The Think-First guide
Students think. AI assists. Students own it.
A practical guide for teachers, counselors and school leaders: how to structure lessons, set AI rules and design assignments so students learn to work with AI without handing it their thinking.
1 · The problem
Why we can’t just catch it
AI detectors don’t work well enough to grade by. OpenAI shut down its own AI-text detector in 2023 because of its “low rate of accuracy.” A Stanford-led study in the journal Patterns found popular detectors wrongly flagged more than half of essays written by non-native English speakers as AI-generated.
Bans protect thinking but leave a gap. Handwritten work and in-class tests are worth keeping. But students still meet AI at home, and they’ll meet it at work. If school never teaches them how to use it well, someone else will teach them how to use it badly.
The fix isn’t catching. It’s designing. When the grade comes from a student’s questions, plans, critiques and spoken explanations, copying from AI stops being useful. You don’t need to police the product if you’re assessing the process.
2 · The method
The 5 P’s, in detail
Use this structure for any assignment where AI is allowed. It works for a 45-minute lesson or a three-week project.
Ponder
○ Screens offStudents think before anything else touches the problem. They ask questions, list ideas, sketch, estimate, recall what they know and notice what they don’t.
- Teacher moves: pose an open question; require a written first answer before discussion; use think-pair-share.
- Good prompts for students: What do I already know? What would a good answer look like? What am I unsure about? What are three possibilities?
- Evidence of thinking: handwritten notes, sketches, estimates and first answers, collected or photographed.
Plan
○ Screens offStudents turn their thinking into something precise: a brief, a hypothesis, a principle, a thesis, success criteria. This is where clear communication is learned, and it’s exactly what makes AI useful later.
- Teacher moves: give a template (see the briefs in each lesson); have partners test whether a plan is clear enough to act on.
- Why it matters: students who can say exactly what they need get far better results, from AI and from people.
Partner
● AI on · short & structuredAI is brought in for one specific job, named in advance: generate options, argue the other side, give feedback, explain a concept a different way, or produce a draft to critique.
- No student devices? Run it on the teacher’s screen. The class decides what to ask; one person types; everyone takes notes.
- Ask for options, not answers. “Give me three ways…” teaches more than “What’s the answer?”
- Keep a receipt. Students record every prompt and what came back.
Probe
○ Screens offStudents interrogate what AI produced. This is the core AI literacy skill, and one that many adults have never practiced.
- What did it get right? What is wrong, missing or invented?
- Is it generic or cliché? Did it make assumptions (about people, places, values)?
- Did it agree too easily? (Next time, ask it to argue the opposite during Partner.)
- Check one claim against a real source.
Prove
○ Screens offStudents make the final work their own and defend it. They revise by hand, then explain their choices in writing or out loud.
- Two-minute defense: “Walk me through why you made this choice.” A student who outsourced the thinking can’t do this.
- Transfer task: solve a twin problem or write a short response with no AI.
- Reflection: what I kept from AI, what I rejected, and why.
3 · Clear rules
AI use levels for every assignment
Label every assignment with a level so students never have to guess what’s allowed. Adapted from the AI Assessment Scale developed by Perkins, Furze, Roe and MacVaugh.
Show what you can do on your own. Handwritten, in class.
Example: in-class essay, unit test, reading response, the “Prove” step.
AI only after you’ve done your own thinking, to test or challenge it. Your ideas come first.
Example: argue with AI after writing your thesis; compare your solution with AI’s.
AI can give feedback or questions on your own draft. It doesn’t write for you.
Example: “What’s the weakest part of my argument?” on a draft you wrote.
AI can help create parts of the work, from your brief. You document everything and own the final choices.
Example: AI concept art from your character brief, revised by you.
The assignment is about exploring what AI can do: testing it, breaking it, comparing tools.
Example: an “AI lab” investigation of where image generators show bias.
Our levels are numbered 0–4. If your school already uses the AIAS (levels 1–5), note that our Level 4 “AI exploration” means studying AI itself, and we leave out the AIAS “Full AI” level. The AI Assessment Scale (AIAS) by Mike Perkins, Leon Furze, Jasper Roe and Jason MacVaugh is licensed under CC BY-NC-SA 4.0. Our level names, numbering and examples are adapted from it.
4 · Assessment
Grade the thinking
Six simple ways to make the process visible and the product less important.
Thinking first, on paper
Collect the Ponder page
Grade the handwritten first ideas, questions and estimates. They prove where the thinking started.
Two-minute defense
Talk it through
A short conversation (“Why this flaw? Why this source?”) reveals ownership instantly, and builds speaking skills.
AI receipts
Show your prompts
Students attach every prompt, what came back, and what they kept or rejected. Honesty becomes part of the assignment.
Critique counts
Grade the Probe
Reward students for finding AI’s mistakes, gaps and biases. It’s analytical thinking, the skill employers rank highest (WEF Future of Jobs Report 2025), applied to AI.
Versions
Draft → AI → final
Compare drafts. Growth between versions shows learning; a perfect first draft with no history raises a question.
Transfer tasks
Twin problems
End with a short, AI-free task that uses the same skill. If they can do it alone, they learned it.
5 · Templates
Classroom AI agreement & AI receipt
Copy, edit and post these. Review the agreement with students and send it home to families.
OUR CLASSROOM AI AGREEMENT 1. I think first. I write down my own ideas before I use AI. 2. I check the level. Every assignment says how much AI is allowed (Level 0–4). If I’m not sure, I ask. 3. I use AI for a job, not for my brain. I ask for options, feedback or a counterargument, not for my answer. 4. I keep receipts. I save every prompt I use and what came back. 5. I question everything. AI can be wrong, biased or made up. I check at least one claim. 6. I own my work. I can explain every choice in my own words. 7. I protect privacy. I don’t share personal information about myself or others with AI tools. 8. I follow the tool’s rules, including age requirements. Student signature: ____________________ Date: ________ Family signature: _____________________
AI RECEIPT (attach to any assignment where AI was used) Assignment: ___________________________ AI level allowed: ____ My thinking BEFORE AI (summarize your first ideas): Tool used: _____________ Date: _______ Prompt 1: What came back (summary): What I kept and why: What I rejected and why: Prompt 2: What came back (summary): What I kept and why: What I rejected and why: One claim I checked, and where: In one sentence, what is MINE in this work?
6 · FAQ
Questions teachers ask us
Short, practical answers.
Our school limits student screens. Can we still teach AI?
Yes. Every ARQIV lesson works with AI only on the teacher’s screen. Students do all their thinking, critiquing and defending on paper. Watching AI make mistakes on a projector, and catching them as a class, is one of the most powerful AI literacy lessons there is.
Should we use AI detectors?
We don’t recommend grading with them. Even OpenAI withdrew its detector for low accuracy, and research has shown detectors unfairly flag non-native English writers. A two-minute conversation about a student’s choices is far more reliable, and fairer.
What if a student uses AI at home anyway?
Design so it doesn’t help much. If the grade comes from in-class Ponder notes, a handwritten plan, a critique and a spoken defense, an AI-written final product earns very little. And students who use AI with an honest receipt are learning exactly what you want them to.
What about younger students (K–5)?
Start unplugged. Teach that AI finds patterns and makes predictions, using sorting games and “guess the next word” activities. Any AI use should be teacher-led and on the teacher’s screen. Many AI tools require users to be at least 13, so check each tool’s age rules and your district’s policy.
Doesn’t this take more class time?
A little at first. But the Ponder and Plan phases are good teaching with or without AI. You’re teaching questioning, planning and revision, skills that already sit in your standards. The AI step is usually 10–15 minutes.
Which AI tool should we use?
Use whatever your district has approved. The 5 P’s work with any chatbot or image tool, because the skill is the thinking around the tool, not the tool itself. Tools change every few months; the method doesn’t.
How does this prepare students for careers?
Employers increasingly need people who can define a problem clearly, direct AI tools, catch their mistakes and take responsibility for the result. That’s Plan, Partner, Probe and Prove. Our AI in… pages show real jobs where people work with AI every day.
Sources for this guide
- OpenAI: AI classifier discontinued for low accuracy (update, July 2023)
- Liang et al., “GPT detectors are biased against non-native English writers,” Patterns (2023)
- The AI Assessment Scale (Perkins, Furze, Roe & MacVaugh), licensed CC BY-NC-SA 4.0
- World Economic Forum, Future of Jobs Report 2025: Skills outlook
- George Pólya, How to Solve It (1945)
Put curiosity to work
Read it. Try it. Question it.
Explore new AI research, or take a paper-first investigation into your classroom.