Resources for teachers
AI without outsourcing student thinking
A practical way to decide where AI can remove needless barriers and where students need to do the intellectual work themselves.
Start with the classroom problem
A polished answer can hide the work that matters
Teachers are often asked to respond to work that looks finished but gives little information about how the student got there. A fluent paragraph may not show how the student interpreted the task, selected an idea, weighed evidence, or decided what to change. An answer can be correct and still leave important questions about understanding.
Open-ended AI can make that gap wider. It can produce a plan, a claim, an explanation, or a complete response before a student has had a chance to make those choices. The useful question is not whether a student used a tool. It is which parts of the learning task the tool supported and which parts the student still had to do.
This is a planning principle, not a detection method. A teacher still needs context, conversation, and professional judgment.
Set the boundary before the task
AI may help with access. Students make the consequential decisions.
Students have always used tools that make learning more accessible: examples, notes, read-aloud support, language supports, calculators, and feedback from another person. AI can sometimes play a similar supporting role when its use is visible, bounded, and allowed by the teacher for that assignment.
| AI may help with | The student still decides |
|---|---|
| Clarifying a direction, defining an unfamiliar word, or offering another way into a difficult concept. | What the task is asking, what question to pursue, and which constraints matter. |
| Pointing out a place where an explanation needs more detail or asking a focused revision question. | Which ideas and evidence are relevant, what claim to make, and how the reasoning supports it. |
| Helping a student notice a pattern in their own draft or practice a skill with a new example. | Whether the feedback is sound, what to revise, and when the work is ready to submit. |
The permitted boundary depends on the learning goal and the assignment. Teachers set the boundary, and students should see it before they begin.
A simple design test
Remove barriers. Preserve the thinking. Build independence.
Remove barriers
Reduce confusing directions, inaccessible formats, long waits for feedback, and unsupported failure. Good support gives a student a way to enter the task without deciding the answer for them.
Preserve the thinking
Keep interpretation, idea generation, evidence selection, judgment, explanation, and deliberate revision with the student when those moves are the goal. A question or next step can support the work without replacing it.
Build independence
Plan for students to explain what they changed, recognize an error, transfer an idea to a new task, and work with less support later. More tool use is not the goal.
Before, during, and after
A planning checklist for a bounded AI task
-
Name the learning goal
Write down the knowledge or reasoning students need to practice. If a tool can perform that exact move, decide whether it is out of bounds for this task.
-
Mark permitted support
Tell students what kind of help is allowed, what information the tool may use, and what it must not do. Keep the explanation short enough to read before starting.
-
Ask for an attempt first
Give students a chance to interpret the question, make an initial choice, or explain their current thinking before substantive help appears, when the learning goal and approved accommodations allow it.
-
Make response to feedback visible
Invite students to say what feedback they considered, what they changed, and why. The response remains their decision; accepting feedback is not the same as understanding it.
-
Plan a later independent check
Use a comparable but not identical task later so students have to retrieve, explain, and apply the idea with equal or reduced support.
Keep the claims modest
What this approach cannot tell you by itself
A process record can help a teacher notice what happened inside one task. It cannot tell the whole story of a student’s learning, and it cannot establish authorship, intent, misconduct, ability, emotional state, mastery, or a grade. Those judgments require context and teacher judgment.
Coach is a developing response to this problem. Its intended direction is bounded support around attempts, explanations, feedback, revision, and retry. The public pages do not claim validated learning outcomes or a finished answer to AI and assessment.
Start small: choose one learning goal, define one useful kind of support, and decide what students should still be able to explain without it.