US Teens and ChatGPT for Schoolwork: What the Data Actually Shows

Quick answer

Two of the most cited US studies on teens and AI point the same direction. Pew Research Center found the share of US teens aged 13 to 17 who say they've used ChatGPT for schoolwork doubled from 13% in 2023 to 26% in 2024. RAND's American Youth Panel research found homework AI use climbed from 48% to 62% of students in a year - and 67% of those students said the habit harms their own critical thinking. Together, the numbers describe rising, largely unstructured use, alongside a majority of those students admitting it worries them. The conclusion isn't that AI is dangerous. It's that the habit built around AI use, not the raw amount, determines the outcome.

Key takeaways

• Pew Research Center's survey-based data shows US teen ChatGPT use for schoolwork doubling from 13% to 26% between 2023 and 2024.
• RAND's youth panel research, tracked over time rather than by single survey, found homework AI use climbing from 48% to 62% of students in a year.
• 67% of AI-using students in RAND's research said the habit harms critical thinking, a concern reported by 75% of girls and 59% of boys who use AI for schoolwork.
• The two data sets measure different things - Pew asks about ChatGPT specifically, RAND about AI use on homework broadly - so they should be read as corroborating trends, not identical figures.
• None of the data distinguishes "used AI" from "let AI do the thinking," which is exactly the distinction parents need to make at home.

Why this data matters for your family

Numbers this size stop being an abstract trend and start describing what is very likely already happening in your own house. If roughly a quarter of US teens report using ChatGPT for schoolwork, and well over half report broader AI use on homework, the realistic starting assumption for most families is not "does my teenager use AI" but "how does my teenager use AI, and does anyone know." The RAND finding that most AI-using students themselves worry about the effect on their thinking is the detail worth sitting with. It means the conversation about AI and schoolwork is not one parents need to introduce from scratch. Many teenagers have already had it with themselves, silently, and are simply waiting for an adult to ask.

What "AI use for schoolwork" actually measures

"AI use for schoolwork" is a broad survey category that covers everything from asking a chatbot to explain a confusing concept, to generating practice questions, to having it draft an entire essay unsupervised. Pew's and RAND's numbers count all of that as "use," which is exactly why the figures should not be read as a cheating rate. A student who used ChatGPT once to check a math answer and a student who used it to write a whole lab report both count identically in the topline statistic. The number that matters for a parent is not in either study - it's the answer to "what did you actually use it for," asked directly and often.

Reading the two studies side by side

Reading them together rather than separately is the useful move. Pew shows the specific behaviour accelerating; RAND shows the broader behaviour is even more common, and that the students doing it are not blind to the risk. Heavy AI reliance is separately associated with cognitive offloading and weaker critical thinking in peer-reviewed research, including Gerlich's 2025 study in the journal Societies involving 666 participants, with the effect strongest in the 17-to-25 age bracket - the exact years these students are entering.

What the gender gap likely reflects

The gap between girls (75%) and boys (59%) reporting that AI harms their critical thinking is one of the more specific findings in the RAND data, and it is worth resisting the urge to over-interpret it. It does not mean girls are more affected by AI use - the data doesn't measure actual cognitive effect by gender at all, only self-reported concern. What it more plausibly reflects is a difference in self-awareness or willingness to name the worry out loud. Either way, the practical takeaway for parents is the same regardless of a teenager's gender: ask the question directly rather than assume silence means the concern isn't there.

Practical examples

• A parent sees their fourteen-year-old's search history full of ChatGPT queries and assumes the worst; asking "what did you use it for this week" reveals mostly concept explanations and one shortcut essay - two very different things worth two very different responses.
• A student who fits RAND's "worried but still using it" profile admits she uses AI for math homework because she's behind, not because she doesn't care about learning it; the fix is closing the underlying gap directly, not banning the tool.
• A family builds one habit from the Pew and RAND numbers alone: after any AI-assisted homework, the student explains the answer out loud without looking at the screen, converting passive output into something they actually own.

Common mistakes when reading this data

• Treating "used AI" as "cheated." The survey categories are far broader than academic dishonesty, and conflating them leads to overreaction.
• Ignoring the self-reported worry. A majority of AI-using students flagging the risk themselves is a rare, useful signal - dismissing it wastes the data's best insight.
• Assuming the gender gap reflects actual differing harm. It reflects differing self-report, which is a different and more limited claim.
• Reading one study in isolation. Pew and RAND measure different things; the trend is stronger evidence than either figure alone.
• Responding to the data with a blanket ban. The research points to habit and verification as the lever, not access.

How the Edison Method applies

Understand: Students first learn what a chatbot is actually doing when it answers a homework question, so "AI got it wrong" stops being a mystery and starts being expected.

Use: Structured practice turns ChatGPT from a shortcut into a study tool - generating practice problems, explaining a concept three different ways, checking working.

Evaluate: Every AI-assisted answer gets checked against a textbook, a teacher, or a worked example before it's trusted.

Build: Real projects, not just chat transcripts, are what convert AI use into evidence of capability a student can show someone else.

Lead: A student who can explain how they used AI and why is already ahead of the disclosure conversation most schools haven't formalised yet.

For the broader picture of how US high schools are handling this shift, see AI education for high school students in the US, and for the wider international picture of programs and evidence, AI education for teenagers in Australia.

The recommendation: use these numbers as a prompt for a specific conversation, not a verdict on your teenager. Ask what they actually used AI for this week, ask how they checked it, and if the honest answer is "I didn't," that's the habit to fix - not the tool to ban. The data says most AI-using teens already sense something is off. Your job is to give that instinct somewhere useful to go.