Best AI for Teaching Mathematics: Research, Practice, and Tools for 2026
Quick Answer: AI tools support mathematics teaching by generating problem-based lesson sequences, developing high-cognitive-demand tasks, creating multiple representations of mathematical concepts, building mathematical discourse protocols, generating differentiated practice with worked examples, and designing formative assessment sequences. Platforms like EduGenius help teachers at Grades KG-9 create mathematics lessons that develop both procedural fluency and conceptual understanding—the combination that NCTM research identifies as the foundation of genuine mathematical proficiency.
Mathematics education has been shaped by more explicit controversy than any other K-12 subject—not by lack of research, but by the difficulty of translating research into classroom practice in the face of persistent disagreements about what math education is for.
The "math wars" that peaked in the United States in the 1990s pitted traditional procedural instruction advocates against reform mathematics advocates. That debate produced enormous heat and somewhat less light, obscuring a research consensus that has actually existed for decades: procedural fluency and conceptual understanding are both essential, and neither alone is sufficient.
The research is consistent on why both matter:
- Procedural fluency without understanding limits students to the exact procedures they've been taught—they cannot adapt to novel situations or diagnose their own errors
- Conceptual understanding without fluency leaves students unable to execute reliably or efficiently
- Genuine proficiency requires both, developed together, using procedures as a vehicle for understanding rather than treating them as alternatives
AI tools support this integration by generating problem tasks that develop conceptual understanding through productive struggle, practice sequences that build procedural fluency efficiently, and mathematical discourse activities that develop the reasoning and communication skills that deepen both.
The Research Foundations of Mathematics Education
Skemp: Relational vs. Instrumental Understanding
Richard Skemp's 1976 paper "Relational Understanding and Instrumental Understanding" in Mathematics Teaching introduced a distinction that has influenced mathematics education research for nearly five decades.
Instrumental understanding: Knowing rules and procedures without understanding why they work—"rules without reasons." Students with only instrumental understanding can execute algorithms they've been shown but cannot adapt to novel situations or diagnose their own errors.
Relational understanding: Understanding what to do, why it works, and how procedures connect to underlying mathematical relationships. Students with relational understanding can construct procedures from understanding, adapt to novel problems, and recognize errors by checking against their conceptual model.
Skemp noted that instrumental understanding is faster to teach and easier to assess, which creates systemic pressure toward instrumental instruction even when teachers know that relational understanding is the more valuable outcome. This pressure is still present in contemporary mathematics education.
Hiebert and Lefevre: Procedural and Conceptual Knowledge
James Hiebert and Patricia Lefevre's Conceptual and Procedural Knowledge: The Case of Mathematics (1986) provided the most systematic theoretical treatment of the distinction that Skemp had identified practically. They defined:
Conceptual knowledge: Knowledge rich in relationships—a web of connections between mathematical facts, ideas, and procedures. Cannot be isolated in memory as a single piece; it is inherently connected.
Procedural knowledge: Knowledge of formal language, symbols, and algorithms—sequences of actions for solving mathematical problems. Can be learned as isolated procedures without connection to conceptual knowledge.
Hiebert and Lefevre's key contribution was showing that conceptual and procedural knowledge can develop independently—students can learn procedures without developing the conceptual connections that give those procedures meaning—but that this independence is educationally undesirable. The research task became identifying how to develop both simultaneously.
NCTM's Principles to Actions
The National Council of Teachers of Mathematics' Principles to Actions: Ensuring Mathematical Success for All (2014) is the most influential recent statement of research-based practice for K-12 mathematics education. The document identifies eight Mathematics Teaching Practices supported by research:
- Establish mathematics goals to focus learning: Clear articulation of what mathematical understanding students are developing
- Implement tasks that promote reasoning and problem solving: High-cognitive-demand tasks that require students to think, not just execute
- Use and connect mathematical representations: Multiple representations (concrete, visual, symbolic) connected to develop conceptual understanding
- Facilitate meaningful mathematical discourse: Purposeful classroom talk that develops mathematical reasoning
- Pose purposeful questions: Questions that reveal students' thinking and advance mathematical reasoning
- Build procedural fluency from conceptual understanding: Procedural fluency developed through understanding, not before it
- Support productive struggle in learning mathematics: Creating conditions for students to grapple productively with challenging mathematics
- Elicit and use evidence of student thinking: Formative assessment used to inform instruction
The NCTM Practices provide the most practically actionable synthesis of mathematics education research for classroom use, and they have become the standard reference for mathematics professional development and curriculum design.
Smith and Stein: Mathematical Task Framework
Margaret Smith and Mary Kay Stein's Mathematical Task Framework (introduced in Stein, Grover, and Henningsen 1996; fully developed in Smith and Stein 5 Practices for Orchestrating Productive Mathematics Discussions, 2011) provides a tool for analyzing the cognitive demand of mathematical tasks.
Tasks are classified into four levels:
- Memorization tasks: Reproduce a fact, definition, or rule—no mathematical reasoning required. ("What is the formula for the area of a rectangle?")
- Procedures without connections: Apply a familiar algorithm to a routine problem—no connection to underlying concepts. ("Calculate 3/4 × 8.")
- Procedures with connections: Apply procedures in ways that connect to underlying concepts and relationships. ("Show three different ways to find 3/4 × 8 and explain which method is most efficient and why.")
- Doing mathematics: Genuine mathematical thinking—complex, non-routine, requiring reasoning and sense-making. ("Investigate: For which values of n does n/4 × 8 produce a whole number? Is there a pattern?")
Smith and Stein's research found that the cognitive demand of tasks as implemented in classrooms frequently declined from the cognitive demand of tasks as designed. Teachers reduced task challenge by breaking down the problem into smaller steps, providing too much scaffolding, or accepting low-level responses.
The five practices they identified for orchestrating productive mathematical discussions are designed to prevent this "funneling" that reduces mathematical thinking.
Ball: Mathematical Knowledge for Teaching
Deborah Ball and colleagues' research on Mathematical Knowledge for Teaching (MKT) (Ball, Thames, and Phelps 2008, Journal of Mathematics Teacher Education) identified a specialized form of mathematical knowledge that is distinct from both general mathematical knowledge and general pedagogical knowledge.
MKT includes:
- Common content knowledge: Mathematical knowledge and skill used by teachers in common with others (knowing mathematics itself)
- Specialized content knowledge: Mathematical knowledge used specifically in teaching—like the ability to evaluate non-standard student solutions, identify mathematical errors and their sources, and explain why procedures work
- Knowledge of content and students: Knowing how students typically think and misunderstand specific mathematical topics
- Knowledge of content and teaching: Knowing how to sequence mathematical ideas, which examples are most pedagogically powerful, and how to structure mathematical tasks
Ball's research demonstrated that teachers with stronger MKT produce significantly better student learning outcomes—and that MKT is distinct from mathematical content knowledge per se. A mathematician who knows calculus may not know how students typically develop misconceptions about functions, and a skilled calculator cannot explain why long division works.
Teaching mathematics effectively requires both mathematical content knowledge and the specialized pedagogical content knowledge that supports learning.
Schoenfeld: Mathematical Problem Solving
Alan Schoenfeld's Mathematical Problem Solving (1985) and subsequent research program investigated what it means to think mathematically and what distinguishes expert mathematical problem solvers from novices.
Schoenfeld identified four categories of knowledge and behavior that distinguish expert mathematical problem solvers:
- Resources: Mathematical facts and procedures available to the problem solver
- Heuristics: Problem-solving strategies (draw a diagram, look for a pattern, work backwards, consider a simpler problem)
- Control: Metacognitive monitoring of one's own problem-solving process—knowing when to abandon an unproductive approach and try a different one
- Beliefs: Conceptions about mathematics and problem solving that affect how solvers engage with problems (e.g., belief that all problems should be solvable in under five minutes, or belief that mathematics involves memorizing formulas rather than reasoning)
Schoenfeld's research showed that novices (students) and experts (mathematicians) have comparable resources for simple problems but differ dramatically in heuristics, control, and beliefs for complex problems. Implications for teaching: problem-solving instruction should explicitly address heuristics, metacognitive control, and mathematical beliefs—not just provide practice on problem types.
Boaler: Mathematical Mindsets and Equity
Jo Boaler's research program (Experiencing School Mathematics, 1997; Mathematical Mindsets, 2016; Limitless Mind, 2019) addressed the relationship between students' mathematical beliefs, classroom practices, and mathematical achievement—with particular attention to equity.
Boaler's key findings:
- Fixed mindset about mathematical ability ("some people are just math people") is strongly correlated with lower achievement and is significantly influenced by classroom practices
- Open-ended, investigative mathematics (project-based, discussion-rich, multiple-approach) produces better equity outcomes than traditional procedural instruction—reducing achievement gaps between demographic groups more effectively
- Timed testing in elementary mathematics produces mathematical anxiety that disproportionately affects girls and students from underrepresented groups, without producing fluency benefits
- Collaborative, heterogeneous group work in mathematics, structured to value diverse approaches rather than speed to correct answers, produces strong achievement outcomes across demographic groups
Boaler's work sits at the intersection of mathematics cognition and social justice in education, arguing that equitable mathematics instruction requires redesigning not just what we teach but how students are positioned in relation to mathematical knowledge.
Stigler and Hiebert: The Teaching Gap
James Stigler and James Hiebert's The Teaching Gap (1999) analyzed videotapes of mathematics lessons in Japan, Germany, and the United States from the 1995 TIMSS (Third International Mathematics and Science Study) video study. Their findings documented stark differences in the structure and cognitive demands of mathematics lessons across countries:
- Japanese lessons centered on a single challenging problem, allowed extended struggle, and focused on comparing and discussing multiple student solution methods
- U.S. lessons covered more topics in each lesson, broke problems into smaller steps, and focused on demonstrating procedures that students then practiced
- German lessons were procedurally focused but with more time on any single topic than U.S. lessons
The Japanese "structured problem solving" approach—which Stigler and Hiebert observed producing both high achievement and genuine mathematical thinking—has influenced U.S. mathematics professional development, particularly around the "launch-explore-discuss" lesson structure.
AI Applications in Mathematics Teaching
High-Cognitive-Demand Task Generation
A prompt for a Grade 5 "doing mathematics" task:
"Generate a 'doing mathematics' level task for Grade 5 students on multiplication and area. The task should: (1) be non-routine—not solvable by applying a procedure students already know; (2) have multiple approaches (visual, algebraic, arithmetic); (3) require students to explain their reasoning, not just calculate; (4) produce a surprising or elegant result when students arrive at it; and (5) connect multiplication to the area model. Include anticipated student approaches and common misconceptions."
A prompt for a Grade 8 investigation sequence on slope:
"Create a mathematical problem sequence for Grade 8 that develops the concept of slope through investigation rather than definition. The sequence should: begin with a real-world context (ramp steepness, rate of change in a data situation); have students develop their own measure of steepness before the formal definition is introduced; connect the informal measure to the slope formula; and end with a challenge problem where students apply slope reasoning to a novel situation. Include teacher facilitation notes for each phase."
Multiple Representations
A prompt for building three connected representations of a single concept:
"Generate three representations of the concept of equivalent fractions for Grade 4 students:
- A visual/area model with folded paper or drawn shapes
- A number line representation
- A ratio table showing multiplicative relationships
For each representation: show how to introduce it, provide 3 practice examples at increasing complexity, and describe how to connect this representation to the others. The goal is for students to see fractions as the same underlying concept, not three different topics."
Mathematical Discourse Facilitation
A prompt for a 5 Practices math talk lesson:
"Design a 'math talk' lesson on Grade 6 ratios using the 5 Practices framework (Anticipate, Monitor, Select, Sequence, Connect). Include: (1) a launch problem worth discussing; (2) anticipated student solution approaches (at least 4 different approaches); (3) suggested monitoring questions for the teacher to use while circulating; (4) guidance on which approaches to select for whole-class sharing and in what order; and (5) connecting questions that help students see mathematical relationships across different approaches."
A prompt for number talks that surface strategy diversity:
"Generate number talks for Grade 3 students on multiplication strategies. Create 5 number talk problems that: (1) encourage mental computation; (2) surface multiple strategies (e.g., arrays, skip counting, doubling); (3) range from accessible to challenging; and (4) include teacher facilitation questions that reveal students' thinking without funneling to a preferred strategy. Include sample student responses and how to record them on the board."
EduGenius for Mathematics
EduGenius (edugenius.app) generates mathematics lesson sequences that balance conceptual development and procedural practice. For Grades KG-9, EduGenius creates: high-cognitive-demand tasks with multiple entry points; multiple representation sequences connecting concrete, visual, and abstract; mathematical discourse protocols for small group and whole class; and formative assessment sequences aligned to specific mathematical standards.
The credit-based system (from $7.99/month, 25 free welcome credits) makes systematic high-quality mathematics lesson development economical for teachers who want to consistently implement research-based practices.
Classroom Scenario: Rodrigo's Problem-Based Unit in Asunción
Rodrigo Benítez teaches secondary mathematics at a school in Asunción, Paraguay's capital and most populous city—a metropolitan area of approximately 2.5 million people on the Paraguay River at the border with Argentina.
A Bilingual Classroom Context
Paraguay has a distinctive linguistic feature unique in South America: it is the only country where an indigenous language, Guaraní, is spoken by the majority of the population and recognized as a co-official language alongside Spanish. Approximately 90% of Paraguayans speak Guaraní, compared to approximately 87% who speak Spanish, with significant overlap between the two groups.
This bilingualism is not merely administrative:
- Guaraní is the language of everyday life, family, market, and affection for most Paraguayans, including educated urban professionals who may use Spanish for formal and professional contexts
- Guaraní has a distinctive grammatical structure (it is an agglutinative polysynthetic language) and a rich oral literary tradition
- Its speakers have developed sophisticated mathematical vocabulary in both languages
The Guaraní-Spanish bilingualism creates an interesting mathematical education context. Word problems and mathematical reasoning that sound natural in one language may feel stilted in the other. Mathematical concepts that connect to everyday life—measurement, proportion, agricultural and commercial mathematics—may be more naturally discussed in Guaraní, while formal mathematical notation is taught in Spanish.
Rodrigo was developing a unit on proportional reasoning for Grade 8 students that connected to Paraguay's dual linguistic and cultural context. He asked EduGenius to help design problem-based tasks using culturally relevant contexts, built around three elements:
- Culturally grounded problem tasks: EduGenius generated proportional reasoning problems drawing on Paraguayan contexts—the traditional yopará dish (a mix of corn and beans with culturally specific proportions), the ratio of Spanish-to-Guaraní speakers in different regions, the proportion of landholdings in Paraguay's highly unequal land distribution (the top 1% of landowners control approximately 77% of agricultural land), and the exchange rates and commercial calculations relevant to the border economy between Paraguay and Argentina
- Multiple approach exploration: Each problem was structured for multiple solution approaches—table of values, graphical representation, unit rate calculation, and scale factor reasoning
- Mathematical discourse protocol: EduGenius generated a discussion protocol, following the Smith and Stein five practices structure, where students with different solution approaches shared their methods and the class identified connections between them
Students were encouraged to choose the entry point that made the most sense to them, embodying the principle that proportional reasoning is a conceptual space with multiple valid approaches rather than a single algorithm to memorize. Rodrigo used the discourse protocol to help students see that the table approach and the unit rate approach are mathematically equivalent, not competing methods.
Rodrigo adapted the EduGenius problems to ensure the Paraguayan contexts were accurate and culturally appropriate, and facilitated discussion partly in Guaraní when students found it easier to reason about the everyday contexts in their home language before formalizing in Spanish mathematical notation.
The Land Distribution Problem
The land distribution problem—in a country where 1% of landowners control 77% of agricultural land—generated the most productive mathematical discussion of the unit and the most student engagement with what the numbers actually meant. Students computed two scenarios:
- If Paraguay's arable land were distributed equally, how much would each farming family receive?
- If the top 1% distributed 20% of their holdings to landless farmers, how many families could receive a viable farm?
These calculations required proportional reasoning, unit conversion, estimation, and interpretation of results in context—all the mathematical practices that Schoenfeld identified as characteristic of genuine mathematical thinking.
The context also raised the normative questions that Boaler identifies as characteristic of equity-oriented mathematics: the numbers mean something about people's lives, and students who engage with that meaning are doing genuine mathematics, not just calculation.
Key Takeaways
- Skemp's relational/instrumental distinction (1976) and Hiebert/Lefevre's conceptual/procedural framework (1986) established that understanding why procedures work is not optional enrichment but the foundation of genuine mathematical competence
- NCTM's eight Mathematics Teaching Practices (Principles to Actions 2014) provide the most research-grounded framework for classroom mathematics instruction: high-cognitive-demand tasks, multiple representations, mathematical discourse, and productive struggle are central
- Smith and Stein's Mathematical Task Framework demonstrates that the cognitive demand level of tasks (memorization → procedures without connections → procedures with connections → doing mathematics) is the single most important instructional design variable in mathematics
- Ball's MKT research (2008) identifies specialized mathematical knowledge for teaching—distinct from both content knowledge and pedagogical knowledge—as a significant predictor of student learning outcomes
- Boaler's equity research demonstrates that open-ended, collaborative mathematics instruction reduces demographic achievement gaps more effectively than traditional procedural instruction, and that fixed mindset about mathematical ability is significantly influenced by classroom practices
- Paraguay's extraordinary Guaraní-Spanish bilingualism — 90% of citizens speak an indigenous language as co-official — illustrates that culturally grounded mathematics problems (drawing on everyday bilingual life) develop both mathematical reasoning and student engagement
- AI most effectively supports mathematics teaching by generating: high-cognitive-demand task sequences, multiple representation exploration activities, mathematical discourse protocols following the five practices structure, and formative assessment items that reveal conceptual understanding rather than only procedural accuracy
Frequently Asked Questions
How do I balance developing conceptual understanding with ensuring procedural fluency?
The research consensus, endorsed by NCTM and the Common Core State Standards for Mathematics, is that procedural fluency should develop from conceptual understanding—not before it or instead of it. Practically, this means:
- Introduce new concepts through exploration and sense-making before teaching the standard algorithm
- Use multiple representations to build understanding before procedural consolidation
- Design practice that develops speed and accuracy with procedures students already understand
"Drill and kill" procedural practice without conceptual grounding produces brittle, non-transferable fluency. Conceptual exploration without procedural consolidation produces understanding that can't be efficiently executed.
What is "productive struggle" and how do I know when to intervene?
Productive struggle is mathematical engagement with a problem that is challenging but not impossible—the student can make progress even if they can't immediately solve the problem.
Signs struggle is productive: students are making attempts, trying multiple approaches, re-reading the problem, and asking mathematical questions.
Signs struggle has become unproductive: students are giving up, have made no progress for an extended period, or show significant frustration without forward movement.
When intervention is needed, aim to restore productive engagement rather than remove the challenge:
- Ask a question that reveals a new entry point
- Connect to a simpler related problem
- Redirect toward a different representation
Avoid providing the solution outright or reducing the cognitive demand of the task.
How do I teach problem-solving strategies explicitly without reducing mathematical thinking to a list of steps?
Heuristics—general problem-solving strategies like "draw a diagram" or "consider a simpler problem"—are worth teaching explicitly. The key is positioning them as tools for thinking rather than algorithms to follow.
Teach heuristics in the context of genuine problem-solving:
- When students are stuck on a problem, model selecting a heuristic and applying it, thinking aloud about why it might help
- Debrief problem-solving experiences by naming the strategies used and discussing when they help
- Avoid teaching heuristics as a sequential checklist; treat them as tools to be selected flexibly based on the problem
How do I create heterogeneous mathematics groups that don't frustrate struggling students or bore advanced students?
Group work in mathematics requires thoughtful task design and role structure. For heterogeneous groups, effective practices include:
- Using tasks with multiple entry points (accessible at a basic level, rich with extension opportunities)
- Assigning specific roles that value different contributions (one role might involve explaining reasoning rather than computing)
- Establishing norms that everyone contributes and everyone can explain
- Using complex instruction techniques (Cohen 1994) to address status differences that affect participation
Research by Boaler and Staples (2008) demonstrates that heterogeneous mathematics groups with proper task design and facilitation produce better outcomes than ability-grouped instruction for students at all levels.
Can AI generate genuinely good math problems, or do they tend to be formulaic?
AI-generated mathematics problems vary significantly in quality based on prompt specificity. Generic prompts ("generate word problems") produce formulaic results.
Specific prompts that specify the following produce substantially higher quality problems:
- The cognitive demand level (using Smith-Stein terminology)
- The mathematical concept to be developed
- The desired multiple approaches
- The context to be used for real-world grounding
Even the most mathematically rich AI-generated tasks typically require teacher adaptation—checking mathematical accuracy, ensuring multiple approaches work as claimed, and calibrating difficulty to the actual class. AI generates starting points; teacher mathematical knowledge and judgment refine them.