AI Storytelling With Young Children: Why Co-Use Beats Solo Use

A five-year-old asks the tablet for a story about a dragon who is scared of the dark. The voice answers instantly, warmly, and with a plot twist. The question for parents is not whether AI storytelling with young children is good or bad. It is who else is in the room. A small study posted in December 2025 put sensors on the heads of 23 kids aged 5 and 6 while they heard stories from an AI chatbot, from a parent, and from both together. Their brains did not respond the same way across those sessions, and the children who saw the AI as most “perceptive” showed the biggest difference. The practical rule that falls out of this, and out of 20 years of earlier research, is short: use it with them, not instead of you.

Key Takeaways

• A 2025 preprint (Kim et al., arXiv, not yet peer-reviewed) found that 5- and 6-year-olds who credited an AI chatbot with more “perceptive” abilities showed more activity in a social-reasoning brain region when alone with it, and less when a parent was present.
• In exploratory analyses, higher brain activity during the AI-only session went with more children reporting feeling “scared” afterward. The sample was 23 children, so treat this as a signal, not a verdict.
• Older, peer-reviewed research is consistent: babies and toddlers learn language from live, responsive interaction far better than from recordings (Kuhl, 2003; Roseberry, 2014).
• AI is not useless for young kids. A 2022 randomized study found a conversational agent could match a human partner’s dialogic-reading benefits for story comprehension in 3- to 6-year-olds.
• The safest default for ages 5 and under is co-use: an adult sitting alongside, talking about the story, and ending the session when it stops being fun.

What the AI Storyteller Study Actually Tested

The study is a preprint titled “Young children’s anthropomorphism of an AI chatbot: Brain activation and the role of parent co-presence,” posted to arXiv on December 1, 2025 by Pilyoung Kim, Jenna H. Chin, Yun Xie, Nolan Brady, Tom Yeh and Sujin Yang. A preprint has not gone through peer review, and that matters for how much weight it can carry.

Twenty-three children aged 5 to 6 did three storytelling sessions: with an AI chatbot alone, with a parent alone, and with the chatbot and the parent together. Afterward the researchers asked the children how human-like each storyteller seemed. That tendency to give human qualities to something that is not human is called anthropomorphism.

The kids rated their parents as more human-like overall, which is reassuring. But they gave the chatbot fairly high marks on two specific things: its perceptive abilities (can it see and hear me?) and its epistemic states (does it know things?). Those beliefs did not line up with how much the children talked. Kids who thought the AI was very perceptive did not chat with it more, as measured by conversational turns.

The interesting part was in the brain data.

How the Brain Measurements Work, and What They Showed

The team used fNIRS, functional near-infrared spectroscopy. A soft cap shines harmless near-infrared light through the scalp. Oxygen-rich and oxygen-poor blood absorb that light differently, so the sensors can estimate when blood flow increases in the outer layer of the brain. More blood flow is a proxy for more neural work. It is quiet, it tolerates wiggling, and children can sit at a table while wearing it, which is why it is popular for studies of young kids. It cannot see deep structures, and it measures blood flow, not thoughts.

The region that stood out was the right dorsomedial prefrontal cortex (dmPFC), an area near the top front of the head that researchers associate with thinking about other minds: what someone knows, wants or intends.

Here is the pattern the authors report:

• Children who rated the chatbot as more perceptive showed more dmPFC activation during the AI-only session.
• The same children showed less dmPFC activation when a parent was there with the AI.
• In exploratory analyses, higher dmPFC activation during the AI-only session was associated with more “scared” mood ratings at the end of that session.

The authors’ interpretation is that a parent’s presence may help some children make sense of, and regulate their reaction to, a new kind of conversational partner. A plain-language version: a child who half-believes the tablet can see her may be working hard to figure out what kind of thing she is talking to. With a parent next to her, she can borrow the parent’s calm instead of solving that puzzle alone.

What this study cannot tell you

Be strict about the limits. Twenty-three children is small. The “scared” link was exploratory, which means the researchers were not testing it as a main hypothesis. One session per condition says nothing about what happens after weeks of nightly use. And more brain activation is not automatically bad; it can mean effort, curiosity or confusion. The study did not measure learning, sleep, or anything about long-term development.

So the honest takeaway is narrow. In this group, the same AI was a different experience for a child alone than for a child sitting next to a parent.

Why Live People Still Matter: Two Decades of Evidence

The preprint lands on top of a much older, peer-reviewed body of research about how young children learn from people versus screens.

In 2003, Patricia Kuhl, Feng-Ming Tsao and Huei-Mei Liu published a study in the Proceedings of the National Academy of Sciences on 9-month-old American babies. Infants who spent 12 lab sessions with live Mandarin speakers kept their ability to tell apart Mandarin speech sounds, an ability that normally fades between 6 and 12 months. Infants who heard the same speakers and materials through audiovisual or audio-only recordings showed no effect at all. Same words, same voices. The difference was a live person.

In 2014, Sarah Roseberry, Kathy Hirsh-Pasek and Roberta Golinkoff tested 36 toddlers aged 24 to 30 months in Child Development. Children learned new verbs from live interaction and from a live video chat, but not from a prerecorded video of the same session. The key ingredient was social contingency: the adult on the other end responded to what the child did, in time.

That word, contingency, is where AI gets complicated. A chatbot is contingent in a technical sense. It waits for the child, and it answers what she said. That is new. A 2022 randomized study by Ying Xu and colleagues in Child Development assigned 117 children aged 3 to 6 to hear a story read by a human or by a conversational agent, with or without dialogic questions (“What do you think the fox will do next?”). The agent asking questions improved story comprehension about as well as a human asking questions, and kept children’s talk more on topic.

So the research is mixed in a useful way. Responsive AI can teach some things. It does not replace the human relationship the child is reading through, and the Kim preprint suggests that, for some kids, the human beside them changes how the AI itself feels.

AI Storytelling With Young Children: Solo Use vs. Co-Use

The table below pulls the evidence together. It is a decision aid, not a grading of AI.

Setup	What the research suggests	Main risk	Best for
Child alone with an AI storyteller	Kids may attribute “seeing” and “knowing” to the AI; one small preprint links this to more effortful brain activity and some “scared” ratings	Confusion about what the AI is; no adult to catch odd content; no shared memory of the story	Not recommended as a default at 5 and under
Parent and child with an AI storyteller	Parent presence went with lower dmPFC activation in kids who saw the AI as perceptive (Kim et al., 2025)	Parent drifts to their own phone, turning co-use into solo use	Occasional novelty: creating a silly story together
AI that asks dialogic questions	Matched a human questioner for comprehension in 3- to 6-year-olds (Xu et al., 2022)	Works only as well as the questions; commercial apps vary	Practice with story recall, with an adult nearby
Parent reading or telling stories	Live, contingent interaction drives early language learning (Kuhl, 2003; Roseberry, 2014)	None beyond the parent’s energy at 8 p.m.	The everyday default
Recorded audio or video story	Recordings alone produced no phonetic learning in 9-month-olds (Kuhl, 2003)	Passive, no back-and-forth	Car rides, quiet time, not as teaching

The American Academy of Pediatrics reached the same direction in its 2016 policy statement “Media and Young Minds” in Pediatrics. It told pediatricians to emphasize “parent-child shared media use” for children 0 to 5, not only content and time limits. That statement predates chatbots, but its logic travels.

What to Do at Home When Your Kid Wants the AI Story

Sit in the same frame

Co-use means you can see the screen and hear the story as it happens. Put the tablet or speaker where both of you face it. If you notice yourself reaching for your own phone, the session has quietly become solo use. Ten focused minutes beat forty minutes in parallel.

Name what the AI is, out loud, in kid words

A 5-year-old does not need a lecture about language models. She needs one true sentence: “This is a computer program that makes up words that fit together. It can’t see you. It doesn’t know you.” Say it calmly and early. Children who credited the chatbot with perception were the ones whose brains worked hardest alone with it, so clearing up “can it see me?” is a small, direct fix.

Make the child the author

Ask your kid for the hero, the problem and the ending before you ask the AI anything. Then let the AI fill in a middle, and stop to judge it together: “Did the dragon do something that makes sense?” The child stays the storyteller, and the AI becomes a prop. Afterward, retell the story with no device, which is the part that builds language.

Watch the face, not the clock

The preprint’s “scared” finding is a reason to watch for unease: going quiet, asking to stop, wanting to be held, or bedtime fears that mention the voice. End the session the moment the fun drains out. For more on spotting these reactions in AI products, the results of recent AI toy safety tests show what a talking device can say when no adult is listening.

What not to do

Do not use an AI storyteller as the bedtime babysitter for a young child behind a closed door. It is the exact solo setup this research flags, it is the moment kids are most tired and suggestible, and it trades away the most reliable language-building routine families have. Common Sense Media’s January 2026 risk assessment of AI companion toys went further than this article does and recommended no AI toys at all for children 5 and under, after finding 27% of tested toy outputs were inappropriate for kids.

What to Watch For Over the Next 3 Months

• Week 4: Does your child describe the AI as a person (“she told me”, “he’s my friend”)? Gently correct it each time. Does she still ask you for stories, or only the device?
• Month 2 red flags: fear of the device, new nighttime worries tied to it, tantrums when it is turned off, or preferring the AI over reading with a parent. Any one of these is a cue to pause AI storytelling for a few weeks.
• Month 3 self-check: Count how many stories this week were told by a person versus generated. If the generated pile is bigger, rebalance. Ask your child to tell you a story she made up herself, start to finish.

If you want a wider frame for these choices, the research on passive versus interactive screen time explains why back-and-forth matters more than the screen itself, and how voice assistants like Alexa and Siri work covers the mechanics behind the voice.

Frequently Asked Questions

Is AI storytelling bad for young children?

Not in itself. A 2022 randomized study found a conversational agent that asked good questions helped 3- to 6-year-olds understand stories as well as a human questioner did. The concern is solo use. A small 2025 preprint suggests young children experience an AI differently alone than with a parent beside them, so make co-use the default.

At what age can a child use an AI storyteller alone?

No study sets a safe age. Common Sense Media advises no AI toys for children 5 and under and caution for ages 6 to 12. UNESCO’s generative AI guidance for education recommends an age limit for independent conversations with these tools. Practically, wait until your child can explain that the AI cannot see or know them.

My 5-year-old thinks the AI can see her. Should I worry?

It is a normal belief at that age, and the Kim study found many kids rated the chatbot high on perception. It is worth correcting simply and repeatedly: “It’s a program. It can’t see you.” Children who held that belief most strongly showed the most effortful brain activity when alone with the AI.

Is an AI story better than a YouTube story video?

For learning, responsiveness matters. Toddlers learned words from live video chat but not from recorded video (Roseberry, 2014), and an AI can respond in a way a recording cannot. But neither replaces a parent. If you choose either, watch or listen together and talk about it afterward.

Does this study prove AI scares kids?

No. The “scared” association was exploratory, in 23 children, in a preprint without peer review. It is a reason to watch your own child’s reactions, not a finding that AI frightens children. Larger, replicated studies are needed before anyone can say that.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Kim, P., Chin, J. H., Xie, Y., Brady, N., Yeh, T., & Yang, S. (2025). “Young children’s anthropomorphism of an AI chatbot: Brain activation and the role of parent co-presence.” arXiv (preprint, not peer-reviewed). https://arxiv.org/abs/2512.02179
2. Kuhl, P. K., Tsao, F.-M., & Liu, H.-M. (2003). “Foreign-language experience in infancy: Effects of short-term exposure and social interaction on phonetic learning.” Proceedings of the National Academy of Sciences, 100(15). https://doi.org/10.1073/pnas.1532872100
3. Roseberry, S., Hirsh-Pasek, K., & Golinkoff, R. M. (2014). “Skype me! Socially contingent interactions help toddlers learn language.” Child Development, 85(3). https://doi.org/10.1111/cdev.12166
4. Xu, Y., Aubele, J., Vigil, V., Bustamante, A. S., Kim, Y.-S., & Warschauer, M. (2022). “Dialogue with a conversational agent promotes children’s story comprehension via enhancing engagement.” Child Development. https://doi.org/10.1111/cdev.13708
5. American Academy of Pediatrics, Council on Communications and Media. (2016). “Media and Young Minds.” Pediatrics, 138(5). https://doi.org/10.1542/peds.2016-2591
6. Common Sense Media. (2026, updated January 22). “AI Toys: Risk Assessment.” https://institute.commonsensemedia.org/risk-assessments/ai-toys
7. UNESCO. (2023). “Guidance for generative AI in education and research.” https://www.unesco.org/en/articles/guidance-generative-ai-education-and-research