2025

AI FOR EDUCATION
&
STUDENT ACHIEVEMENT PARTNERS

Guide to
A Guide to
Integrating
Generative AI for
Deeper
Guide toMath
Learning
Integrating
Generative AI for
Deeper Math
Learning
Guide to Integrating
Generative AI for Deeper
Math Learning
This document represents an initial exploration of how to harness the
power of generative artificial intelligence (GenAI) to enhance, not replace,
the cognitive lift and meaningful learning in K-12 math classrooms.
Developed with an understanding that both the technology and its
educational applications are rapidly evolving, this document explores
practical strategies for using GenAI to support students’ productive
struggle while maintaining the essential cognitive work that drives
learning. We offer these insights as a conversation starter and support as
educators continue to wrestle with how GenAI can serve instructional
needs and how content and pedagogy intersect with this evolving
technology.

Goals of This Resource

While ChatGPT and other GenAI tools are at the earliest stages of their development,
recent research shows teachers have begun exploring ways to integrate them into
planning and classroom instruction. Leaders and teachers are grappling with
determining how to integrate GenAI in a way that effectively supports learning and
strengthens student outcomes at a time when there is little evidence or research to
guide their approaches.

Leveraging our collective expertise in the current uses of GenAI in math classrooms and
high-quality math instruction, AI for Education and Student Achievement Partners have
created this guide focused on how mathematics educators can begin to use these tools
to enhance, not replace, students’ cognitive lift in the classroom.

This guide is a work in progress. We aim to gather input, learn alongside our teacher and
instructional leader community, refine our understanding, and amplify best practices.
We will be seeking feedback from classroom teachers who apply these strategies so that
we can collectively better understand the impact on student learning.

© 2025 AI for Education, Student Achievement Partners

 Key Assumptions
This resource assumes teachers and students have a basic foundation in AI literacy.
The suggested activities serve a dual purpose: supporting both math instruction
and the development of AI literacy skills.

This resource also assumes a shared definition of high-quality math instruction that
encompasses essential mathematics, connected mathematics, meaningful
mathematics; and student-centered mathematics. For more on this topic, see the e2
Instructional Practice FrameworkTM.

Definitions of Key Concepts

Mathematics instruction: instruction that is focused on developing students’
understanding of mathematical concepts, proficiency with procedural skills,
and demonstration of problem-solving abilities.
Cognitive lift: mental effort required to engage with content while learning to
demonstrate deep understanding and critical thinking skills.
Productive struggle: meaningful interactions with high-quality math content
that poses challenges through which to learn.

Counterproductive struggle: challenges with content or tasks that create

barriers to learning.

AI literacy: knowledge, skills, and mindsets that enable individuals to use AI

safely, ethically, and effectively.

Generative AI: subset of AI that can generate new content such as text, code,
images, audio, and video. Generative AI models are developed using massive
amounts of data, primarily from the internet, which allows them to recognize
patterns and predict what content might come next in a sequence.

Large Language Model (LLM): an artificial intelligence system trained to

recognize patterns in large amounts of text data and replicate those patterns
to generate human-like text. LLMs can perform language-based tasks such as
writing essays, summarizing texts, and translating by calculating the most
probable response to a text or voice input.

Chatbot: AI tool designed specifically to simulate conversation with human

users through text or voice interactions, often powered by LLMs.

© 2025 AI for Education, Student Achievement Partners

Productive (and
Counterproductive) Struggle
in Math Classrooms
When math instruction is structured to focus on procedural skills and memorization,
absent conceptual understanding or opportunities for student agency in selecting
problem-solving methodologies, students are more likely to internalize the beliefs that
there is one right approach to a solution, that mistakes are failures, and that peers are
competitors. When students approach math with this mindset, they are more likely to
engage in counterproductive struggle, which increases math anxiety and hinders
positive development of math identities.

Alternatively, when math instruction is focused on building conceptual understanding,

providing multiple pathways towards a solution, fostering student agency, and
leveraging student assets in approximations and partial solutions, students are more
likely to deepen engagement and build positive math identities.

The concept of “productive struggle” is often articulated in

education circles, but it is not frequently enough attended to in
terms of answering these questions:

What is a productive struggle that supports content-specific

learning?
Conversely, what is counterproductive struggle that negates
or detracts from learning?

Building on the core components of high-quality mathematics instruction, the following

chart illustrates how productive and counterproductive struggle manifest in math
classrooms. For each key area, specific examples demonstrate the crucial difference
between challenges that advance learning and those that may impede it. This chart
serves two essential purposes:

1 It provides concrete examples of how productive and counterproductive

struggle appear within each component of high-quality math instruction as a
foundation for the GenAI use cases outlined below.

2 It serves as a practical tool for teachers and leaders to reflect on and

evaluate struggles in their own math classrooms, helping them ensure their
instructional choices promote meaningful learning.

© 2025 AI for Education, Student Achievement Partners

Essential Tasks offer multiple entry Scaffolds that “make sense”
Mathematics: points and scaffolds when of the problem for students
necessary to access the (e.g., providing prompts that
Engage in cognitively grade- or course-level break down the steps of a
challenging mathematics content. multi-step word problem,
that focuses on grade- providing fill-in-the-blank
level or course-level Tasks allow for multiple number sentences that
content standard(s) pathways so that students represent the problem).
can demonstrate
understanding of grade-level
mathematics concepts in a
variety of ways.

Connected Tasks leverage the Practice is of procedural

Mathematics: coherence of math concepts skills or fluency, absent
across and within grades and conceptual understanding.
Relate new content to courses.
math content within or Instruction prioritizes only
across grades that builds Instruction builds on answer-getting and/or
towards future students’ prior math solving a problem in one
mathematical concepts; knowledge, ideas, way.
build procedural skill and assets, and goals as they
fluency from conceptual develop grade-level and
understanding course-level understanding.

Students share their

reasoning in multiple ways,
including using multiple
representations.

Meaningful Tasks center on current Supports “simplify” the

Mathematics: events relevant to students’ problem (e.g., removing
lives and that can be context, highlighting key
Engage in mathematical analyzed mathematically. words/phrases) as a way of
reasoning, discourse, making the problem more
communication, problem- Tasks connect with the real accessible.
solving, and modeling world and provide
while engaging in the math opportunities for students to Tasks do not include data or
content; position students develop skills and knowledge contexts relevant to
as the mathematical relevant to their academic students and their
experts, thinkers, and and professional goals. communities when they are
doers in the classroom exploring mathematical
concepts.

© 2025 AI for Education, Student Achievement Partners

 Supports “simplify” the Students are forced to only
problem (e.g., removing make meaning within a
context, highlighting key problem’s provided context
words/phrases) as a way of (such that students’
making the problem more identities, communities and
accessible. cultures are not valued).

Student-Centered Instruction provides access Scaffolds/supports select

Mathematics: to math tools and offers the tools for students to use,
ample time to explore math provide the representation
Engage in mathematical concepts. for students to use, and/or
reasoning, discourse, dictate the steps that
communication, problem- Students co-construct, students will take.
solving, and modeling share, and own the
while engaging in the math mathematical ideas Teachers are the only voices
content; position students produced in the lesson. summarizing the lesson
as the mathematical takeaways.
experts, thinkers, and Students build on one
doers in the classroom another’s shared ideas and Instruction provides little or
math knowledge and ask no opportunities for student
questions about each other’s discourse; students are
thinking. discouraged or prohibited
from utilizing home and
Instruction positions everyday language to learn
multiple languages and mathematics.
cultures as valuable assets.

© 2025 AI for Education, Student Achievement Partners

Enhancing Math
Instruction with GenAI
GenAI can be viewed as a tool for extending thinking, similar to the view of calculators
in mathematics instruction. Over time, norms and standards for using calculators in
ways that support meaningful engagement with computation have been developed.
Just as math instruction must be carefully designed to position the calculator as a tool
to support productive struggle and not as a shortcut, GenAI tools must be integrated
with similar intentionality—considering how these tools can enhance rather than
replace student access to productive struggle.

Specialized AI Tools for Mathematics

While GenAI tools show definite promise in enhancing mathematics education, our field
tests have revealed significant limitations1 in calculation reliability 2 and accuracy,
particularly in general-purpose (foundational) models like ChatGPT. These Large
Language Models (LLMs) operate by predicting the next word in response to a prompt,
not by performing actual calculations.

Therefore, it is our strong recommendation that educators

use specialized tools such as Khanmigo or Desmos for any
task involving computation.

These tools have been developed with additional fine-tuning to support numerical
calculations beyond the capabilities of foundational LLMs. They also typically have
structured interfaces that support the input of mathematical symbols and equations.
Foundational LLMs are most useful for language-based tasks such as lesson planning,
conceptual exploration, and generating alternative explanations and examples for
students.

1
https://www.forbes.com/sites/johnwerner/2024/10/07/ai-is-usually-bad-at-math-heres-
what-will-happen-if-it-gets-better/
2
https:/arxiv.org/pdf/2503.21934v1

© 2025 AI for Education, Student Achievement Partners

 AI Tool Categories and Their Optimal
Uses
Foundational Large Language Models
Foundational LLMs such as ChatGPT, Claude, Google Gemini, and Microsoft Copilot are
most suitable for language-based tasks, and should never be relied upon for
computational accuracy. These models excel at:

Brainstorming: generate alternative teaching strategies, creative ways to

address standards, or metaphors to clarify abstract mathematical concepts.
They can also act as a conversational thought partner, helping educators
explore different instructional approaches, refine ideas through back-and-
forth dialogue, and consider perspectives they might not have otherwise
thought of.

Designing innovative projects and assignments: generate ideas or outlines

for projects, tiered assignments, or performance tasks that integrate
mathematical thinking with real-world applications. They can suggest
themes, prompts, or scaffolding structures based on standards and learning
goals.

Drafting student-friendly communications: draft clear definitions, age-

appropriate learning objectives, and simplified explanations of complex
terms or procedures. This can help make math content more accessible and
relatable for students.

Identifying common misconceptions: list typical errors or partial

understandings that students might have with a given concept (e.g.,
confusing the numerator and denominator in fractions), along with
suggestions for addressing these issues through targeted instruction or
scaffolding.

Translating: translate word problems or teacher communications such as

emails, newsletters, and assignment instructions into other languages. Note-
it is recommended that a native speaker review the output whenever
possible to ensure cultural and linguistic accuracy

© 2025 AI for Education, Student Achievement Partners

 Specialized Tools
Specialized AI-powered math tools are purpose-built applications that integrate AI
capabilities specifically for mathematical tasks. While powered by LLMs, these tools
receive additional fine-tuning during the development process that allow them to perform
actual calculations rather than merely predict text patterns. Khanmigo, for example,
incorporates a calculator in its operations rather than relying solely on LLM outputs.
These tools have received additional training specifically on mathematical concepts, and
often contain specialized interfaces that can accommodate symbols, notation, and
structured equations. For any task involving calculations, specialized math tools should
be the preferred choice for both teachers and students.

Disclaimer: the tools listed below are provided only as examples of AI-powered math tools.
This list is not exhaustive, and inclusion of a tool does not constitute endorsement. We
encourage educators to explore a range of tools to find those that best fit their instructional
goals and students’ needs.

Khanmigo: an AI-powered tutor developed by Khan Academy, integrated

with ChatGPT-4. Offers conversational, Socratic-style guidance to help
students think through problems rather than simply providing answers.
Mathway: provides step-by-step solutions to a wide variety of math
problems, from basic arithmetic to calculus. Supports photo uploads of math
problems.
Photomath: mobile app that supports photo uploads of handwritten or
printed math problems. Provides animated, step-by-step explanations.
Gauthmath: AI-powered step-by-step problem solver with human tutor
support when needed. Supports photo and PDF uploads of math problems.
Snorkl: formative assessment tool that analyzes student work, identifies
misconceptions, and provides actionable insights for teachers. Allows
students to insert voice recordings of their thinking as they work through
problems on the platform.
Mathos AI: provides both solutions (Math Solver) and step-by-step AI
tutoring (Math Tutor). Supports photo and PDF uploads, as well as voice
inputs.
Wolfram Alpha: an advanced computational engine that solves equations,
visualizes data, and integrates math with science, engineering, and other
fields to explore real-world math applications. Provides a wide variety of
practice problems.
Julius AI: a generative AI assistant for analyzing and interacting with data.
Allows teachers to ask questions of data sets directly using natural language.

© 2025 AI for Education, Student Achievement Partners

Key Principles for GenAI Integration
When leveraging AI tools in math instruction, consider the following:

Tools should support, not circumvent, productive struggle for

students.

Integration should enhance, not replace, proven instructional

practices.

Usage should align with students’ developmental readiness and

mathematical learning goals.

GenAI should augment educators’ pedagogical expertise,

content knowledge, and knowledge of students.

LLMs should be used for language-based tasks like conceptual

exploration, planning, and generating examples.

Specialized, math-specific tools should be used for tasks

requiring computation.

A way to consider meaningful use of GenAI that supports student learning is to

move away from activity- and technology-driven tasks, and instead focus on
student learning. Ask:

What are students learning through this task?

Where is productive struggle essential to student learning?

Where might GenAI benefit learners or learning outcomes?

Where could it mitigate counterproductive struggle?

© 2025 AI for Education, Student Achievement Partners

A Note About GenAI Use for Elementary
Teachers
At the elementary level, GenAI tools should be used primarily by teachers for lesson
planning, preparation, and selective modeling. This use aligns with elementary
students' developmental stages—when children are still building foundational
cognitive skills like causal reasoning, source evaluation, and metacognitive
awareness. These developmental factors make it premature for young students to
independently evaluate AI-generated content or critically assess its limitations.

Until there is compelling evidence of positive impacts on learning through direct

student interaction with GenAI tools at the elementary level, as well as adequate
safeguards in place to completely eliminate the risk of exposure to inappropriate
content, elementary teachers are advised to restrict GenAI use to their own
workflows. This is an evolving field, and while there may be some direct interaction
that is supportive of students, we encourage caution here.

Many GenAI tools impose age restrictions that prevent elementary students from
engaging with them directly. Those that do allow elementary use require close
oversight and adult guidance to help students interpret responses critically.

Strategies for Using GenAI to Support Deeper

Math Learning
Following are examples of effective and ineffective applications of GenAI, including
worked chatbot transcripts that illustrate these practices.

Reminder: Use specialized tools as described above for any tasks involving
calculations, equation solving, or mathematical modeling. General-purpose
language models like ChatGPT or Claude are not designed to perform reliable
mathematical computations and may produce incorrect results.

Within the following tables, student use cases are highlighted. Examples that
assume direct student use are targeted to middle and high school students with
parental permission.

© 2025 AI for Education, Student Achievement Partners

For Students

Engage with AI AI math tutoring Positions AI as a tool Avoid using general-

tools have been to help students purpose language
Tutoring think through models such as
specially developed
problems and reveal ChatGPT. Use
to help the students
Worked example various entry points specialized tools (see
think through to reinforce list above).
with problems, asking underlying
computational questions and mathematical Recognize that
errors providing feedback processes responses are limited
to guide thinking to the tool’s interface
Provides private and training data.
without providing forum for asking
Use specialized
answers directly. questions without Structure use of AI
tools fear of judgment tutors to encourage
self-reflection and
Offers 24/7 access to gradual skills release;
feedback and strive for students to
support shift to self-checking
after mastering the
concept under study
to prevent over-
dependence.

Ensure the AI tutor is

aligned with specific
curriculum to avoid
conflicts

AI can check Scaffolds conceptual Avoid using general-

Conduct error completed student understanding by purpose language
analysis and work for accuracy, targeting specific models such as
areas of breakdown ChatGPT. Use
feedback provide immediate
specialized tools (see
feedback, and Provides a safe space above).
Use specialized answer questions, for students to
guiding students to a identify gaps in their Recognize that
tools clear understanding understanding and responses are limited
of where and how ask clarifying to the tool’s interface
questions without and training data.
errors occurred, as fear of judgment
well as how to
correct them.

© 2025 AI for Education, Student Achievement Partners

For Students (continued)

AI can generate a Provides Note risk of cognitive

step-by-step opportunities for offloading and over-
explanation of a students to explore dependence; structure
multiple AI interactions in the
solution to a math explanations of manner of tutoring
problem and prompt mathematical sessions by explicitly
students to concepts that can be prompting the model
understand the easily tailored to not to provide direct
process behind an their interests and answers, but rather to
answer. experiences guide students’
thinking to determine
Allows for analyzing the correct solution
AI can prompt drawings and themselves when they
students to do an handwritten get stuck
error analysis of a equations through
worked example multimodal
functionality.
(with errors).

Practice with Students can scale Increases access to Avoid using general-
up their practice practice purpose language
procedural skills opportunities for models such as
with procedural skills
specific skills ChatGPT. Use
by using AI to
Use specialized specialized tools (see
generate Invites scaffolded above).
tools personalized practice that build
problems related to on prior math Recognize that
the current skills knowledge and/or responses are limited
under study. skills and connect to to the tool’s interface
grade-level content and training data.

Allows students to Note risk of cognitive

practice in a time offloading and over-
and place of their dependence; structure
choosing AI interactions in the
manner of tutoring
Allows students to sessions by explicitly
co-construct prompting the model
meaningful practice not to provide direct
problems by answers, but rather to
customizing the generate relevant
content to their practice problems and
background guide students’
knowledge and thinking to determine
interests the correct solution
themselves when they
get stuck.

© 2025 AI for Education, Student Achievement Partners

For Teachers

Create lessons, AI can serve as a Adds capacity to the Carefully screen AI-
thought partner to lesson planning generated content for
materials, and process bias and
help teachers
interventions developmental
develop lesson Ensures alignment of appropriateness
plans, instructional interventions and
Worked example content, and extensions to Since lack of
targeted primary learning contextual awareness
interventions or objectives can lead to
extensions by misalignment with
Generates adopted curricula and
suggesting activities suggestions for department/school/
aligned to learning diverse examples district goals, ensure
objectives and and scenarios these are in alignment
filtered through the connected to real- before deploying AI-
teacher’s world math generated content
applications with students
professional
judgment. Redesigns
instruction and
assessment to
incorporate math
practices such as
mathematical
reasoning and
communication

Generate AI can create highly Quickly creates Always check AI-

contextual practice engaging, relevant generated problems
authentic practice problems at for accuracy and
problems at volume,
practice problems incorporating real-
scale solution viability
world scenarios, Generates complex Guard against bias and
Use specialized student interests, problem sets that cultural
tools and standardized provide multiple misrepresentation in
test referents. pathways to AI-generated word
demonstrate problems
understanding
Ensure that AI-
Creates coherent generated problems
assessments aligned are in alignment with
to reference formal curriculum and
documents like learning objectives
curricula, standards,
and standardized
test exemplars

© 2025 AI for Education, Student Achievement Partners

For Teachers (continued)

AI can assist with Creates materials Ensure that

Provide with multiple entry modifications
developing modified
differentiation/ assignments, points for diverse maintain grade-level-
Individualized learners appropriate task
alternative complexity and
instruction assessment Facilitates efficient rigorous learning
methods, and scaffolding of expectations
differentiated instruction for a
Worked example variety of readiness Always check AI-
teaching
strategies/explanatio levels and generated problems
backgrounds for accuracy and
ns based on specific solution viability
student learning Brainstorms varied
needs. approaches to Ensure that AI-
content presentation generated materials
and alternative are in alignment with
teaching strategies, formal curriculum and
examples, and learning objective
explanations to
facilitate student
understanding

Break down tasks AI can help unpack Assists with Ensure that AI-
challenging tasks and activating students’ generated content is
and concepts prior math accurate, aligned to
concepts by
knowledge by learning objectives
generating related planning activities
Worked example and established
content such as that activate trajectories, and free
prerequisite skills, prerequisite skills from bias
warm-up questions,
common errors or Anticipates and
partial proactively plans to
address common
understandings, errors and partial
helpful understandings
interventions, and
other pedagogical Brainstorms lists of
strategies. hints and alternative
explanations for
specific problem
types that do guide
students who are
stuck without giving
away answers
directly

Offers insights of
how to break down
complex problems
into manageable
chunks to facilitate
understanding

© 2025 AI for Education, Student Achievement Partners

For Teachers (continued)

Analyze student AI can analyze Assesses student Risk of compromising

student data and/or performance on student data privacy;
data and student close-ended never enter students’
student work to
work assessment tasks personally identifiable
identify trends and and identifies information into an AI
action items related trends/action items tool; always
Use specialized to topics such as anonymize data before
tools student thinking, Efficiently scores uploading
common errors and constructed
partial response data, Be aware of biases
including drawings inherent to the
understandings, training data that can
areas of strength, the Identifies and awards be reflected in action
effectiveness of a partial credit for items (e.g., small
given teaching correct thinking in sample sizes for
strategy, student some areas of an marginalized student
assessment task, populations can lead
interests, and more.
providing deeper to overestimation or
insight into student underestimation of
understanding their performance)

Create graphs, AI can generate Enhances concept Ensure that digital

visual visualization interactives are
charts, and accessible to all
representations of
interactives Supports multiple students, including
data and become an representations of those using assistive
interactive digital mathematical technology such as
Use specialized tools to illustrate concepts and data screen readers or
tools mathematical alternative input
concepts for Supports a wider devices
instruction and variety of
instructional content Verify accuracy and
assessment creation alignment to
curriculum and
learning objectives

© 2025 AI for Education, Student Achievement Partners

Who We Are

AI for Education is committed to leading the responsible adoption of

Generative AI in education by providing comprehensive AI literacy training
and AI guidance and policy development to schools and educators worldwide.
As a pioneering organization that has partnered with some of the largest
school districts across the U.S., they develop and publish a wealth of free,
practical, and innovative resources while working with academic institutions
to conduct training, develop curriculum, define policy, and implement
customized AI solutions. With the goal of providing AI literacy training to 1
million educators, their mission focuses on empowering teachers to expertly
navigate AI technology, ultimately improving student outcomes and preparing
them for the future.
aiforeducation.io

Student Achievement Partners (historically known as Achieve the Core), is a

central voice in the conversation about high-quality instructional practices
and materials in K12 Mathematics and Literacy in the US. Founded in 2011,
members of the original SAP team were the lead writers of the Common Core
State Standards and began developing tools and resources designed to give
more students access to high-quality, standards-aligned math and literacy
instruction. As a team of former educators, SAP understands that the most
useful and effective tools and guidance are grounded in both the latest
pedagogical research and educator insights
2 into what works in the classroom.
In the last four years, SAP has developed an expanded framework for
instructional excellence, e .

learnwithsap.org