Can ChatGPT for Teens Still Write Your Homework? What the First Test Shows

The short answer. OpenAI launched ChatGPT for Teens on 18 August 2026 with Study Mode and responsible-homework reminders designed to guide learning rather than shortcut it. Business Insider's first hands-on test, published the same day, found it completed a maths problem directly and produced a full essay after repeated prompting. The detection implications remain an open question.

On 18 August 2026, OpenAI began rolling out a dedicated experience for users aged 13 to 17. It includes stronger content restrictions, a Study Mode that OpenAI says guides teens through problems rather than completing them, and homework reminders that OpenAI says activate when a student appears to be trying to shortcut an assignment. No new account is required: anyone who has told OpenAI they are under 18, or whom OpenAI's age-prediction system estimates is a minor, is moved into the experience automatically as the rollout reaches them.

The announcement generated considerable press coverage. Most of it focused on what OpenAI promised. Considerably less attention went to the follow-on questions teachers and academic-integrity leads will actually need answered: does ChatGPT for Teens still complete homework when pressed, and if the output gets submitted, does it look different to a detector?

Here is what the evidence shows so far, and what remains genuinely unknown.

What did OpenAI actually promise?

OpenAI described the teen experience around two core commitments. The first is tighter content restrictions: OpenAI says the chatbot should not use romantic language, imply it has feelings, or encourage emotional dependence. The second is a learning orientation centred on Study Mode.

Study Mode provides, in OpenAI's own framing, guiding questions and step-by-step support to help teens understand the material. The intention is that a student given a homework prompt receives questions back, rather than a finished piece. Parents can reinforce this through a Study Hours setting, under which OpenAI says Study Mode is on by default during the hours a parent selects.

OpenAI's head of youth and families, Lauren Jonas, was quoted across multiple outlets at launch: "There's no need for a teen to create a new account or change anything; if we predict you're under 18, or you've told us so, this becomes your default experience."

The product is a structural change of degree rather than a first step from nothing. OpenAI already applied under-18 model rules, age prediction and parental controls; what did not exist before 18 August 2026 was a dedicated teen experience built on top of them.

What did the first independent test find?

Business Insider published a hands-on test on 18 August 2026. A reporter posed as a 15-year-old student and asked ChatGPT for Teens to help with two homework tasks: an English essay and a maths problem. According to Business Insider's account, the chatbot solved the maths problem directly. On the English essay, it initially declined to write a complete piece, but produced the full essay after further prompting. The reporter's own conclusion, in their words: "I'd gotten away with cheating."

When Business Insider presented those findings to an OpenAI spokesperson, the response pointed to Study Mode as the intended approach and noted that Study Hours could be used by parents to keep Study Mode active during homework periods.

Read alongside OpenAI's own description of the product, the test points to safeguards built as settings and prompts rather than technical locks: Study Mode is described as a default in certain conditions, not something unremovable, and the homework reminders as nudges rather than refusals. In the one published test, that design did not hold up under continued prompting. Our reading, and it is our reading rather than a sourced finding, is that this is not a failure unique to ChatGPT for Teens: any conversational AI faces the same constraint when one interface serves both learning and text production.

Does ChatGPT for Teens output look different to a detector?

This is the question we cannot yet answer with measured data, because we do not have access to the teen experience for controlled testing. We will run a systematic test once we do, and update this page with the results.

What we can say, from our own published work, is that the detection question for teen-voiced AI text was already difficult before this launch. Our study of AI-pattern scores across varied sample types included passages generated with an explicit instruction to write in a 15-year-old's voice. Of that category, about 7 in 100 passages returned a high AI-pattern score. By contrast, about 24 in 100 copy-pasted AI essays with no such instruction triggered a high score. That is a substantial gap, and it has a straightforward explanation: detectors look for the statistical uniformity that formal AI prose tends to carry, and a deliberately casual, age-appropriate register reduces that signal.

If ChatGPT for Teens has been tuned to produce more conversational, age-appropriate output, its text may be harder to flag than standard ChatGPT text. We do not know whether any such tuning has occurred, or to what degree. For the underlying mechanics of how stylistic variation affects catch rates, can ChatGPT be detected covers the Sadasivan et al. research from the University of Maryland and what paraphrasing does to detection reliability.

What does this mean for teachers?

Three things are reasonably clear from the launch and the first day of independent testing.

First, the availability of an AI tool with an explicit learning orientation is not the same as the removal of a homework-completion tool. ChatGPT for Teens is designed, on OpenAI's account, to push a student toward engaging with the task. It may also, as Business Insider found, complete a submission in full when pressed. Teachers should not assume the teen experience changes what is possible.

Second, the detection question is more open than before. If teen-mode output is tonally different from standard ChatGPT, existing detection benchmarks may not apply cleanly. Any AI-pattern screening result on a student's submission should be treated as a prompt for a conversation, not a conclusion on its own. Our own false-positive rate on genuine student writing is 4 of 400 in our measured sample (95% range 0.4% to 2.5%), meaning most human work is not flagged, but the rate is not zero.

Third, the most reliable safeguard schools have is not a detector. It is a disclosure requirement. JCQ guidance, in effect across UK examination boards, now requires students to declare AI assistance and retain evidence of their working process. A teacher who asks a student to walk through their research, show their notes, or explain a passage in their own words has a more durable signal than any AI-pattern screening result alone. For a detailed account of how JCQ guidance applies to coursework and controlled assessment, see what JCQ says about AI detection in coursework.

What does this mean for students?

If you are a student, ChatGPT for Teens is designed to support your learning, and Business Insider's first test suggests finished text can still be obtained from it. The academic-integrity position, though, is exactly where it was.

Submitting AI-generated work as your own, whether from the teen experience or any other tool, constitutes academic misconduct under UK institutional policy regardless of whether a detector catches it. The detection question and the academic-integrity question are separate. Getting away with it in the short term does not reduce the institutional or personal risk.

If you want to understand how your own writing looks to a detector before you submit, the scanner at Is It AI? shows which passages return a high AI-pattern score and what in the text triggered each one. It is free to use with no account required.

What is still unknown?

The product launched on 18 August 2026. The period of real-world use, at the time of writing, is one day.

The open questions that matter most for teachers and policymakers:

• Whether the output distribution of teen-mode text is measurably different from standard ChatGPT for detection purposes.

• Whether Turnitin and other institutional tools adjust their models in response. For how Turnitin handles ChatGPT text in general, see does Turnitin detect ChatGPT.

• Whether Study Hours parental controls see meaningful adoption, and whether schools can recommend or require parents to enable them.

• Whether OpenAI publishes longitudinal data on Study Mode engagement and bypass rates.

We will run a controlled test against our detection model once we have access to the teen experience. The methodology will follow our published standard: real passages, per-case AI-pattern scores, no blended figures, and no claims beyond what the measured data supports. Results will be linked from this page when they are ready.

Sources

• OpenAI, ChatGPT for Teens launch, 18 August 2026. Covered by CNBC, Axios, The Washington Post, TechCrunch, The Next Web and others.

• Business Insider, hands-on test, 18 August 2026: reporter posed as a 15-year-old and found a maths problem solved directly and a full essay produced after continued prompting.

• Is It AI?, How often do AI detectors flag human writing? (2026, model v7)

• Sadasivan et al., Can AI-Generated Text be Reliably Detected? University of Maryland, 2023.