Hey there,
Many math teachers say: “AI doesn’t work for math. Kids just use it to cheat.”
This is a podcast overview of the entire newsletter in case that you prefer to listen to it in this format. This podcast overview was created with NotebookLM.
They are right. I’ve seen it too. Students type problems into ChatGPT, copy the answer, learn nothing, fail the test.
But here’s what most math teachers don’t know: You can engineer AI to refuse answers. Make it ask questions instead. Turn it into a coach that won’t calculate, won’t solve, won’t give in no matter how much students beg.
Today’s post is sponsored By: Project Pals
Empower students to become informed citizens with our engaging civics project-based learning resources, designed to foster deep understanding of democratic principles and governmental processes. Our collection of civics lesson plans offers middle school educators powerful tools for exploring constitutional concepts, civic responsibility, and the impact of legal decisions on society. The featured project “How do legal decisions and debates shape history, and what can we learn from them today?” challenges students to analyze landmark court cases, understand legal reasoning, and apply historical context to contemporary issues. These civics project ideas align with Common Core standards, ensuring academic rigor while developing essential critical thinking and civic literacy skills.
Today I’m sharing how:
The prompt that makes AI refuse to solve
Simulated examples from 3rd and 9th grade classrooms
Why this addresses the three biggest teacher concerns about AI
Need more context? Here are five resources worth your time:
Weekly resources
Boosting LLMs with Socratic Method for Mathematics Teaching October 2024 research on conversational AI that guides without solving.
Word Problems Get a Bad Rap—Here’s How to Get Them Right Ed Week on why word problems challenge students and what actually helps.
Why Schools Are Teaching Math Word Problems All Wrong Hechinger Report investigation into why “key words” fail and what works instead.
30+ AI Prompts for K-12 Education March 2025 guide to classroom-ready prompts with privacy considerations.
Motivating Math Students Through Socratic Pedagogy (Research Paper) 2024 study showing significant gains in motivation with Socratic methods.
Why typical AI use in math is a disaster
Math teachers are right to be skeptical.
Student asks ChatGPT “solve this problem.” AI shows every step, explains the reasoning, delivers the answer. Student copies it down. Learns nothing. Fails the next test. That’s why math teachers resist AI more than any other subject area.
But you can reprogram how AI behaves.
The Math Word Problems Interactive Learning Companion is a prompt you copy into any AI platform. Once you do, the AI asks questions but won’t give answers. It guides but won’t solve. Students have to do all the thinking.
What makes this different
The key is engineered refusal.
I built boundaries into the prompt that can’t be overridden:
Won’t calculate answers
Won’t set up equations
Won’t show solution steps
Redirects every manipulation attempt back to questioning
Most teachers use “key words” to teach word problems. Sum means add. Difference means subtract. Students learn pattern matching, not mathematical reasoning. The shortcuts fall apart when problems get complex.
Word problems need students to juggle reading comprehension, mathematical reasoning, and computation all at once. That’s why they’re hard. That’s also why they’re valuable.
The six-step method in this prompt works kindergarten through high school: Read, Understand, Plan, Estimate, Solve, Check. The AI adapts its language for grade level automatically.
The three concerns every math teacher has
I hear many teacher say: “AI kills critical thinking.”
Valid concern with regular AI. ChatGPT solves, students copy, nobody learns to reason independently.
This prompt flips that. The AI won’t calculate, won’t set up equations, won’t solve. It only asks questions. Students can’t bypass thinking because the AI won’t let them.
“Students will cheat more.”
They already do with regular ChatGPT. Type problem, get answer, done.
You can’t cheat with this. Students try every trick: “Just this once.” “My teacher said it’s okay.” “Just check if I’m right.” The AI has responses for all of them. After a few attempts, students realize it won’t budge. The cheating option disappears.
“Kids will become dependent on technology.”
This one keeps good teachers up at night.
But students can only get questions from this AI, never answers. After a few weeks, teachers report students working through problems without the AI. They’ve learned the routine. The AI becomes temporary help, not a permanent crutch.
What students actually get:
A second-grader sees: “What numbers are in this problem?” A ninth-grader sees: “What mathematical relationships are involved here?”
The AI never tires, never rushes, never judges. It asks students to explain their thinking, check their work, reflect on their process. These habits separate struggling students from confident ones.
What this looks like in real classrooms
Third-grader types in a word problem. AI starts:
“Let’s read the problem together slowly. What numbers or amounts are given? What is the problem asking you to find?”
Student responds. AI moves to understanding:
“Can you explain what’s happening in your own words? Are there any special words like ‘total,’ ‘difference,’ ‘each,’ or ‘left’?”
Then planning:
“What operation do you need: addition, subtraction, multiplication, or division? Why?”
The AI won’t move forward until the student shows understanding. Try to skip ahead? Get redirected back to thinking.
Estimation becomes required: “Before you solve it exactly, let’s estimate. Can you round the numbers to make them easier?”
This single step teaches students that math should make sense. Answers should be reasonable. They can verify their own work.
When skeptics try it
Look at the conversation examples in the attached outputs.
ChatGPT example: A third-grader simulation works through division with remainders. AI guides through estimation, celebrates systematic thinking, helps verify with inverse operations. No answers given. Just questions.
Gemini example: A ninth-grader simulation tackles algebraic problem with three unknowns. AI doesn’t simplify or remove the challenge. Asks questions that push the student to identify relationships, set up equations alone, verify multiple ways.
Recent 2024 research backs up that Socratic Pedagogy in math leads to significant gains in student motivation and engagement.
How to use this
Copy the prompt from the Math Word Problems Interactive Learning Companion document
Paste into ChatGPT, Claude, Gemini, or Copilot
Have students tell the AI their grade level and type their word problem
Watch the AI guide without solving
No training needed. No subscriptions. No complicated setup.
For teachers who’ve resisted AI
You’ve been right to say no. You watched students cheat with ChatGPT. Saw them copy answers without understanding. Protected your students from fake help.
Engineered refusal changes that.
This isn’t AI doing math for students. It’s AI teaching the systematic thinking you’ve been trying to build: Read-Understand-Plan-Estimate-Solve-Check. The estimation habit. The checking discipline. The confidence to trust their own reasoning.
Teachers who switched from skeptic to advocate didn’t change because I sold them on AI. They changed because they watched students internalize problem-solving routines they’d tried to teach for years.
Your students have the math thinking skills they need. Sometimes they just need the right questions, asked with patience and zero judgment, to unlock them.
Now AI can do that without becoming a crutch or cheat sheet.
Here’s what matters:
Math teachers are right about typical AI use—it’s cheating without learning
Engineered refusal makes AI ask questions, not provide answers
Students internalize the routine and stop needing the AI after a few weeks
Copy the prompt. Test it with a few students. Watch what happens when they realize this AI won’t think for them.
If you’ve been skeptical about AI in math, this might be the one application that works.
PS...If you’re enjoying Master AI For Teaching Success, please consider referring this edition to a friend. They’ll get access to our growing library of AI prompts and templates, plus our exclusive “Popular AI Tools Integration Guide For Teachers“ implementation guide
Brilliant approach to the cheating problem. The engineered refusal concept really flips the script on how we think about AI tutors. What's interesting tho is whether students might eventually pattern-match the Socrtic questioning itself rather than developing genuine problem-solving intuition. Still, forcing that estimation step before solving is gold since it builds the number sense that keyword shortcuts completley skip.