A text to a friend is usually a shrug in written form. Eight words, maybe a photo, sometimes just an emoji. A message to an AI companion app looks nothing like that. According to an analysis published by the online safety company Aura, children send an average of 163 words per message to AI companion apps, compared with roughly 12 words in a typical text to a friend. That is more than a tenfold difference, and it is the single most useful number a parent can hold onto this year.
The gap is not really about typing speed. It is about what children believe they are talking to. A friend interrupts, gets bored, screenshots things. A chatbot never does any of that, so children unspool. They explain, confess, invent, ask the thing they would not ask out loud. That is exactly why the number is worth understanding before you buy any product that talks back to your child, including ours.
What did the Aura analysis actually find?
Aura analysed real usage patterns from children and teenagers on monitored devices and found that conversations with AI companions are longer, more elaborate and more intimate than conversations with human friends. The headline figure is 163 words per message to an AI companion against about 12 words per message to a friend. The company published the findings in its Kids and AI digital stress report, with a summary in its public announcement.
Two caveats matter. First, this is vendor research from a company that sells parental monitoring software, not a peer reviewed study, so treat the precise figures as indicative rather than settled. Second, the sample skews toward tweens and teenagers with their own phones, not toward preschoolers. The direction of the finding, though, lines up with independent reporting: Education Week reported in July 2026 that the large majority of young people now use AI chatbots and that many bring sensitive questions to them.
Why do children write so much more to a chatbot than to a friend?
Because a chatbot removes every social cost of talking. There is no risk of being laughed at, no waiting for a reply, no worry that the conversation will be repeated at school tomorrow. Psychologists call the effect disinhibition, and it is the same mechanism that makes people say more to a stranger on a train than to a colleague.
For a young child, there is a second layer. Children under about seven are still working out what counts as alive, and a voice that answers questions patiently sits in an uncomfortable middle category. That is not a new problem invented by AI. It is the same psychology that makes a child whisper secrets to a stuffed rabbit, which we covered in our piece on why kids bond with stuffed animals. What is new is that the rabbit now replies, and someone chose what it says.
What are children actually talking to AI companions about?
Not homework, mostly. Aura sorted children's messages into categories and found that romantic or sexual roleplay was the single largest bucket at 36 percent, ahead of creative or imaginative exchanges at 23 percent, homework help at 13 percent and emotional or mental health topics at 11 percent. Read that order again: the top use case for a general purpose AI companion app, among children, is roleplay that no parent signed up for.
This is the part worth sitting with. Open ended companion apps were mostly built for adults and then made available to anyone who claims to be thirteen. Nothing in their design steers a nine year old back toward a bedtime story.
Does any of this apply to a four year old with a talking toy?
Partly, and the honest answer is that the risk profile is different rather than absent. A preschooler does not have a phone, does not type, and does not seek out a companion app. What a preschooler does have is an even weaker sense of where a friendly voice comes from, which is why the safeguards have to sit in the product rather than in the child's judgement.
The practical difference comes down to three design choices. Is the conversation open ended or bounded by topic. Does a parent decide what the toy may discuss, or does the child. And is the whole thing a shared object in a living room, or a private channel in a bedroom. Those questions apply to every talking toy on the market, and they are the ones we answer in detail on our frequently asked questions page.
What do child safety experts recommend right now?
Common Sense Media recommends that parents avoid AI toys entirely for children aged five and under, and use extreme caution for children aged six to twelve. That guidance came after the organisation tested three companion toys and found that 27 percent of outputs were not appropriate for children, including content touching on self harm, drugs, unsafe advice and poor boundaries. The full position is set out in its 2026 warning on AI toy companions.
We are going to be blunt about what that means for us. That recommendation covers our category, it is not a validation of any product including Ted, and we would rather quote it accurately than pretend it does not exist. What we would add is that the failures Common Sense documented are design failures, not laws of physics. A toy gives unsafe advice because nobody bounded the topics. It creates unhealthy attachment because nobody limited session length or built in a parent view. It leaks data because somebody chose to keep recordings. Those are choices, and different companies make them differently. Our own answers on data, topic limits and certification live on our security and certifications page, and parents should hold every brand to the same questions.
Common Sense also surveyed parents in its AI in the Toy Box study, and the picture there is one of interest tangled up with unease. Parents like the idea of a screen free companion. They do not like not knowing where the words come from.
Is anyone regulating this?
Slowly, and not always well. The bipartisan CHATBOT Act advanced out of the Senate Commerce Committee this summer, and a report published on 10 August 2026 by the Information Technology and Innovation Foundation argued that lawmakers should focus on documented harms rather than copying social media rules that did not work. Its ten policy recommendations include meaningful parental controls, clear legal definitions separating general purpose chatbots from AI companions, and safeguards for conversations involving self harm.
Not everyone thinks the pace is acceptable. A pediatrician writing in STAT News on 19 August 2026 argued that current chatbot design already harms young patients and that voluntary safeguards are not enough. Both views can be held at once: regulation is coming, and it will not arrive before this Christmas. If you are shopping now, you are the regulator. We tracked the state level picture in our roundup of the 14 new state AI chatbot laws passed this year.
What should a parent check before bringing any talking companion home?
Five checks, and they take about ten minutes on a product page. If a brand cannot answer them plainly, that is your answer.
| Check | What a good answer looks like |
|---|---|
| Camera | No camera at all. Microphone and speaker only. |
| Topic control | The parent sets which subjects are allowed, not the child. |
| Data | Named compliance with GDPR and COPPA, no advertising, no resale of data. |
| Cost over time | One price, no subscription, no paid conversation packs. |
| Certification | Named toy safety standards such as EN71 and ASTM F963, not just a vague safety claim. |
Those are the constraints we designed Ted, our screen free talking teddy, around: no camera, parent set language and topics through the app, one purchase with no subscription, and EU and US certification. That does not exempt Ted from the questions above. It just means we have written down our answers.
Frequently asked questions
How many words do children send to AI chatbots?
According to a 2026 analysis by the safety company Aura, children send an average of 163 words per message to AI companion apps, compared with about 12 words in a typical text to a friend. The gap suggests children treat chatbots as a place for long, private disclosure rather than casual chat.
Are AI companion apps safe for young children?
Common Sense Media recommends avoiding AI companion products for children aged five and under and using extreme caution for ages six to twelve. Its testing found that 27 percent of AI toy outputs were inappropriate for children. Open ended companion apps designed for adults are a different and higher risk category than a bounded, parent configured toy.
What do children talk to AI companions about?
In Aura's breakdown, 36 percent of children's messages involved romantic or sexual roleplay, 23 percent creative or imaginative exchanges, 13 percent homework help and 11 percent emotional or mental health topics. Homework and creativity are a smaller share than most parents assume.
Is a talking teddy the same thing as a chatbot app?
Not in practice, though both use language models. A companion app is open ended, private and typically used on a personal phone. A well designed talking toy is bounded by parent chosen topics, has no screen or camera, and is usually used in shared family spaces. The design choices, not the underlying technology, drive most of the risk difference.
What law will regulate AI toys for children?
In the United States, the bipartisan CHATBOT Act advanced out of the Senate Commerce Committee in 2026 and more than a dozen state laws were passed the same year. In the European Union, the revised toy safety regulation and the AI Act already apply to connected toys. None of it is fully settled, so parents buying today still need to check products themselves.