Preparing a Lesson Starts With What Students Should Be Able to Do
A lesson on confounds begins with a sentence about what students will do, not with a generated plan. AI help can speed some steps and quietly shape others, so preparation has to record which was which.
Suppose the claim a lesson serves is this: after the session, a student can name a plausible third variable that could explain an association and say how they would check it. That sentence belongs to the instructor, and everything downstream, including the example, the activity and the follow-up, is a means to it. Preparing with AI help goes wrong when the order flips, so that a generated plan arrives first and the claim gets reverse-engineered from it. If the claim cannot be stated before the tool is opened, the lesson is being designed by whatever the tool happens to produce first.
Candidates and choices
Ask for scenarios, and treat each as a candidate. Take one: students who walk to campus earn higher grades. Does the story have the structure the lesson needs? It should leave open several explanations, such as living closer to campus, having more flexible mornings or a different amount of time to study. A generated version may include a detail that settles the question, for example mentioning that students were assigned to walk or ride. That changes what the scenario can teach. Level matters as well. A scenario suited to a university methods course may be too dense for a first-year survey, and one that leans on a stereotype teaches something else along with the intended point.
Anything presented as a fact needs a source the instructor can open. A generated paragraph may name a study, a researcher or a figure, and a confident tone is no evidence that they exist or say what is claimed. A simple discipline is to match claims to sources: for each factual statement, write down where it comes from and where in that source it appears. Statements that cannot be matched are cut, or replaced with an original illustration that is labeled as one. This takes time, and for some materials the checking costs about as much as writing the passage directly. That cost belongs in the decision about whether to use help at all.
Likely wrong answers
One use that fits teaching well is asking for the answers students are likely to give. For the walking scenario, the likely ones include that walking improves grades, that good grades make people walk, and a vague appeal to other factors. These correspond to a causal claim, a reversed direction and an unspecific third variable, which are different things to teach. The generated list is a hypothesis about the class. The first stack of real answers tests it. Where the list was wrong, the instructor learns something about these students. Where it was right, feedback has a head start, provided each entry was checked against the concept.
Preparation can end with a short record beside the lesson: which materials were drafted with help, which sources were opened, which choices the instructor made and which were inherited from a draft. Drafting help touches wording, examples and source suggestions alike, so the record should not assume a clean split between what the tool did and what the instructor did. The note serves the next revision and anyone who takes over the course. It also keeps the preparation honest about its own method.
An original classroom example
Original, unevaluated example. The lesson claim: students can name a plausible third variable and say how they would check it. An instructor requests five scenarios, keeps the walking-to-campus case, rewrites a second so that a third variable is plausible, discards one that states assignment, and replaces a named study that cannot be found with a labeled invented illustration. Write the claim, the likely wrong answers (walking causes grades, grades cause walking, vague other factors) and a record of what was drafted with help and what was checked.
Questions for your course
- What is the one-sentence claim your lesson serves, and when was it written?
- Which details in a generated scenario could settle the question it was meant to leave open?
- How would you check a named study before using it?
- What does your record of tool use need to say for a successor to trust the lesson?