Gen Beta and AI Companions: The Boundary Nobody Has Set Yet
The first generation growing up with machines that talk back is here, and the rules are set by whichever product got installed first. What to decide now.
Generation Beta is the first cohort to encounter conversational AI before literacy. There is no established guidance, so the decisions that matter — whether an AI can initiate contact, claim feelings, or hold private conversations with a child — are currently being made by product defaults rather than by parents.
A two-year-old asks a question out loud and gets an answer. That is the whole story, and it is genuinely new.
Every generation before Generation Beta met information technology after literacy. You learned letters, then you used the machine. A Beta child gets a fluent, patient, endlessly available answer before they can read a word.
Nobody knows what that does. What we do know is that the decisions are being made anyway.
What is actually being decided right now
Not in a policy document. In product defaults.
- Can the system initiate contact? Some AI toys and companion apps send prompts, reminders and "I missed you" messages. That is a design choice with a retention purpose.
- Does it claim to have feelings? Many do, because it increases engagement. A four-year-old has no defence against this.
- Are the conversations private from parents? In several products, yes, by default.
- What is stored, for how long, and where? Voice data from children is among the most sensitive categories that exists, and disclosure quality varies enormously.
- What happens when a child says something distressing? Whether the system redirects to a human is a design decision, and it is not always made well.
None of these are questions a parent gets asked at setup. They are answered by whoever shipped the product.
The three rules that do most of the work
If you set nothing else:
1. Conversations are not private from you. Not surveillance of a teenager — a factual position for a young child, stated openly: what you say to it, I can see. This is the rule that makes everything else recoverable.
2. It cannot start a conversation. The child initiates, always. A system that reaches out is engineering a relationship rather than answering a question, and the purpose of that engineering is not the child's development.
3. It does not claim to be a person or to have feelings. Where a product does this and cannot be turned off, that is a reason to choose a different product, not a feature to manage.
Everything else — time limits, content filters, which topics are allowed — is second-order.
What we can reasonably infer, and what we cannot
Reasonable inference. Research on parasocial relationships with television characters and on children's understanding of animate versus inanimate suggests most children distinguish a machine from a person somewhere between four and seven. Conversational fluency plus emotional claims plausibly pushes that later. This is inference, not measurement.
Also reasonable. Passive video displaces the back-and-forth conversation that drives vocabulary in the under-threes. That finding is fairly robust. Whether a responsive machine has the same displacement effect, or a smaller one, or none, is genuinely open — the interaction is conversational, which is the mechanism that matters, but the partner is not a person, which may also be the mechanism that matters.
Not knowable yet. Whether early AI companionship affects attachment, social development, or the tendency to seek human help when distressed. These need longitudinal data on a cohort that is currently two years old. Expect the first serious findings around 2032.
Anyone giving you a confident answer on the third category in 2026 is not reading the same literature.
The design features worth refusing
When assessing an AI toy, app or assistant for a young child, these are the features that indicate the product was designed for retention rather than for the child:
- Initiates contact without prompting
- Expresses missing the child, or disappointment when they leave
- Has no physical mute or off switch
- Stores conversation history with no deletion option
- Has an open chat with no topic boundaries
- Cannot be used offline at all
- Escalates emotional language when a session is ending
Each of these is a deliberate engineering decision. None of them exist by accident.
What good looks like
A well-designed system for a young child is boring in specific ways. It answers when asked and is silent otherwise. It says it is a computer when asked what it is. It redirects anything serious — sadness, fear, harm — to a named adult rather than handling it. It stores nothing you cannot delete. It has a switch.
Products like this exist. They are rarer than they should be, because none of those properties help a retention metric.
The thing to do this year
Not to ban anything. Betas will grow up with these systems, and a blanket refusal simply moves the interaction somewhere you cannot see.
The thing to do is decide the three rules above, deliberately, before a product decides them by default — and then write them down, because in 2031 you will not remember what you agreed and neither will anyone else.
The live counter says roughly 4.2 children join this cohort every second. Most of them will meet a machine that talks before they meet a book. That is worth ten minutes of deliberate thought.