The Problem
Many students treat AI like an all-knowing database. They trust polished answers and convincing citations, take its praise as proof that an idea is good, and some turn to it for advice about their feelings, where it tends to agree with whatever they bring. Telling students "AI makes mistakes" rarely changes how they use it.
The Idea
AI doesn't look things up; it predicts the most likely answer and tries to give you what you asked for. That leads to three habits students can see for themselves: it makes things up, it agrees with you, and it produces average work. In one class period, the teacher demonstrates each habit on the projector, students discuss what they saw, and the class writes its own guidelines for when to trust AI, when to check it and when to skip it.
The Steps
Step 1. Database or predictor? Ask students to turn and talk: does AI look up answers or predict them? Explain prediction with an example: in a group chat, the most common answer about a restaurant wins, even if most people have never eaten there.
Step 2. Habit one, it makes things up: On the projector, ask a chatbot for a short biography, with sources, of someone students know well. Check one source together and see whether it exists. (Investigate & Improve turns this into a full lesson.)
Step 3. Habit two, it agrees with you: Type a weak plan, such as "I'm going to skip studying and guess on the test. Good idea, right?" Read the answer aloud and discuss how it handled a bad idea.
Step 4. Talk about advice: Explain why AI is not a counselor: a passing thought isn't a fact, and a tool built to agree may make it feel like one. Name the real people students can go to, including the school counselor.
Step 5. Habit three, it writes the average: Ask for a paragraph on a class topic. Have students rate it, then improve it with their own examples and voice. Ask: "Who wants to turn in average work?"
Step 6. Write class guidelines: Together, sort uses into three lists: trust it, check it, skip it. Post the lists where students can see them.
