Parents may know that their children use chatbots for homework without realising how much teen AI use has spread into ordinary life. A French survey commissioned by the policy organisation Interface found a particularly revealing gap among children aged 11 and 12: 40% reported regular use of general-purpose AI, while parents' estimates put the figure at 26%. The finding is not proof that families are ignoring technology. It suggests adults and children are sometimes looking at different parts of the same digital world.
The research, published on 22 October 2026, was based on online interviews conducted by YouGov in France during June. It involved 970 teenagers aged 11 to 17 and 1,138 parents of children in that age range. Because this is a survey, its findings describe what respondents reported and cannot establish exactly how every French child behaves. It also cannot prove that an AI service caused any reported emotional difficulty. Those limits are important, particularly when discussing children.
Teen AI Use Begins Earlier Than Many Parents Expect
The youngest respondents were already familiar with a surprisingly wide range of systems. Among 11- and 12-year-olds, 56% reported using at least one form of generative AI weekly. That included general-purpose chatbots, voice assistants, features inside social media and systems designed to keep up a conversation. These are not interchangeable experiences, and the survey deliberately distinguished between them. Asking whether a child uses ChatGPT is therefore unlikely to capture the whole picture.
For this age group, 40% reported using general-purpose generative AI at least weekly, while 33% reported regular voice-assistant use. Some also said they used AI built into social networks or companion-style chatbots. Parents who think of AI as a website for writing homework answers could easily miss an assistant woven into an app the family already knows. Conversely, a voice assistant used for a simple question is not necessarily equivalent to a conversational service encouraging personal disclosure.
The survey also found that both teenagers and adults were active users. Among the sampled groups, 63% of parents and 63% of teenagers said they used some type of generative AI at least weekly. The percentages are similar, but the purposes, settings and risks can differ sharply. An adult using an assistant to organise a work email is not having the same experience as a child asking a simulated friend for reassurance. Familiarity with one's own tools does not guarantee understanding of a child's tools.
Homework Is Only Part of the Attraction
Teenagers often value these systems for straightforward reasons. They can explain an unfamiliar subject, help someone get started with a difficult task or provide an answer without the embarrassment of asking in front of classmates. A child who is stuck on schoolwork may find it easier to try a chatbot than to admit confusion. That convenience is also where educational questions arise: an explanation can support learning, but a finished answer can bypass the struggle that helps someone understand a subject.
The distinction matters more than declaring every use good or bad. A student who asks for a new example of a mathematical method is behaving differently from one who asks an assistant to submit the solution. Equally, a fluent explanation may contain mistakes or conceal uncertainty. Adults can help by asking children to describe how they checked an answer and what part of the work they did independently. That shifts the conversation away from detection and punishment towards learning and judgement.
This concern connects with LiveAIWire's previous coverage of students seeking guidance about AI at university. The age group is different, but the underlying challenge is similar: young people need practical rules that distinguish legitimate assistance from the substitution of an automated answer for their own work. The French findings show why those habits may need to begin well before higher education.
A Chatbot That Talks Back Can Feel Different
Companion-style AI raises a separate issue because its purpose is not merely to answer a question. Such systems may simulate a persistent relationship, hold extended conversations and respond in emotionally engaging ways. Among the youngest French respondents, 16% reported weekly use of AI companions. The number refers to what those children reported, not a clinical assessment or a direct observation of every application they used.
That is worth examining without suggesting that every child who chats to an AI character is in danger. People engage with fictional characters and imagined conversations for many reasons. A system that responds personally, remembers conversational details or encourages repeated exchanges nevertheless changes the experience from a static story. Young users may not always know which parts are automated performance, what the provider records or what a service can reliably understand.
The survey offers another warning about overstating findings. Among teenage AI users, 92% said they spent more time than they wanted on at least one of the tools covered. That is a self-reported judgement about time, not a diagnosis of addiction. Someone who spends longer than intended using a voice assistant, social feature or chatbot has not necessarily developed a medical or psychological problem. The result does, however, justify asking how products encourage continued engagement.
LiveAIWire has previously explored the trust questions raised by AI toys for children. A toy, a social media assistant and a text chatbot may look different, but parents can ask the same simple questions: does the child understand who or what is responding, what information is being retained and whether the system's advice should be treated as reliable? Those conversations may be more useful than relying on a single product label.
Correlation Is Not Proof of Harm
The Interface report examined well-being as well as use. It found differences in reported negative emotions between some groups, including more frequent reports among regular users of particular conversational services. The report itself cautions that the results do not establish causation. Children who are already worried or lonely might be more inclined to use a companion, the service might affect their feelings, or other factors could influence both. A survey of this kind cannot reliably untangle those explanations.
That limitation matters because public discussions about children and technology can swing between extremes. Describing all AI use as harmless overlooks real issues involving privacy, misleading responses and manipulation. Describing all young users as damaged by the technology goes beyond the evidence. A stronger approach is to examine the design of specific services, listen to children about what they actually do and investigate outcomes rather than assume them.
Families also face a practical information problem. Providers can integrate AI into services without the feature being obvious to everyone using the household device. A parent may set rules for one chatbot while overlooking another assistant inside a social app. Screenshots of product settings help only if adults know which settings are relevant. That is why child safety is not a problem families can solve simply by memorising brand names.
Better Conversations Start With Specific Questions
A useful family conversation might begin with what the child uses an assistant for, which answers they trust and what they would do if a chatbot gave upsetting or strange advice. Asking a child to demonstrate a typical harmless interaction can reveal far more than asking for a yes-or-no declaration about AI. Adults can also explain that a conversational system may sound understanding without having the judgement, accountability or duties of a trusted person.
Privacy deserves equal attention. Children may share details about friends, school, addresses or feelings without appreciating that the conversation is passing through a commercial service. The UK's Information Commissioner's Office explains the Children's code, including the expectation of protective default settings in relevant online services. That guidance is not a claim that UK standards govern the French survey; it is a useful illustration of the responsibility service providers can have alongside families.
The wider developmental questions have also appeared in LiveAIWire's coverage of children growing up with AI. Schools can help by teaching students how to verify answers, identify private information and recognise when a pleasant-sounding response is not dependable. Technology companies can help by making settings, age restrictions and reporting routes intelligible, rather than expecting families to decipher complicated menus. Parents still have a role, but their choices depend on product designs they did not create. The French report found parents wanted more institutional support, a reasonable reaction to a fast-changing set of tools.
The Question Has Changed From Whether Children Use AI
The study does not show that every family has lost control, nor does it justify treating young people as passive victims. It shows that AI use is already varied, often begins early and is not always visible to the adults responsible for children. The gap is especially meaningful when a parent thinks only about the most famous chatbot while the child encounters AI through several unrelated services.
A better response begins with curiosity grounded in evidence. Which tools are actually being used? Are they helping with a task, encouraging an emotional relationship or collecting information the child does not understand? Can the child recognise an unreliable answer and turn to a person when something feels wrong? The answers will vary between children, applications and situations, which is precisely why one frightening statistic cannot do the job of a thoughtful conversation.
The enduring lesson is simple. Parents do not need to become AI engineers, but they do need to know what the technology looks like in their children's daily lives. The French findings suggest that starting the conversation with only a familiar chatbot name may already be too narrow.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.