What Happens When Your Child's New Best Friend Is an AI?
A UNICEF policy brief published this year draws a distinction I hadn't seen stated so plainly before, but that bubbled around in my mind at BETT 2026 as I saw just how much AI was being touted: an AI tutor and an AI companion are not the same thing, and they don’t carry the same category of risk.
I read a lot of AI-in-education material for obvious reasons. Most of it is about tutoring, feedback, and lesson planning etc. This year, a different strand of writing has been growing alongside it, and it's a harder one to think clearly about because there is a blend between the very functions of the AI.
AI companions. Not tools that help a child learn something. Systems designed, quite deliberately, to be someone.
UNICEF published a policy brief this year specifically on this category, and it's worth reading in full if you build anything for children. What stood out to me is how carefully it separates companion apps from assistant chatbots. The distinguishing feature is that companion apps are built, quite deliberately, to initiate and sustain something that feels like a relationship: persistent memory, a consistent personality, the ability to be shaped to the user's preferences over time.
The scale of this cannot be ignored. Between 2022 and mid-2025, the number of AI companion apps grew by roughly 700%, according to reporting cited in the American Psychological Association's own coverage of the trend. Common Sense Media, the children's media watchdog, assessed some of the major companion products last year and found they repeatedly failed to respond appropriately when teenagers disclosed thoughts of self-harm.
Terms such as “Synthetic relationships are filling the void to satisfy the fundamental human need for social connection.” really hit home. This is BIG and is hitting the streets fast, with post-fact scrutiny scrabbling to catch up… again. Does that pattern sound familiar yet?
That's a documented failure mode, already showing up in a product category being marketed, in part, to teenagers.
I want to be careful here, because this sits outside my area of authority. I'm not a psychologist, so I'll stop short of offering a confident view on the mental health mechanics of such intimate attachment to a chatbot. What I can speak to, a little, is the design side of the question, because it overlaps with decisions we make about oodlü.
Here's the question I keep turning over.
Every design choice that makes an AI companion feel more real, persistent memory, a consistent personality, responses tuned to keep a conversation going, is also a design choice that increases engagement. In a companion product, those are one and the same lever. The World Economic Forum's own framing this year put it simply: these systems can be a dangerous mix for young people, specifically when age verification is weak, and the underlying system is optimised for engagement.
That's uncomfortable, because it means the incentive that makes a companion app commercially successful is close to the same incentive that makes it psychologically risky for a child using it.
I don't think oodlü has a clean answer to offer here, and I'd be suspicious of anyone who claimed they did. Where we've landed, for now, is a narrower one: AI in oodlü assists a responsible adult in generating content for children to explore. It sticks to that job, rather than role-playing as a friend or being built to be remembered as one. That's a deliberate boundary we've drawn, and an evolving one rather than a settled answer, one we'll keep revisiting as the underlying technology changes faster than any of us can write policy for it.
What I'd genuinely like to know, from anyone reading this who works with children, whether that's teaching, home education, or child psychology, is where you think the line actually sits.
Is it persistent memory that's the risky ingredient?
Is it the anthropomorphism?
Is it simply that no chatbot, however designed, should be marketed to a lonely teenager as company?
I don't think anyone has a settled answer to that yet, and I'd rather say so than pretend otherwise.
We'd love to hear your thoughts on this. Find us on the social channels linked at the top of the page.