Generative AI has moved into school life faster than many classrooms were prepared to absorb it. Chatbots can answer homework questions, draft essays, and offer instant help from a student’s phone, leaving teachers to decide when the technology supports learning and when it gets in the way.
At Cheshire Academy, a private boarding and day school in Connecticut, the response has not been a single mandated tool or a blanket ban. The school is using a mix of teacher training, classroom experiments, and simple assignment labels to make AI expectations visible.
A school choosing guidance over mandates
Cheshire Academy has about 400 students in grades 9 through 12. Administrators there do not require instructors to use AI, but the school’s librarian and technology coordinator George Aiello says the “vast majority” of instructors use it in some way.
That use is intentionally varied. Teachers are working with general-purpose chatbots such as ChatGPT and Perplexity, while some are also trying MagicSchool, a platform built specifically for educators. The result is a patchwork of tools rather than a single approved system.
The school took that route after consultants advised it to train staff in general AI techniques instead of prescribing specific products. Training covered how to write useful prompts, but it also emphasized limits. Staff were warned about incorrect and biased responses, a key concern when AI output is used near classroom material.
That approach gives teachers room to choose whether AI fits their work. It also reflects the uneven reality of classroom adoption: some educators see AI as useful for planning and preparation, while others remain cautious about putting AI-generated text directly in front of students.
How teachers are using AI behind the scenes
At Cheshire Academy, many teachers use generative AI to help prepare class materials. That can include asking a chatbot to assist with lesson planning or to create grading rubrics.
Some teachers are interested in using AI to help provide feedback to students. But that has not yet happened at the school because of concerns about quality, personalization, and privacy. Those concerns matter because feedback is not only about whether an answer is correct. It also reflects a teacher’s knowledge of a student, the assignment, and the learning goals.
MagicSchool is one of the tools available to staff. Its appeal is the number of education-focused functions it puts in one platform. It can generate questions and assignments, including quizzes and worksheets, across many subjects and grade levels. It also includes a grading rubric generator that produces a table teachers can use to score assignments.
The platform can also create presentations, lesson plans, and administrative reports. Teachers enter task-specific prompts. For an assignment, they can specify grade level, number of questions, question types such as multiple choice or short answer, and documents that the questions should align with.
MagicSchool has free and paid versions. Teachers who want unlimited access and complete records in the system need to pay just under $100 per year for an individual plan. At the same time, many Cheshire Academy teachers use general-purpose AI tools instead, including chatbots from Anthropic, Google, OpenAI, and others, especially for administrative tasks.
Teaching students to examine AI, not just use it
French teacher Miriam Przybyla-Baum shows another side of the school’s approach. She does not use AI herself, saying she has built up enough classroom materials across nearly 30 years of teaching. But she does address AI directly with students.
Her experience with AI-assisted shortcuts began before ChatGPT’s launch. Students were already using tools such as Google Translate on assignments, which raised questions about what the tools could and could not teach them about language.
Instead of treating every AI-assisted edit as automatically helpful, Przybyla-Baum asks students to evaluate the machine’s work. In one assignment, students let a large language model edit their homework. Then they review the changes and decide which edits were correct and which ones removed their voice.
In another exercise, students anonymously grade each other’s AI-assisted assignments. They add annotations identifying which parts they think were supported by AI. The point is not simply detection. It is reflection: students have to consider how AI changes the work and what learning may be lost when a tool does too much.
The traffic-light policy makes expectations visible
Cheshire Academy has adopted a schoolwide version of this reflective approach through assignment labels modeled on traffic lights. The system gives students a quick signal about what kind of AI use is allowed.
- Green means AI is fully allowed for the assignment.
- Yellow means the teacher permits some tools while banning others, such as allowing spell-check but not a chatbot.
- Red means no AI use is allowed.
The strength of this system is its clarity. It does not require every assignment to follow the same AI rule. A writing task, a language exercise, a quiz, and a planning activity can each carry different expectations, and students can see those expectations before they begin.
It also gives teachers a practical way to talk about AI without turning every assignment into a debate. The label becomes part of the instructions. Students know whether the task is meant to measure unaided work, allow limited support, or invite broader experimentation.
Why the classroom question remains unsettled
The broader challenge is that schools are still adapting to a technology that arrived quickly. Teachers were already balancing lesson planning, homework, and grading before generative AI added a new layer of decisions. Even when teachers can spot signs of AI-generated text, that does not solve the larger question of how the tools should fit into learning.
Organizations including OpenAI and UNESCO encourage AI use in the classroom, but many teachers remain unsure how to handle it. Cheshire Academy’s example suggests one practical path: train educators on the technology’s limits, let teachers choose tools that fit their needs, and make student-facing rules explicit.
The school is continuing to experiment. It is piloting a Student AI Council in which students create media and lead discussions about healthy AI use. The goal is to push students to consider when AI benefits the community around them and when it should be held back.
For schools facing the same uncertainty, the lesson is less about any one platform and more about structure. AI in the classroom becomes easier to manage when teachers have room to experiment, students are asked to think critically, and assignments clearly state what kind of help is allowed.