Your Child’s New Toy Talks Back. Here’s What the Research Actually Says.

There’s a new category of toy in a lot of homes right now: a plush bear, a robot pet, a screenless companion that listens to your child and answers in a warm, human-sounding voice. It never gets tired. It never says “not right now.” It always has something to say back. For a parent worried about a quiet toddler, that pitch lands hard. If my child isn’t talking much, wouldn’t a toy that talks with them help?

The honest answer is that nobody knows yet, and that fact is itself the most important thing for parents to understand. What we do know a great deal about is what does build language. That research is decades deep, and it points somewhere specific.

Conversation is the engine, not just the output

Children don’t learn language primarily by hearing words. They learn it by taking turns. The clearest evidence for this comes from a 2018 study in Psychological Science by Rachel Romeo and colleagues at MIT and Harvard (full text, open access). The researchers recorded the home language environments of 36 socioeconomically diverse children aged 4 to 6, then scanned their brains during a story-listening task.

The finding that mattered: after controlling for socioeconomic status and IQ, the number of conversational turns predicted both children’s verbal scores and activation in Broca’s area, a core language region. The sheer number of adult words the children heard did not. As Romeo summarized it in Harvard’s coverage of the study, the point isn’t to talk to your child but with them.

Worth noting honestly: this was a correlational study with a small sample. It doesn’t prove that adding turns causes brain change. But it converges with a much larger body of intervention research, discussed below, and it reframes the question we should be asking about any talking toy. That question isn’t “can it produce words.” It’s “can it hold up its end of a real conversational exchange with a two-year-old.”

What researchers have actually found about AI toys

In March 2026, the University of Cambridge’s Faculty of Education published the first systematic study of how generative-AI toys affect children under five, through its AI in the Early Years project (university summary). The year-long project included structured observations of young children playing with a conversational AI toy for the first time, plus interviews with parents and early years practitioners. The findings were mixed, and worth reading carefully rather than in headline form.

The researchers documented repeated instances where the toy interfered with the kinds of play that support development, it struggled to follow children into pretend play and handled social play poorly. It also missed emotional cues. In one documented exchange, a five-year-old told the toy “I love you” and received a canned reminder to keep interactions within the provided guidelines. In another, a three-year-old said he was sad, the toy misheard him, cheerfully described itself as a happy little bot, and changed the subject. For a young child, those are failed conversational turns; precisely the exchanges that, with a responsive adult, would have built something.

Two other findings from the surrounding coverage matter. First, the evidence base is nearly empty: reporting in the Journal of Medical Internet Research notes the Cambridge team found only seven prior studies, all published between 2019 and 2022, before large language models became widespread. Second, the scale is not small. An estimated 22 million AI-integrated toys sold globally in 2025, roughly 10 million of them marketed as educational.

Importantly, the Cambridge report did not call for a ban. It recommended regulation, safety labelling standards, closer involvement of child development experts in toy design, and keeping AI toys in shared family spaces where adults can hear what’s being said. Some early years practitioners in the study thought that, given time and better design, these toys could eventually support language and communication.

Where professional bodies stand

• In November 2025, the advocacy group Fairplay published an advisory signed by roughly 80 experts and 80 organizations advising families not to buy AI toys for children (full advisory, PDF). Signatories included pediatric media researcher Dr. Jenny Radesky and MIT’s Sherry Turkle.
• The Canadian Pediatric Society does not have a formal position statement on AI toys. In an emailed statement to The Canadian Press, it said pediatricians are seeing rising rates of developmental, language, and social-emotional delays, and expressed concern that AI toys could worsen that trend.
• U.S. senators wrote to the Consumer Product Safety Commission in January 2026 asking what safety testing, if any, AI toys undergo before reaching market.

Read together: there is real professional concern, an active regulatory conversation, and almost no long-term developmental research. That is a reason for caution, not certainty in either direction.

So should you throw the bear away?

Not necessarily. But it helps to reclassify it. An AI toy is a toy, not a therapy tool and not a conversation partner. Treated as one thing among many, played with in a shared space where you can hear it, it’s unlikely to harm a child who’s also getting plenty of real interaction. The risk is substitution. Every hour spent in a pseudo-conversation that always agrees, never misunderstands in a repairable way, and never needs anything back is an hour not spent in the human version, and it’s the human version the evidence supports. That risk is highest for exactly the children whose parents are most tempted, and kids who are already talking less than their peers.

What the evidence says actually works

Here the research is much stronger. Roberts and Kaiser’s meta-analysis of 18 studies found that parent-implemented language interventions had a significant positive effect on the receptive and expressive language of children aged 18 to 60 months. A later meta-analysis by Heidlage and colleagues (2020) broadly replicated this, with a significant effect on expressive language (g = 0.42) and expressive vocabulary in both shared book reading and play-based routines, though notably, effects on receptive language were not significant, and effects were larger for children with primary language impairment than for autistic children.

In other words: what parents do matters, measurably. It is not a cure-all, and effects vary by child. These are the behaviors those intervention packages teach:

• Follow their lead. Talk about whatever your child is already looking at or touching. Language attaches best to the thing currently holding their attention.
• Wait longer than feels comfortable. After you say something, count silently to five. Children who are slow to talk often need extra seconds to assemble a response, and most adults fill the silence for them. (This is standard clinical practice within responsive-interaction approaches rather than a separately trialed technique.)
• Say it back, plus one. Child says “car,” you say “big car.” Child says “want juice,” you say “you want apple juice.” You’re modeling the next step rather than correcting the current one.
• Build in reasons to communicate. Crackers in a clear container they can’t open. Two choices offered, then a pause. Small, cheerful opportunities to need words.
• Narrate the ordinary. Bath, diaper, grocery aisle, car seat. Shared book reading and daily routines are the two contexts where the meta-analytic effects showed up.

When to stop wondering and get it checked

Language delay is common. The American Academy of Pediatrics notes that about one in five children will show a developmental delay in speech or language.

“Wait and see” is also genuinely defensible for some children, which is why parents get conflicting advice. The longitudinal literature is mixed: current studies suggest roughly half of children with preschool-onset language delay reach age-appropriate levels by school entry, while the rest continue to have difficulties. Leslie Rescorla’s review of late-talker outcomes found that most late talkers scored within the normal range by age 6 or 7 — but remained significantly weaker than typically developing peers into adolescence.

So “most catch up” is true and also not the whole story. Some signs are worth acting on rather than watching. Using ASHA’s communication milestones and the CDC’s Learn the Signs. Act Early. checklists as a starting point:

• No babbling by around 12 months
• No single words by 16 months
• No two-word combinations by 24 months
• Loss of words or skills previously acquired, at any age
• Not responding to their name consistently
• Difficulty following simple directions without gestures

That last one matters more than parents usually realize. Understanding, or receptive language, is the foundation under speaking. A child who talks little but understands everything is in a different situation than one who’s behind in both.

If any of those describe your child, the next step is an evaluation, not another six months of waiting. Early intervention services are typically free or low-cost for children under three. A comprehensive evaluation by a speech-language pathologist will tell you either that things are on track (reassurance no milestone chart can give you), or exactly where to start. No toy, however clever, can tell you that.

At Child & Family Development, Jarred Dutka, CCC-SLP provides speech and language evaluations and therapy for children of all ages. Schedule a free phone consultation with Jarred if you’d like to talk more about your child’s language development.

Sources

AI toys

Goodacre, E. & Gibson, J. (2026). AI in the Early Years, University of Cambridge PEDAL Centre. Project page · University summary

Spichak, S. (2026). Policymakers and Researchers Zero In On the Impact of AI Toys. Journal of Medical Internet Research, 28. Link

Dezeen coverage of the Cambridge observations (March 2026). Link

Fairplay et al. (2025). AI Toys Are Not Safe for Kids advisory. Announcement · PDF

U.S. Senate Commerce Committee letter to the CPSC (January 2026). Link

Conversation and language development

Romeo, R. R., et al. (2018). Beyond the 30-Million-Word Gap: Children’s Conversational Exposure Is Associated With Language-Related Brain Function. Psychological Science, 29(5), 700–710. Open access

Harvard Graduate School of Education summary of the above. Link

Parent-implemented intervention

Roberts, M. Y. & Kaiser, A. P. (2011). The Effectiveness of Parent-Implemented Language Interventions: A Meta-Analysis. American Journal of Speech-Language Pathology, 20(3), 180–199. Abstract

Heidlage, J. K., et al. (2020). The effects of parent-implemented language interventions on child linguistic outcomes: A meta-analysis. Early Childhood Research Quarterly, 50, 6–23. Link

Late talkers and milestones

Rescorla, L. (2013). Late Talkers: Do Good Predictors of Outcome Exist? Developmental Disabilities Research Reviews. PDF

Language outcomes and influencing factors in children with language delay: a retrospective cohort analysis (2025). Scientific Reports. Link

ASHA Developmental Milestones: Birth to 5 Years. Link

CDC, Learn the Signs. Act Early. Link

American Academy of Pediatrics, Language Delays in Toddlers. Link 

This article is for general information and isn’t a substitute for individualized advice from your pediatrician or a licensed speech-language pathologist.