Read a sample: 1. Start With the Learning, Not the Tool
A teacher assigns a paragraph explaining why a character makes a difficult choice. One student asks ChatGPT for possible interpretations, chooses one, and finds evidence in the text. Another asks it to write the paragraph, changes a few words, and submits the result. A third writes a draft independently and asks for feedback on whether the explanation is clear.
All three students used the same kind of tool. Whether their use supported or displaced the learning depends on what the teacher wanted the assignment to accomplish.
If the goal was to practice finding textual evidence, asking for possible interpretations might be acceptable, provided the student still located and evaluated the evidence. If the goal was to demonstrate an independent interpretation, even that initial suggestion might interfere with the evidence the teacher hoped to collect. Feedback on clarity could be useful for a revision exercise, but less appropriate during a final assessment of what a student can produce without assistance.
This is why a useful classroom guideline begins with a question about learning rather than a blanket decision about technology:
What do students need to know, practice, or demonstrate in this activity?
Once that question has an answer, a teacher can make a more defensible decision about where assistance belongs. The aim is not to classify every interaction with ChatGPT as inherently helpful or harmful. It is to decide whether a particular use leaves the essential thinking with the student and gives the teacher the evidence needed for the instructional purpose.
Classroom tasks often combine several goals. A research presentation, for example, may ask students to locate sources, judge their usefulness, organize information, speak clearly, and answer questions. A math assignment may involve recalling a procedure, choosing a strategy, performing calculations, and explaining an answer. If directions simply say “complete the presentation” or “solve the problems,” students may not know which parts are intended as practice and which kinds of help are acceptable.
Start by identifying the learning that matters most for this activity. A sentence frame can help:
Students will practice or demonstrate the ability to by .
For a writing lesson, the sentence might become: “Students will practice developing a claim by writing a paragraph that connects an interpretation to evidence from a story.” For an introductory science course: “Students will demonstrate their ability to distinguish an observation from an inference by analyzing a short investigation.”
These statements do not settle every question about AI. They do make the central work visible. If students are supposed to develop a claim, a tool that supplies the claim may take over a meaningful part of the assignment. If they are supposed to distinguish observation from inference, a tool that labels every statement for them may remove the very decision the teacher needs to see.
The same task can serve different purposes on different days. Early in a unit, students might benefit from discussing possible claims before choosing one. Later, the teacher may need to know whether each student can develop a claim independently. The assignment may look similar, but the role of assistance can reasonably change.
That distinction is worth explaining to students. Otherwise, a change in the rules may feel arbitrary: “We could use it yesterday, so why not today?” The answer is not that the tool has suddenly become good or bad. The answer is that yesterday’s activity was designed for supported practice, while today’s activity is designed to show what each student can do independently.
A learning goal describes what students are working toward. Evidence is what the teacher can observe to judge their progress.
Suppose the goal is to solve a multistep equation. A correct final answer is one piece of evidence, but it may not reveal whether a student selected the right operations, understood why those operations were valid, or recognized an error along the way. Depending on the purpose, the teacher might also ask for written steps, an explanation of one decision, or a response to a similar problem completed without assistance.
Now consider a student who enters the equation into ChatGPT and copies the worked solution. The page may contain correct steps, but it gives the teacher limited evidence of the student’s own reasoning. The problem is not merely that an outside tool was involved. It is that the submitted work no longer clearly shows the ability the assignment was intended to reveal.
By contrast, a student might attempt the problem first, then ask for a hint about the next step and annotate where the hint was used. That process can produce useful learning evidence in a practice setting: the original attempt shows the student’s starting point, and the revision may show how the student responded to support. It would not necessarily provide the same evidence as an independent assessment.
Before deciding on an AI boundary, ask:
- What part of the student’s work must come from the student for me to interpret it meaningfully?
- What assistance would still leave that work visible?
- What assistance would make the finished product hard to interpret?
- Do I need evidence of the process, the final product, or both?
- Is this an opportunity to learn with support, or an occasion to demonstrate current independent performance?
The answers may be brief. A teacher does not need a lengthy analysis for every worksheet. Even a few moments of deliberate planning can turn a vague rule into…