ChatGPT is a large language model that generates text by predicting the next word, trained on a huge body of internet text. Kids use it for homework, conversation, image generation, and code. There are three real risks (hallucination, over-reliance, content exposure) and three real uses (research starting point, draft-then-revise, conversation partner for stuck problems). The last section is the conversation script: what to ask your kid about their use of it.
What ChatGPT actually is, in one paragraph.
ChatGPT is a chat interface to a family of large language models built by OpenAI. The model is a neural network trained on a huge body of text scraped and licensed from across the internet, plus additional rounds of human feedback that shaped its responses. When you type a message, the model generates a reply one token at a time, choosing each next token based on the patterns it learned. It doesn’t look anything up. It doesn’t consult a database. It generates plausible-sounding continuations of the prompt you gave it, based on what kinds of text usually follow what kinds of prompts.
You don’t need to dumb this down for a kid. They can hold the idea that the model is good at sounding right and that sounding right is not the same as being right. That single distinction is most of the AI-literacy curriculum, and Duri Long and Brian Magerko’s 2020 framework at CHI made it the second of their five competencies.1
The other thing worth knowing: ChatGPT is one product in a category. Anthropic’s Claude, Google’s Gemini, Meta’s Llama, and a few others share the same basic shape. When you hear someone say “AI” in 2026, they almost always mean a large language model in the same shape as ChatGPT. The category is roughly three years old in its consumer form and changing fast.
How kids actually use it.
Reports from the Pew Research Center and Common Sense Media in 2023 and 2024 documented that a substantial share of US teenagers had used ChatGPT for schoolwork within months of its consumer release, with adoption climbing in subsequent surveys.2 We’re not in a position to pin exact percentages and the numbers move quickly, but the qualitative shape is well-established: if your kid is school-age and online, the probability they’ve used ChatGPT is high.
The patterns of use fall into a few buckets. Most parents won’t have seen all of them, so it’s worth listing.
Homework. The most common and the most worried-about. Kids paste a question or a writing prompt into ChatGPT and use the result either to understand a topic, to draft an answer, or, less helpfully, to just turn in. Schools have responded across the spectrum from outright bans to required disclosure to active integration.
Conversation. Some kids treat ChatGPT as a thinking partner. They ask it to explain something they didn’t understand in class, to debate a position with them, to help them work through a feeling. The shape of this use varies enormously by kid and by topic.
Image generation. Through DALL-E inside ChatGPT or through standalone tools like Midjourney, kids generate images for fun, for assignments, or to illustrate stories they’re writing. This is also where the most explicit content exposure risks live, and the platforms have filters of varying quality.
Code and creative projects. Some kids, especially older ones, use ChatGPT to write small programs, scripts, or game prototypes. This is closest to what we build the studio for, but ChatGPT alone is a thinner version of the experience because the kid never sees what their code is doing in context.
The three real risks.
There are a lot of imaginary risks in circulation and a few real ones. The real ones are worth being precise about.
Hallucination. The model generates fluent text that is sometimes confidently wrong. It invents citations. It misremembers facts. It produces plausible-sounding answers to questions it doesn’t actually have the information to answer. The research community has documented this pattern across language models for years and it has not been solved.3 A kid who treats ChatGPT output as authoritative is going to be misled some of the time, and the polish of the prose hides the failure. The pedagogical mitigation is to teach the kid to check, but most consumer AI products are wired to discourage checking by smoothing over rough edges.
The model is good at sounding right. Sounding right is not the same as being right. That single distinction is most of the AI-literacy curriculum. On the one idea every kid using ChatGPT needs to internalize
Over-reliance. The deeper risk is cognitive offloading. A kid who has ChatGPT write every essay, summarize every reading, and answer every hard question builds thinner writing muscles, thinner reading muscles, thinner endurance for the productive struggle that learning actually requires. Stefania Druga’s work at the MIT Media Lab on kids and conversational agents has been documenting how kids form mental models of AI partners, and the patterns are sensitive to how the tool positions itself.4 A tool that positions itself as the answer-giver trains a different kid than one that positions itself as a thinking partner. We’ve written more about the laziness question in Will AI make my kid lazy?.
Content exposure. ChatGPT is general-purpose. It will respond to almost any prompt, with filters that catch the worst categories but inevitably miss edge cases. A kid one curious search away from a topic a parent wouldn’t have chosen is a real shape of risk. The mitigation isn’t a ban; it’s either supervision or a scoped tool whose surface area is smaller. We made the latter choice in our own product, for reasons covered in Is this just ChatGPT with a wrapper?.
The three real uses.
It would be easy to write a piece that just lists the risks and stops there. That’s the takedown shape and we don’t think it’s honest. ChatGPT is a useful tool for several specific things, and a kid who learns to use those well is in a better position than a kid who avoids the tool entirely.
Research starting point. ChatGPT is good at the first step of understanding something unfamiliar. Ask it what a term means, ask for a summary of a topic, ask for the main arguments on both sides of a debate. The kid then verifies through a real source. The pattern that works is use it to get oriented, not to get answers. The orienting is fast. The verifying is the actual learning.
Draft, then revise. Using ChatGPT to produce a rough first draft and then rewriting it in your own voice is a defensible workflow, especially for kids who struggle with the blank page. The trap is using the draft as the final. The pattern that works requires the kid to read the draft critically, change what doesn’t sound like them, push back on what feels wrong. This is the same loop a thoughtful adult writer uses with a research assistant.
Conversation partner for stuck problems. When a kid is stuck on a math problem, a chess position, or a logic puzzle, having a partner who can ask questions and propose moves is genuinely helpful. The pattern that works is the kid driving the conversation, not the kid asking for the answer. “What should I try next?” works. “Just tell me the answer” doesn’t. Mitchel Resnick’s work on creative learning environments has long argued that scaffolded support beats hand-the-answer instruction, and that observation extends to AI partners too.5
What to ask your kid.
The most useful thing a parent can do is have a conversation about how the kid actually uses the tool. Not a lecture. A conversation. Here’s a short script that has worked for us. Pick the questions that sound like you.
“Have you ever used ChatGPT? What for?” This is the opener. You’re not in trouble mode. You’re curious. Let the kid describe their actual use, not the use they think you want to hear.
“Has it ever been wrong about something?” A kid who has used it for a while will have stories. If they don’t, that’s a signal too: they may not be checking. The follow-up is “how would you know?”
“If you used it for a school assignment, did you tell anyone?” This is the disclosure question. It’s not about catching the kid. It’s about teaching that disclosure is a normal part of using powerful tools. A kid who can tell their teacher they used ChatGPT to draft and then revised in their own voice is doing something honest.
“Is there something you’ve used it for that you wouldn’t want a friend to know?” This is the deepest question and it sometimes opens up the conversation about emotional use, which is the under-discussed shape of teen AI use. The American Academy of Pediatrics’ 2016 guidance reminded us that screens work best when they supplement human relationships rather than replace them.6 A kid who is using ChatGPT as a confidant for things they can’t tell anyone needs more than a parental control; they need a person.
The shape of the right parental position, we think, is informed and present. Not banning, not abdicating. Knowing what your kid is doing with the tool and helping them learn to do it well. ChatGPT is going to be in your kid’s life. The version of you that has read this and asked the questions is the version that helps.
References
- Duri Long & Brian Magerko, “What is AI Literacy? Competencies and Design Considerations,” Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems, ACM, 2020. The five-competency framework, including the “knowing what AI can and cannot do” competency.
- Pew Research Center, surveys on teen ChatGPT use, 2023–2024, available at pewresearch.org; and Common Sense Media, The Common Sense Census: Media Use by Tweens and Teens, ongoing series at commonsensemedia.org/research. Both organizations have published repeated measurements of US teen AI adoption.
- Ziwei Ji et al., “Survey of Hallucination in Natural Language Generation,” ACM Computing Surveys, 2023. A widely-cited survey of LLM failure modes.
- Stefania Druga and colleagues, work on the Cognimates platform and on kids’ mental models of conversational AI, MIT Media Lab. See cognimates.me and Druga’s thesis work on AI literacy for children.
- Mitchel Resnick, Lifelong Kindergarten: Cultivating Creativity through Projects, Passion, Peers, and Play, MIT Press, 2017. See also the MIT Media Lab’s Lifelong Kindergarten group at media.mit.edu/groups/lifelong-kindergarten.
- American Academy of Pediatrics Council on Communications and Media, “Media and Young Minds,” Pediatrics, vol. 138, no. 5, November 2016.