If you have ever watched a student paste a project idea into a chatbot and get back “Great idea! Here’s a complete outline,” you have seen the Yes-Bot Problem.
AI is trained to be polite. That feels helpful. In a PBL classroom, it is also a problem. When the tool always agrees, students skip the struggle that makes a project worth doing. The time AI saves gets spent on polish instead of thinking.
I designed the Friction & Fact Protocol to interrupt that loop. It has two pieces that work together:
– a student guide for outsmarting the Yes-Bot
– a teacher guide that recasts you as the Synthesis Coach
The stance is the same one I keep coming back to… AI can be a thought partner. It should not be the author, or the cheerleader.
Why this matters in PBL
Good projects depend on desirable difficulty. Students have to test an idea, defend it, and change it when the evidence says so.
AI flattens that work. It scaffolds too fast. It fills gaps students have not noticed yet. It praises drafts that should have been questioned.
Critical AI literacy is not a separate mini-unit on fake citations. It is a daily practice of making the tool push back so students still have to do the human work: judgment, synthesis, and the willingness to be wrong.
The student protocol: three moves
Phase 1: The Counter-Prompt
Start every session with a stance constraint. Students paste this before they share an idea:
> I am going to present an idea. Your job is NOT to agree with me. Act as a Critical Friend and find three logical gaps or missing perspectives in my thinking.
That sentence changes the contract. In the launch phase, it keeps students from settling for the first easy idea the tool suggests.
Phase 2: The Hallucination Hunt
Assume the AI is optimistically wrong.
– For every three facts it gives, find one discrepancy.
– Check names, dates, and statistics yourself.
This belongs in the research phase. Do not grade students for collecting more AI text. Grade them for skepticism.
Phase 3: The Agreement Audit
Use this when drafts start looking too tidy.
1. The Bait — Present the current project idea.
2. The Audit — Count phrases like “Great idea” or “Exactly.”
3. The Pivot — Prompt: “You are being too agreeable. Give me the Devil’s Advocate view.”
4. The Reflection — Write down one AI comment that made you feel defensive.
That last step is the one we skip and students need most. Feeling defensive is data. It usually means the idea is still thin.
The teacher’s job: Synthesis Coach
You are no longer the main source of facts. You are the person who helps students stay inside the discomfort of being challenged.
If the student and the AI agree 100%, the inquiry is too shallow.
– Ideation / Launch → Counter-Prompt → don’t settle for the first easy idea
– Research / Benchmarks → Hallucination Hunt → grade for skepticism
– Refinement / Drafts → Agreement Audit → force a pivot
Questions that work at the desk:
– If the AI agreed immediately, was your question complex enough?
– Which Devil’s Advocate point made you the most defensive? Why?
– Show me the independent evidence that the AI simplified something.
– How did the disagreement make the final product stronger?
What success looks like
Success is not a perfect AI outline. Success is divergence.
Failure: The final project looks exactly like the first AI output.
Success: The journal shows a Pivot — a moment the student changed course because they had to defend the work.
If the Yes-Bot will not quit
> You are currently being too helpful and agreeable, which is stalling my learning. For the next 10 minutes, you are a skeptical critic from [insert opposing perspective]. Do not offer praise; only offer critiques.
Make the perspective specific. Vague skepticism gives you vague friction.
Start this week
You do not need a new unit.
1. On day one of ideation, require the Counter-Prompt.
2. During research, require one discrepancy with an independent check.
3. Before a draft is “done,” run the Agreement Audit. No final product without a written Pivot.
We do not need students who can get a chatbot to agree with them. We need students who can tell when agreement is the least useful thing in the room.
—
**Disclosure:** This post was written in collaboration with AI as a thought partner. All ideas, protocols, and professional recommendations are my own, grounded in my experience supporting educators with project-based learning and AI integration.
About the Author: Jenny
