verification and bias
Most teens describe chatbots as logical and accurate, and US teens in particular describe them as unbiased. But fewer than half of teens in any country have been taught about AI bias, and rates of fact-checking vary considerably across countries. Product design choices around source citation, bias surfacing, and literacy touchpoints can help close the gap between how teens perceive these tools and how they actually work.
Finding
Many teens describe chatbots as “logical” and “accurate”. A majority of US teens also describe their frequently-used chatbot as “unbiased,” compared with fewer than half of Brazilian and French teens.
A majority of teens in the US and Brazil have received some AI instruction; this education was less common for teens in France.
Within each country, “understanding bias” was the least common AI literacy topic addressed.
Recommendation
Flag where AI is likely to show bias, especially in contested, health-related, and identity-related responses. Because bias is the area teens are least likely to have received education in, prioritize surfacing accuracy concerns and opportunities for further information verification.
Finding
44% of US, 50% of Brazilian, and 26% of French teens reported often or almost always checking chatbot information against other sources. French teens also reported the lowest rates of instruction on identifying or verifying AI-generated content, and the lowest rates of AI literacy instruction overall.
Recommendation
Embed AI literacy educational opportunities throughout the product, not only at onboarding. Short, plain-language explanations of how the model works, how AI answers are surfaced, etc. should appear at relevant moments in use.
Link responses to vetted external sources where possible, and reframe verification nudges around depth of learning. For example, replace language like, “AI can be wrong, check other sources” with “These sources have more on this topic” (and provide links). Also, indicate confidence on responses that may be uncertain, incomplete, or outdated
3. Personal advice, emotional expression, and social use
Teens describe chatbots as warm and nonjudgmental, and a meaningful share uses them for personal advice, emotional expression, and social interaction. While most US and Brazilian teens find this advice helpful, the conversational warmth that makes these interactions feel supportive can also blur the line between a tool and a confidant. Product behavior in these contexts—tone, response patterns, and pathways to human support—has direct implications for how teens come to rely on these tools.
Finding
At least two-thirds of teens across all three countries described a frequently-used chatbot as “warm”. Three-quarters of teens in the US and France, and almost half of Brazilian teens, described it as “nonjudgmental”. Teens were divided on whether their chatbot felt more “humanlike” or “machinelike”.
Recommendation
Avoid language that positions the chatbot as an observer or advisor in personal or emotional contexts. These phrasings imply a capacity for perception, judgment, and clinical insight that the product does not have, and can reinforce teens’ tendency to treat chatbot responses as expert guidance. For example:
- Instead of “here’s what I’d recommend,” use “here are some options.”
- Instead of “what I see happening here,”use “here’s what you’ve described” or “based on what you’ve shared.”
Finding
Between one-quarter and one-third of teens in the US and Brazil who used AI for personal advice did so frequently, while only 10–15% of French teens did so on any topic. Most US and Brazilian teens found this advice helpful; French teens were more divided.
Recommendation
Treat emotional-support conversations as on-ramps to human connection. Where teens use chatbots for support, in-product prompts can suggest how to have similar conversations with friends, family, or counselors.
Finding
Roughly one-third of teens in the US and Brazil frequently used chatbots to roleplay, express thoughts and feelings, or socialize. Fewer than one-quarter of French teens did so.
Recommendation
Build active redirection for mental health, physical health, and crisis-adjacent topics. Provide in-product pathways to professional resources and trusted adults rather than relying on disclaimers alone.
4. Gender gaps in US adolescent use
Among US teens, boys use AI chatbots more frequently than girls, and do so across a wider range of activities, including learning, creativity, advice-seeking, and social engagements. These gaps were not observed at the same magnitude in Brazil or France. Default product design choices, from personas and voices to onboarding prompts and featured use cases, can either reinforce or help close the gap in these patterns of engagement.
Finding
Among US teens, boys used chatbots significantly more often than girls (42% vs. 30% daily) and rated them as significantly more important to their daily lives (48% vs. 28%).
Recommendation
Audit default personas, voices, and framing for gender bias. Many assistant-style products default to feminine voices and personas; meaningful customization reduces reliance on stereotyped defaults.
Finding
US boys reported using AI significantly more often than girls across a wide range of activities, including learning (“research,” “discover,” “practice”) and creative tasks (“create,” “brainstorm,” “edit and refine”).
Recommendation
Broaden the use cases and prompts surfaced to new users. If featured use cases skew toward stereotypically male interests, girls may not see themselves as the intended audience. Onboarding suggestions should reflect a broad range of needs, including independent learning and creative applications.
Finding
US boys reported asking AI for advice significantly more often than girls across all categories.
Recommendation
Design early interactions to build confidence rather than showcase capability. Given evidence that girls may underestimate their own AI competency, low-stakes early experiences matter.
For questions about these findings and recommendations, please email us.