How to Teach Problem Solving With AI
Teaching problem solving with AI works best when AI generates the specific raw material that enables each phase of problem-solving instruction: unfamiliar problems (for exploration), worked examples with explicit strategy labels (for modelling), problems designed to elicit common errors (for diagnosis), and student-accessible explanation scripts (for metacognitive discussion). AI cannot teach problem solving — but it can eliminate the most time-consuming preparation work so teachers can do more actual problem-solving instruction.
Quick Answer: Use AI to generate four types of problem-solving teaching material: open-ended entry problems for exploration (multiple solution paths exist), explicitly labelled strategy demonstrations (think-aloud scripts showing one named strategy), error-analysis problems (a worked solution with one deliberate strategic error), and reflection prompts for post-problem metacognitive discussion. These four materials map directly to the four phases of effective problem-solving instruction.
What Problem-Solving Instruction Actually Requires
There is a fundamental confusion in mathematics education between assigning word problems and teaching problem solving. Assigning word problems at the end of a chapter — after students have been taught the relevant calculation — is computation practice in narrative form. Teaching problem solving means teaching students to navigate mathematical situations where the solution path is not obvious in advance.
What Works Clearinghouse (2024) distinguishes these clearly: effective problem-solving instruction involves (1) teaching problem-solving strategies explicitly, (2) exposing students to diverse problem types that require strategy selection, (3) modelling the think-aloud process (externalising the internal reasoning), and (4) creating structured opportunities for students to discuss and evaluate their approaches after solving. This is fundamentally different from "do these ten word problems for homework."
AI supports this genuinely instructional model in ways that go beyond material generation:
- Strategy-labelled examples: AI generates worked examples with each step labelled by the strategy used, not just the calculation performed
- Open-ended entry problems: AI generates problems with multiple valid approaches, so students must choose rather than follow a recipe
- Error-designed problems: AI creates worked solutions with deliberate strategic errors (not arithmetic errors) that provoke metacognitive discussion
- Reflection prompts: AI generates structured discussion questions for after-solving conversations
Each type requires a different prompt approach.
The Four-Phase Problem-Solving Instructional Model
| Phase | Teacher Purpose | AI Material Type | Classroom Time |
|---|---|---|---|
| Exploration | Students encounter the problem without instruction | Open-ended entry problem with multiple approaches | 10-15 min independent/partner work |
| Modelling | Teacher models a named strategy explicitly | Think-aloud script with strategy labels | 10-15 min whole-class demonstration |
| Strategy Practice | Students apply named strategies to fresh problems | Strategy-targeted problem set | 20-30 min guided or independent practice |
| Reflection | Students evaluate their approach and compare alternatives | Post-solving reflection prompts; error-analysis discussions | 10-15 min whole-class debrief |
AI generates the material for all four phases. The teacher provides the instruction — the facilitation, the think-aloud, the questioning, the discussion leadership. AI can save preparation time; the teacher provides the pedagogical expertise.
Phase 1: Generating Open-Ended Entry Problems
An open-ended entry problem is one where more than one solution path leads to a correct answer. Students who have been taught only algorithmic mathematics may struggle initially with the ambiguity — which is precisely the point.
"Write 3 open-ended problem-solving tasks for Grade 6 students on the topic of multiplication and division. Each task should have at least two valid solution approaches (e.g., one using ratio reasoning and one using multiplication). Problems should not have a single obvious method — the challenge should be in deciding how to approach the problem, not in executing the calculation. Use contexts: planning a school event (budgeting materials), organising a school sports day, comparing travel options. Provide teacher notes listing two or three different valid solution approaches for each problem."
Why teacher notes are essential for entry problems: A teacher who has not worked through the multiple approaches cannot facilitate the exploration discussion productively. Teacher notes that describe two or three valid approaches give the teacher a mental map of what students might do — without scripting the lesson.
Sample output: "The school canteen plans to buy 120 sandwiches. A box of 8 costs $4.80. A box of 12 costs $6.60. Which is better value? Show your working." Valid approach 1: unit price comparison (each box per sandwich). Valid approach 2: total cost for 120 using each option. Valid approach 3: ratio — compare the price-per-quantity ratios.
Phase 2: Generating Think-Aloud Strategy Scripts
A think-aloud script externalises the internal reasoning process. It is not a solution — it is a narration of how a mathematical thinker approaches a problem, including the moments of uncertainty, the strategy selection decisions, and the verification steps.
"Write a think-aloud script for a Grade 5 teacher demonstrating the 'draw a model' strategy for this problem: 'A school garden is 24 metres long and its perimeter is 72 metres. What is the width?' The script should: (1) read the problem aloud; (2) identify what is known and what is unknown; (3) explain the decision to use a drawn model; (4) draw and label the model (describing what to draw); (5) write the equation from the model; (6) solve and verify. Write it in first-person teacher voice, including moments where the teacher says 'I'm not sure yet' or 'let me check' — model genuine mathematical thinking, not perfect performance."
The "imperfect performance" requirement: A think-aloud that shows a teacher confidently executing every step without uncertainty does not model problem-solving — it models performance. Students watching a perfectly executed solution do not see what to do when they are uncertain. Requesting moments of explicit uncertainty makes the think-aloud authentically instructional.
"Write a think-aloud script for a Grade 7 teacher demonstrating the 'work backwards' strategy for this problem: 'After buying 3 books and paying $5 postage, Maya had $12 left from her $32 gift card. How much did each book cost?' Show how a teacher would identify that working backwards is the right strategy, start from the end state, and reverse each operation to find the original cost."
Phase 3: Generating Strategy-Targeted Problem Sets
Once a specific strategy has been modelled, students need practice problems that are well-suited to that strategy. A problem set for the "draw a model" strategy should contain problems where drawing a model is the most efficient approach; it should not contain problems where direct calculation is faster and the model is unnecessary.
"Write 8 problems suitable for the 'draw a diagram' strategy for Grade 5 students. Problems should involve: area and perimeter where dimensions are given in parts (not direct formulas); comparison of two quantities where a bar model clarifies the relationship; multi-step problems involving rate or ratio. Problems should not be solvable by direct formula recall — students should need to represent the structure visually before calculating. Provide the model drawing description in the answer key."
"Write 6 problems suitable for the 'find a pattern' strategy for Grade 6-7 students. Each problem presents a sequence or relationship and asks students to identify the pattern, extend it, and use it to find a far term or total. Include at least two problems where the pattern is not immediately obvious (requires looking at second differences, not first differences). Provide the pattern rule and the answer in the teacher notes."
Phase 4: Reflection Prompts and Error Analysis
The most underused phase of problem-solving instruction is the structured reflection after solving. Students who solve a problem and move on do not build strategic awareness; students who compare their approach with an alternative and evaluate both develop metacognitive problem-solving capacity.
"Write 6 post-problem reflection prompts for Grade 7 students to use after solving a multi-step mathematics problem. Prompts should encourage: (a) naming the strategy used; (b) identifying the point where the student felt most uncertain; (c) considering whether a different approach would have been more efficient; (d) checking whether the answer makes sense in context; (e) connecting this problem type to a previous one; (f) identifying what they would do differently next time. Write prompts as question stems that work for any problem."
Error-analysis problems for reflection:
"Write 5 error-analysis problems for Grade 6 students. Each shows a multi-step word problem and a student's attempted solution. The error in each solution should be a strategic error (chose the wrong strategy or applied the right strategy incorrectly to this context) — not a calculation error. Examples: used the total rather than the unit amount; found the area when perimeter was needed; subtracted when addition was required by the comparison context. Label the error type in the teacher answer key but not in the student version. Ask students to identify the error and explain why it is an error."
A Classroom Scenario: A Three-Week Grade 5 Problem-Solving Unit
Say you teach Grade 5 mathematics and you're beginning a three-week problem-solving unit specifically designed to develop your students' strategic flexibility — their ability to choose between strategies rather than applying a single memorised procedure.
Imagine your students are strong at algorithmic calculation but freeze when problems don't match a familiar format. Your goal is to build comfort with strategic uncertainty.
A possible AI-generated four-phase sequence:
Week 1 — "Draw a Model" strategy:
- Monday: Open-ended entry problem (school garden problem with no strategy hint)
- Tuesday: Think-aloud script — teacher models "draw a model" using a perimeter and area problem
- Wednesday-Thursday: 8-problem "draw a model" practice set
- Friday: Reflection discussion + 2 error-analysis problems where a student used calculation instead of a model and got confused
Week 2 — "Work Backwards" strategy: Same four-phase structure, different strategy.
Week 3 — Mixed strategy selection:
- Problems that require students to identify which strategy is appropriate before applying it
- Post-solving comparison discussions: "Which strategy did you use? Did anyone use a different approach? Which was more efficient?"
Generating the entry problem, think-aloud script, practice set, and reflection materials for a week with AI might take roughly 45 minutes. Preparing the same four phases by hand — writing each problem, the think-aloud, the practice set, and the reflection prompts from scratch — could otherwise take several hours a week.
The intended shift: by Week 3, you want students regularly asking "what strategy should I use?" before starting, rather than immediately writing calculations. That change in behaviour is the goal of problem-solving instruction.
RAND Corporation (2024) found that explicit strategy instruction — naming strategies, modelling their application with think-alouds, and providing structured practice — produces significantly stronger problem-solving performance at Grades 4-7 than problem practice without explicit strategy labelling. The AI-generated strategy-labelled materials make this evidence-based approach practical at a single-teacher scale.
Pro Tips for AI Problem-Solving Materials
- Always request "at least two valid solution approaches" in teacher notes for entry problems. Without this, the teacher risks not knowing how students might approach a problem differently — which undermines the exploration discussion.
- Name the strategy in every think-aloud script prompt. "Think-aloud for the 'work backwards' strategy" is far more useful than "think-aloud for a multi-step problem." The named strategy becomes the vocabulary students use when discussing their approach.
- Generate error-analysis problems with strategic errors, not arithmetic errors. Arithmetic errors are caught by checking; strategic errors require understanding why a different approach was needed. These are the most instructionally rich errors for problem-solving discussion.
- Use EduGenius for generating structured problem-solving worksheets. When you need formatted worksheets with problem statement, workspace, reflection questions, and answer key on a separate page, EduGenius produces structured output that can save formatting time. The Bloom's Taxonomy alignment also ensures that practice problems progress from recall through application through analysis — natural for a problem-solving unit.
- Generate problems in your students' language and cultural context. "School canteen" may be unfamiliar to students who call it a "cafeteria"; "sports day" is British English; local market contexts are more motivating than US retail examples. Add your context preference to every problem-generation prompt.
What to Avoid
Avoid Calling Word Problems at Chapter Ends "Problem Solving Instruction"
Post-chapter word problems test whether students can apply a calculation they just learned. Genuine problem-solving instruction requires problems where the solution path is not obvious. If every student in the class immediately knows which operation to apply, the problem is not teaching problem-solving — it is testing procedural recall in a context.
Avoid Think-Aloud Scripts That Show Perfect Performance
A teacher think-aloud that moves confidently from problem to solution without visible uncertainty does not model how mathematical thinkers actually work. Include moments of "I'm not sure which strategy to try first," "let me check if this makes sense," and "this approach didn't simplify things — let me try a different direction." These moments are the most instructionally valuable.
Avoid Skipping the Reflection Phase
The exploration and practice phases generate mathematical experience. The reflection phase converts that experience into transferable insight. Students who never reflect on their approach after solving do not generalise their strategies to new contexts. Even a five-minute structured discussion using AI-generated reflection prompts is more valuable than five additional practice problems.
Avoid Generating All Problems With a Single Context
A problem-solving unit that always uses shopping contexts teaches students that "problem solving" means "shopping math" — not an unreasonable inference for a 10-year-old. Vary contexts deliberately: scientific, sporting, geographic, social. Problem-solving strategies transfer across contexts; using varied contexts teaches and tests that transfer.
Key Takeaways
- Teaching problem solving with AI requires four material types: open-ended entry problems, strategy think-aloud scripts, strategy-targeted practice problems, and post-solving reflection prompts/error-analysis tasks.
- AI cannot teach problem solving — it generates the materials that enable teachers to teach it. The think-aloud, the facilitation, and the discussion remain the teacher's work.
- Think-aloud scripts should include moments of strategic uncertainty — a perfectly confident think-aloud does not model genuine mathematical problem-solving.
- Error-analysis problems with strategic errors (wrong strategy chosen, not arithmetic mistakes) produce the richest problem-solving discussion.
- Explicit strategy naming — "draw a model," "work backwards," "find a pattern" — gives students vocabulary to reflect on their own approach after solving.
- The four-phase instructional cycle (explore → model → practice → reflect) maps directly to the four AI material types; each phase has a different AI prompt approach.
FAQ
What is the difference between problem solving and word problems?
Problem solving is instruction where the solution path is not immediately known and students must select and apply strategies. Word problems — especially post-chapter exercises — are typically computation practice where the operation is implied by the chapter topic. Genuine problem-solving instruction uses unfamiliar problem contexts where strategy selection is part of the challenge, not something students can infer from context clues.
Which problem-solving strategies should I teach at Grades 5-7?
Core strategies for Grades 5-7: draw a model/diagram, work backwards, find a pattern, make an organised list, guess and check with systematic adjustment, simplify the problem, and use logical reasoning (eliminate and narrow). NCTM (2025) recommends introducing 2-3 strategies per year with explicit modelling rather than introducing many strategies superficially. For percentage-specific problem-solving strategies, see Best AI for Percentages in 2026-2027.
How do I evaluate problem-solving strategies on an assessment?
Evaluate on three dimensions, not just the answer: (1) Is the strategy appropriate for this problem type? (2) Is the strategy applied correctly? (3) Does the student verify the answer and check it makes sense in context? A student who chooses an appropriate strategy, applies it with one small error, and checks their answer has demonstrated stronger problem-solving competence than a student who gets the right answer by guessing. Design marking criteria before the assessment and share them with students in advance. For multi-step word problem AI resources, see How AI Helps Students Master Multi-Step Word Problems.
Can students use AI directly for problem solving practice?
This is a nuanced question. Students who use AI to check their own approach after solving — "I got this answer; can you show me another way to solve this problem?" — develop strategic awareness. Students who use AI to find the answer before attempting the problem bypass the problem-solving process entirely. The distinction is whether AI is a checking tool or a replacement tool. At Grades 5-9, teacher guidance on appropriate AI use in mathematics is part of responsible AI literacy education. For Grade 2 money problems as problem-solving entry points, see AI Word Problems for Money Math in Grade 2.
For the complete overview of AI in mathematics teaching, see the AI for Math Education: The Complete 2026 Guide. For place value tool comparisons, see Best AI for Place Value in 2026-2027. For percentage problem solving specifically, see Best AI for Percentages in 2026-2027. For cross-subject study guide generation, see Best AI Study Guide Generators in 2026.