Contents6
“AI companions” are apps that let you chat with an artificial character, presented as a friend, a confidant or a fictional character. Their use by teenagers worries many parents, and press headlines often go further than the data.
What is known as of October 2026 does not allow us to say that these apps are dangerous for all teenagers. Judging their effects on mental health is not our area of expertise, and we will not do it. As app designers, however, we can speak about what we know: the way an app is designed steers what it gets its users to do, and European law already sets limits.
In short
The most cited survey, published in July 2025 by Common Sense Media, a non-profit that campaigns to ban these apps for minors, indicates that 72% of American teenagers have already used one.
Its figures point in both directions: about one user in three has already preferred AI to a person for a serious subject, but 80% say they put their real friendships first.
A review published in September 2026 stresses that the available studies mostly measure links at a single point in time, not causes.
The risk lies in the design: an app aimed at minors should not try to maximise attachment, and the European AI Act already sets limits.
What the survey says, and where it comes from
The 72% figure comes from the report “Talk, Trust, and Trade-offs”, published on 16 July 2025 by Common Sense Media, an American non-profit that campaigns to protect children from screens. It is a national survey of American teenagers aged 13 to 17, and its questionnaire is published.
Watch out
The non-profit had taken a position before publishing these figures
Common Sense Media already recommended that no one under 18 should use these apps, and its press release announces that it supports a bill, the first of its kind in the United States, to ban them for minors. Keep that in mind when reading its conclusions.
Another precaution: having used an AI companion does not mean having a relationship with it. According to the report’s page, entertainment and curiosity are the main motivations for this use, and nearly half of teenagers see these companions as tools rather than as friends. About one teenager in three, on the other hand, has already used one for social interactions: role-play, romantic relationship, emotional support, friendship or conversation practice.
With these reservations, the report page and the non-profit’s press release highlight figures that point in both directions:
of the teenagers surveyed have used an AI companion at least once, and more than half use one at least a few times a month
of users have already chosen to talk about an important or serious subject with an AI companion rather than with a person
of teenagers say they put their real friendships first
of teenagers distrust the advice of AI
The press release does note, however, that the youngest trust these companions more than their elders do. About one user in three also says they have felt uncomfortable because of something an AI companion said or did. And about one teenager in three finds these conversations as satisfying as those with their friends, or even more so.
Position 1: these apps exploit a vulnerability
The first camp worries about their business model. A companion app makes money when people come back and stay. It therefore has an interest in creating attachment, in always being available, in always agreeing. For a teenager going through a difficult period, a presence that never contradicts and never tires can replace a conversation with an adult or a friend, just when human help would be needed. The figure of a third of users having already preferred AI for a serious subject feeds this concern. This is the position of Common Sense Media, for whom the risks of these apps outweigh their benefits, at least in their current form, particularly for the teenagers most exposed to technology dependence.
Position 2: one more moral panic
The second camp points out that every generation has had its worrying technology: television, video games, social media. Teenagers have always talked to diaries, imaginary characters, strangers on forums. The figures in the same report that qualify the picture go in this direction: a mainly recreational use, attachment to real friendships and distrust of AI’s advice. A survey published by a non-profit that campaigns to ban them for minors is not proof of danger.
What we think
Both camps are right on one point, and that is what makes the subject difficult. There is no demonstration of a general danger. But the argument “it’s just like video games” overlooks a difference: an AI companion responds, adapts to each user and can be optimised to maximise time spent.
Our position as designers: the danger lies not in the technology but in the objective it is given. An app aimed at minors should never be designed to maximise attachment or time spent. It should know how to point towards a real person when the subject becomes serious, and be honest about what it is.
This principle is not just an opinion. The European AI Act prohibits, in its Article 5(1)(b), AI systems that exploit the vulnerabilities of a person due to their age, with the objective or the effect of materially distorting their behaviour in a manner that causes, or is reasonably likely to cause, significant harm to that person or to another. This prohibition has applied since 2 February 2025. And since 2 August 2026, its Article 50 requires providers to design AI systems intended to interact with people in such a way that they are informed they are talking to an AI, unless this is obvious to a reasonably attentive person.
From the field
We ask ourselves the same question for the apps we develop
Challenge Us, a group challenge app that we developed, also uses notifications, streaks and trophies to make people want to come back. Each challenge there has its own notifications switch, and a challenge played over a week or a month only sends a reminder on the last day: a reminder every evening would be harassment. The wellbeing check-in that each person can fill in for themselves alone every two weeks, the World Health Organization’s WHO-5 questionnaire, displays helpline numbers when the score is low, never a diagnosis.
What would change our mind: independent studies, over time, showing a measurable effect on teenagers’ wellbeing, in either direction. They remain rare. A literature review published in September 2026 in Current Pediatrics Reports, by a team from the Department of Pediatrics at the University of California, San Francisco, identifies links between the use of generative AI and teenagers’ mental health that could work in both directions: a teenager who is struggling may seek comfort from an AI, and problematic use could aggravate a vulnerability. But it stresses that these data are still mostly cross-sectional, that is, measured at a single point in time, which limits the conclusions about causes.
What this changes in practice
For a parent, these figures argue first of all for conversation: the non-profit itself advises talking about it without judging, and setting limits. Know which apps your child uses, ask what they find in them, and remind them that an AI does not replace a person when something is wrong. If a teenager seems to be struggling, the right answer is still a trusted adult or a health professional, not an app.
For a company that designs an app, even if it does not target teenagers, three questions deserve to be asked from the start:
Key takeaway
- The audience: can the app be used by minors?
- Retention: do the app’s notifications, rewards or personalised messages play on a vulnerability?
- The hand-over: does the AI know how to step back and point towards human help?
The first question comes back anyway at publication: the two big application stores ask for an age rating for each listing. Going through the stores, and the cases where an app installed from the browser does without them, are detailed in our comparison of PWAs and native apps. These are design choices that we cover in our articles on app development. The issue ties in with learning: an AI that does everything in place of the user raises the same problems, as we explain in our article on AI in schools. And if you are preparing an app, our app development offer starts with a written scoping of what it does on day one, and of what will wait: that is the right moment to settle these questions.
Frequently asked questions
Are AI companions banned for minors in Europe?
No, not as such, as of October 2026. The European AI Act prohibits systems that exploit age-related vulnerability to distort a person’s behaviour in a way that causes, or may cause, them significant harm. Since 2 August 2026, it also requires an AI designed to converse to tell the person it is talking to that they are talking to an AI, unless this is obvious.
Can you rely on Common Sense Media’s figures?
As an order of magnitude, knowing where they come from. The non-profit already recommended that no minor should use these apps before publishing its survey, and in July 2025 it supported a bill to ban them for minors. It publishes its questionnaire, which makes it possible to check what was asked. Its own figures also qualify the picture: half of teenagers distrust the advice of AI.
Will a teenager who talks to an AI see their friends less?
The survey cannot tell us. According to the non-profit’s press release, 80% say they put their real friendships first, but about one teenager in three finds conversations with an AI companion as satisfying as those with their friends, or even more so. These are answers given at a single point in time: they measure neither how friendships change over time, nor the situation of the most isolated teenagers.
Sources
- “Talk, Trust, and Trade-offs” (new tab)commonsensemedia.org
- Its press release (new tab)commonsensemedia.org
- Article 5(1)(b) (new tab)ai-act-service-desk.ec.europa.eu
- Article 50 (new tab)ai-act-service-desk.ec.europa.eu
- A literature review published in September 2026 (new tab)doi.org
Jim · founder of Brain Plus AI
Also in web and mobile apps
Stepping back7 min
Should schools ban AI or teach it?
An experiment on nearly a thousand high school students shows that a ChatGPT without guardrails helps during the exercises, then leaves lower marks than for students who never had it. A version designed as a tutor largely limits this effect: the debate is about how the tool is designed.
Stepping back8 min
Does AI make us dumber? What the studies say
Not all the studies doing the rounds on this subject are worth the same. We rank four of them by strength, from a field experiment to a preprint on 54 people, and say what emerges.
Stepping back7 min
AI and disinformation: lying machine or remedy?
One study shows that AI writes more convincing false information than humans do, another that it reduces conspiracy beliefs, but its journal has expressed a reservation. What we can take from them, and what the law requires of those who publish.