What Teenagers' AI Use Reveals About the EU Kids Act
Jessica Galissaire / Nov 4, 2026
Parents are underestimating how frequently their children use generative artificial intelligence, especially at younger ages. According to a new YouGov study commissioned by interface, 40% of French 11- and 12-year-olds report using general-purpose generative AI at least weekly, compared with just 26% according to parents’ own estimates.
More strikingly, among the 11- and 12-year-olds who use generative AI, 14% say they turn to these tools to confide in them. These findings illustrate how generative AI is becoming part of young people's lives in ways that go well beyond homework assistance or information searches, while parents are not necessarily equipped to support them in navigating these new experiences.
They also come at a particularly timely moment for European regulation. In September, the European Commission presented its EU KIDS Act, a proposal introducing obligations across a range of digital services, including social media, video-sharing platforms, online games, and AI chatbots and companions.
But while the Commission has proposed a clear minimum age for social media access, its approach to AI chatbots and companions is notably different. This raises an important question: how well does the EU's emerging regulatory framework reflect the ways children actually use these technologies, and the risks they may encounter?
Children are already growing up with generative AI
Between June 24 and 30, YouGov surveyed 970 French children and teenagers aged 11 to 17, alongside 1,138 parents of children in the same age range. The study examined their use of generative AI, including general-purpose chatbots such as ChatGPT, AI companions such as Character.AI, AI embedded in social media such as My AI on Snapchat, and voice assistants such as Siri.
One finding stands out: generative AI use is already widespread, including among the youngest respondents. Perhaps more surprisingly, 23% of children in the 11-12 age group report regularly using generative AI embedded in social media, while 16% regularly use AI companions, despite most of these services' terms and conditions restricting access under the age of 13.
For teenagers, generative AI is primarily a practical resource, helping them find information, save time, learn and complete schoolwork. Among those familiar with it, 89% identify at least one benefit. Parents do recognize its potential, but they also express concerns, particularly about children's ability to think independently, the reliability of AI-generated information and the protection of personal data. Moreover, 44% of parents whose children use generative AI say they do not feel at ease with that use.
The findings therefore paint a more nuanced picture than one of children simply being exposed to a new category of online harms. Generative AI is already serving different purposes in their daily lives, some beneficial, others potentially concerning. Understanding these differences is essential to developing an appropriate regulatory response.
Not all AI interactions are created equal
Another striking finding concerns the time teenagers spend interacting with these systems.
Among surveyed teenagers who use generative AI, 92% report spending more time than they would like on at least one of the tools covered by the study. This feeling is particularly pronounced among users of AI companions (59%) and generative AI embedded in social media (57%), compared with 45% for general-purpose generative AI users.
These figures should not be interpreted as evidence of addiction. Spending more time than intended is a subjective experience, not a clinical diagnosis. Nevertheless, the differences between categories suggest that the design and purpose of these systems deserve closer attention. In particular, AI companions are explicitly designed to sustain ongoing conversations, often through personalized interactions that can resemble human relationships. AI embedded in social media, meanwhile, operates within environments whose business models already rely on encouraging repeated and maximized engagement.
The survey also reveals a relationship worth investigating between generative AI use and emotional well-being. While 38% of all surveyed teenagers report experiencing negative emotions at least weekly, the proportion rises to 50% among regular users of AI companions. Of course, correlation does not imply causation. It may be that teenagers already experiencing emotional difficulties are more likely to turn to conversational AI for reassurance or companionship. This uncertainty is precisely why further research is needed.
The KIDS Act's uneven approach to age restrictions
So how does the EU KIDS Act address these different uses and risks?
One of the proposal's main additions to the existing EU digital rulebook is a three-tier system for accessing social media and video-sharing services presenting certain risks. Children under 13 would generally be prohibited from holding accounts on these services, those aged 13 and 14 could use limited accounts supervised by a parent or guardian, while teenagers aged 15 and above could open their own accounts.
AI chatbots and companions, however, are treated differently. Although the proposal does not establish an equivalent general prohibition on access for children under 13, Article 14 requires their access to be enabled and controlled through parental tools. These services are also subject to specific safety-by-design obligations, including measures addressing potentially addictive designs and emotionally dependent relationships.
The reliance on parental controls deserves scrutiny. According to the survey, only 32% of parents feel sufficiently supported in helping their children navigate generative AI, while 42% report receiving no support at all. Parental controls can be useful, but policymakers should take into account that parents do not always have a clear picture of how their children use these technologies. The primary responsibility for mitigating risks should remain with the companies designing the systems.
This difference in treatment is particularly interesting in light of another of the survey’s key findings. Younger children already report using generative AI for emotional support, while users of AI companions are especially likely to feel they spend more time on these systems than intended. The KIDS Act does not overlook these risks. Article 14 introduces specific obligations for AI companions and conversational chatbots. Providers would have to avoid designs that simulate interpersonal relationships in ways likely to create emotional dependency, refrain by default from using information derived from previous interactions in subsequent conversations, and conduct risk evaluations before placing their systems on the market.
However, the difference in the treatment of access remains difficult to ignore. Why does the KIDS Act prohibit children under 13 from holding accounts on certain platforms, while allowing them to access AI companions under parental supervision? There may be legitimate reasons. Social media and conversational AI present distinct risks, and the same safeguards will not necessarily be appropriate for both. But those distinctions should be supported by evidence, not assumptions about which categories of technology are inherently more dangerous.
The boundaries are also becoming increasingly blurred. Generative AI is now embedded in social media services, meaning that children may encounter similar conversational functionalities across different types of platforms. The KIDS Act recognizes this convergence, introducing additional safeguards for AI companions and chatbots embedded in social media, video-sharing services and online games, including restrictions on their automatic activation. Nevertheless, the differences in age-based access requirements raise questions about how consistently the proposal addresses risks associated with similar functionalities.
What this means for the KIDS Act negotiations
The survey offers several lessons for European lawmakers as negotiations on the KIDS Act get underway.
First, policymakers should pay greater attention to the characteristics of AI systems that encourage prolonged, personalized or emotionally significant interactions. Article 14 provides an important starting point, but the effectiveness of its safeguards will depend on how risks are assessed, how compliance is demonstrated, and whether enforcement authorities have the expertise needed to evaluate these systems.
Second, the EU should ensure greater consistency in how it approaches risks across different categories of digital services. Age restrictions may be justified in certain circumstances, but they should be proportionate and grounded in evidence. More importantly, restricting access should not become a substitute for addressing the features that make a service potentially harmful.
Third, policymakers need to invest in independent, longitudinal research examining how children's interactions with generative AI evolve over time, particularly among younger and potentially more vulnerable users. The current evidence remains limited, and self-reported surveys such as this one cannot establish causal relationships between AI use and well-being. Still, the absence of conclusive evidence should neither justify inaction nor serve as a blank cheque for sweeping restrictions.
Finally, regulation must preserve the opportunities generative AI can offer children, including learning, creativity and access to information, while ensuring they are not left alone to navigate potentially harmful systems.
The KIDS Act represents an important opportunity to move European child-safety regulation beyond managing access and towards greater accountability for digital product design. Its success, however, will depend on whether its provisions adequately reflect children's actual experiences. The fundamental question is not simply at what age children should be allowed to use a particular technology, but under what conditions that technology can be made safe and beneficial for them. If the EU wants to develop an effective framework for protecting children in the age of generative AI, that is where the debate should begin.
Authors
