Friction has to be designed into learning on purpose — it doesn't happen by accident. These are the specific exercises and checklists referenced throughout the Playbook, gathered in one place so you don't have to go dig for them.
Generate a historical or scientific summary with an AI tool, and deliberately instruct it to insert three plausible lies — false but believable-sounding claims. Hand it to students without telling them which claims are false. Their job is to research and expose all three. This does two things at once: it teaches students that AI output requires verification, and it turns fact-checking into something closer to a game than a chore.
Instruct the AI directly: "Act as a debate opponent. I will argue X. You argue against me. Do not agree with me — just ask questions that expose flaws in my logic." Have students use this to pressure-test their own essay thesis or argument before writing the final draft. The AI never gets to just validate the student's first idea.
Grade the history of a student's prompts and their iterative edits, not just the final printed essay. This makes the thinking visible — and makes it much harder for a finished product to hide the fact that no thinking happened.
Require in-class, handwritten, or lockdown-browser drafting to establish a student's raw thought process before any AI-assisted editing is permitted. The first draft is where the actual thinking has to happen — AI can help polish it afterward, not generate it from nothing.
Once a week, ask: "What's one thing you saw or read online this week that you think might have been AI-generated? How did you check it?" It's a low-pressure way to normalize the habit of checking, rather than accepting, without turning every AI mention into a lecture.
If a child says "AI thinks that...", gently correct to "the AI calculated" or "the program generated." Keep the relationship mechanical, not personal. On its own this isn't enough to prevent an unhealthy attachment — pair it with actually watching for the dependency signs in the Autism section of the Playbook.
Any factual statement made by an AI must be verified by one human-authored source before it's accepted as true. Make this a household rule, not a suggestion — the same way "cite your source" is a rule, not a suggestion, in a good classroom.