AI in Education: A Practical Guide to Implementation in Schools

AI in education has moved from curiosity to commitment. Districts are writing policies. Schools are piloting tools. Teachers are experimenting—some enthusiastically, some reluctantly, most with questions no vendor demo fully answers. The conversation has shifted from Should we use AI? to How do we implement it in ways that actually improve teaching and learning?

That second question is harder than the first, and most of the current guidance doesn’t address it well. The resources tend to fall into two camps: breathless product announcements that overpromise, or cautious policy documents that Under-specify. What schools need sits between those extremes—a structured approach to AI implementation that starts with pedagogy, accounts for context, and builds capacity over time.

This guide lays out that approach. It is built on four dimensions of AI implementation—leadership, coaching, tool evaluation, and special education—and grounded in what we know about how adults learn new practices and how schools actually change. It draws on cognitive science, implementation research, and the realities of classroom instruction rather than vendor roadmaps or speculative futures.

AI in Education: Where We Are Now

The adoption numbers tell a story of uneven momentum. According to the 2024 Walton Family Foundation survey, over 60% of teachers reported using AI tools at least occasionally in their planning or instruction. The Consortium for School Networking’s 2024 State of EdTech report found that nearly half of districts had adopted or were piloting some form of AI-enabled tool. Usage is happening. Alignment is not.

Most AI adoption in schools right now is tool-driven, not strategy-driven. A teacher hears about ChatGPT from a colleague, tries it for lesson planning, and begins using it personally. That is adoption, but it is not implementation. Implementation implies intentionality: clear goals, deliberate integration into existing workflows, evidence of impact, and structures for sustainability. By that standard, genuine AI implementation in schools remains early-stage at most institutions.

The gap between hype and reality is significant. AI can already support teachers in drafting materials, generating formative assessment items, and differentiating content for learners at different levels. These are real efficiencies. But AI cannot replace professional judgment about what students need, design coherent curricula on its own, or substitute for the relational work of teaching. The tools are useful. The narrative that they will transform education on their own is not.

This matters because how schools enter the AI conversation shapes where they end up. Districts that start with What should we buy? tend to accumulate disconnected tools. Districts that start with What problems are we solving and how would we know if AI helped? tend to build something that lasts. AI implementation in schools requires the second orientation. Tools change. Frameworks for thinking about tools endure.

What Schools Actually Need from AI

The typical starting point for AI in education is a tool list: “Top 20 AI Tools for Teachers.” These lists serve a purpose—they raise awareness and reduce the search cost of discovery. But they also reinforce a fundamentally reactive posture. The question becomes Which tool should I pick? rather than What does my school need AI to do, and how will we measure whether it’s working?

What schools actually need is an evaluation lens—a way to assess any AI tool or practice against criteria that reflect their values, constraints, and instructional priorities. Dr. Matt Rhoads uses a framework called VATT to structure that evaluation:

• Value: Does this AI application solve a real problem that teachers or students experience, or does it create new work without clear benefit? Value means starting from the instructional need, not the tool’s feature set.
• Accessibility: Can all intended users—teachers, students, families—actually use this tool with the devices, connectivity, language support, and technical skill they have? Accessibility failures are implementation failures.
• Trust: Is the tool transparent about how it uses data? Does it produce reliable outputs? Can educators verify what it generates? Trust is not abstract—it is built or eroded in every interaction teachers have with a tool.
• Teaching Quality: Does this use of AI preserve or enhance the instructional core—meaningful teacher-student interactions, cognitively demanding tasks, formative feedback? If AI saves time but degrades instruction, it fails the most important test.

VATT is not a checklist you apply once. It is a mindset that shifts the conversation from Is this tool good? to Is this tool right for our context, our students, and our goals? That distinction matters because the answer changes by school, by subject, and by the adults who will use it.

This lens also exposes the limits of tool-first thinking. A tool that scores well on VATT for one district’s coaching model may fail in another district where teachers have less release time for collaboration. Implementation is local. The framework ensures you are asking the local questions.

Four Dimensions of AI Implementation

AI in education is not one thing. It looks different depending on your role, your context, and the problems you are trying to solve. Trying to address it as a single initiative almost always produces shallow adoption: a few enthusiasts use tools deeply while most colleagues barely engage.

A more productive approach is to organize AI implementation around four dimensions, each with its own stakeholders, competencies, and success measures. These four dimensions correspond to the sub-pillars of this guide and represent the areas where schools need deliberate strategy, not just permission to explore.

AI for District Leaders: Policy, Ethics, and Measuring Impact

District leaders face a different set of questions than teachers do. Their job is not to pick the best AI tool for tomorrow’s lesson. It is to create the conditions where AI implementation in schools can proceed responsibly, equitably, and at a pace that matches capacity rather than pressure.

The core responsibilities for leaders include:

• Developing usable AI policy — Not a one-page aspirational statement, but guidance that specifies what teachers can and cannot do with AI, what data protections are required, and how emerging tools will be evaluated before adoption. Policy should enable thoughtful experimentation, not just restrict risk.
• Building implementation teams — AI adoption affects IT, curriculum, special education, and facilities. No single department owns it. Cross-functional teams with clear roles and decision-making authority prevent both paralysis and fragmented adoption.
• Measuring impact — Most districts cannot answer the basic question: Is our AI investment making a difference? Defining success metrics before launch—teacher time saved, student engagement, differentiation quality, assessment efficiency—creates accountability without overstandardizing.
• Establishing ethical frameworks — AI ethics in education is not just about data privacy (though that is essential). It includes questions about algorithmic bias, equitable access, the displacement of professional judgment, and the responsibilities districts have when AI tools interact with minors.
• Communicating with stakeholders — Parents, school boards, and community members have legitimate concerns about AI in classrooms. Proactive, clear communication about what AI is doing and not doing builds the trust that implementation requires.

The AI School Leadership Guide provides a deeper framework for leaders navigating these decisions, including policy templates, team structures, and impact measurement tools.

Related reading: How Data Decision-Making and AI Can Future-Proof Your Organization · The Rise of In-House AI Built Learning Applications · Creating AI Agent Safeguards

AI for Instructional Coaches: Building Teacher Capacity

If district leaders create the conditions for AI implementation, instructional coaches do the daily work of making it live in classrooms. Coaches are the bridge between policy and practice—and that bridge is where most AI initiatives either take hold or stall.

AI changes the coaching conversation in productive ways. Coaches can use AI to:

• Prepare observation feedback — AI-generated summaries of lesson observations give coaches a starting draft, freeing them to focus on the interpretive and relational work that only a human coach can do.
• Co-plan differentiated instruction — When a coach and teacher use AI together to generate multiple entry points for a lesson, they are not outsourcing pedagogy. They are accelerating the iterative work that good planning already requires.
• Demonstrate cognitive science strategies — AI tools can produce retrieval practice activities, interleaved problem sets, and worked examples on demand. The AI Instructional Coaching Guide details how coaches can use these outputs within the I Do/We Do/You Do instructional framework to model, co-construct, and release teacher independence.
• Support data-driven reflection — Coaches can help teachers analyze patterns in AI-generated student feedback, assessment results, or engagement data without replacing the teacher’s judgment about what those patterns mean.

The critical insight is that AI does not replace the coaching relationship—it amplifies it. But only if the coach has a framework for integrating AI into their practice rather than adding it as an extra task. The I Do/We Do/You Do cycle, adapted for adult learning, provides that structure (detailed in the Implementation Cycle section below).

Related reading: Using AI for Retrieval Practice · Using AI for Interleaving, Spaced Practice, and Retrieval · Boost Student Learning with Interactive Worked Examples · Using Generative AI for Planning Workflow

AI Tools Evaluation Framework: Choosing Systematically

The sheer volume of AI tools marketed to schools is overwhelming. Every week brings new options, and the feature sets blur together. Teachers and leaders need a systematic way to cut through the noise—not another top-ten list, but a repeatable process for evaluating whether a tool deserves a place in their classroom.

An effective evaluation framework should address five dimensions:

• Pedagogy — Does the tool support evidence-based instructional strategies, or does it default to low-level tasks (fill-in-the-blank, recall-only, passive consumption)? The best AI tools for teachers extend what is instructionally possible, not just what is fast.
• Privacy — What data does the tool collect from students and teachers? Where is it stored? Who has access? Does the tool comply with FERPA, COPPA, and your district’s data governance policies? Privacy is non-negotiable, and the burden of proof should be on the vendor.
• Accessibility — Does the tool work across devices commonly available in your school? Does it support multiple languages? Is it usable for students with visual, auditory, or motor impairments? Accessibility is not a nice-to-have; it is a legal and ethical requirement.
• Evidence — Has the tool been tested in contexts similar to yours? Are there peer-reviewed studies, pilot data, or credible case studies? Absence of evidence is not evidence of absence, but it is a reason for caution and small-scale piloting before broad adoption.
• Integration — Does the tool connect with your existing systems—LMS, SIS, Google Workspace, assessment platforms? Or does it create a new silo that teachers must manage separately? Tools that do not integrate become tools that get abandoned.

The AI Tools Evaluation Guide provides a structured rubric for applying these five dimensions, including scoring criteria and decision protocols for moving from evaluation to pilot to scale.

Related reading: The Best AI Tools for Teachers · What’s All the Talk About ChatGPT · 5 Ways Agentic AI Can Transform Your Teaching Workflow

AI and Special Education & Co-Teaching: Meeting Vulnerable Populations Where They Are

AI in education has a particular responsibility toward students who are most often underserved by one-size-fits-all approaches: students with IEPs and 504 plans, multilingual learners, and students in co-taught classrooms where differentiation is not optional but structural.

The opportunity is real. AI can:

• Streamline IEP workflows — Drafting present levels of performance, generating goal options, and writing accommodation suggestions are time-intensive tasks where AI provides legitimate efficiency gains. The key is using AI to produce initial drafts that teachers then refine based on their knowledge of the student—not replacing professional judgment with generated text. Amplifying Special Education Teachers’ IEP Workflow with AI Tools details the process.
• Support differentiation in co-taught classrooms — Co-teachers can use AI to generate tiered materials, adapted texts, and scaffolded tasks for the same lesson objective. This is not new pedagogy; it is existing co-teaching strategy accelerated by AI. The AI Special Education & Co-Teaching Guide shows how to integrate AI into established co-teaching models without disrupting the partnership dynamic.
• Serve multilingual learners — AI translation, vocabulary scaffolding, and content adaptation tools can lower linguistic barriers to grade-level content. But careful evaluation is essential: AI translations are not always accurate, cultural nuance is often lost, and over-reliance on translation can limit language development. Amplifying Multilingual Learners with AI Tools provides guardrails.
• Navigate ethical complexity — Using AI tools with vulnerable populations raises questions that do not arise in general education contexts. Bias in algorithmic recommendations, reduced human oversight, data sensitivity around disability and language status, and the risk of automating care that requires relational judgment—these require explicit ethical frameworks, not just good intentions.

The AI Special Education & Co-Teaching Guide provides comprehensive guidance on all of these dimensions, including practical workflows and ethical considerations specific to vulnerable populations.

Related reading: Amplifying Special Education Teachers’ IEP Workflow with AI Tools · Amplifying Multilingual Learners with AI Tools · #WhyKnowledgeMatters Podcast: Co-Teaching Evolved

The Implementation Cycle: I Do, We Do, You Do for AI Adoption

Most AI professional development fails because it tells teachers about AI rather than helping them learn AI. A presentation on ChatGPT’s capabilities produces awareness but not competence. What produces competence is structured practice with scaffolded support—the same principle that works for students learning new skills.

The I Do/We Do/You Do framework, widely used in explicit instruction, adapts naturally to adult learning and AI adoption. Here is how it works when the skill being taught is AI integration:

I Do: The Leader or Coach Models

In this phase, the instructional leader or coach demonstrates an AI workflow while making their thinking visible. This is not a polished demo; it is a think-aloud. The modeler shows what they prompt, why they prompt it that way, what the AI generates, where they revise, and what they ultimately use versus discard.

The cognitive science rationale: worked examples reduce cognitive load during initial skill acquisition (Sweller, 1988; Renkl, 2014). Learners who see a complete model of a process before attempting it themselves perform better on transfer tasks than learners who attempt the task unassisted. This applies to teachers learning AI just as it applies to students learning to solve equations.

What modeling looks like in practice:

• A coach demonstrates using AI to draft a differentiated reading passage, narrating their prompt design, their revisions, and their final instructional decisions.
• A principal shares how they used AI to prepare a data summary for a board meeting, including the prompts that worked and the outputs they corrected.
• A tech coordinator models building a retrieval practice activity with AI, showing the iteration process rather than the polished result.

We Do: The Coach and Teacher Co-Plan

In this phase, coach and teacher work together on an AI-integrated task. The coach provides scaffolding—suggesting prompts, questioning output quality, helping evaluate whether the AI-generated result meets the instructional goal—but the teacher is doing the work alongside the coach.

The cognitive science rationale: This is guided practice, or the “fading” phase of scaffolding. Research on cognitive apprenticeship (Collins, Brown, & Newman, 1989) shows that learners need a supported practice phase where they can make errors in a low-stakes environment, receive immediate feedback, and gradually assume more responsibility. The deliberate practice literature (Ericsson, 2006) reinforces that skilled performance develops through focused repetition with feedback, not exposure alone.

What co-planning looks like in practice:

• A teacher and coach use AI together to generate three versions of an assessment item at different Depth of Knowledge levels, then discuss which versions are usable and why.
• A co-teaching pair uses AI to draft accommodations for an upcoming lesson, then evaluates each accommodation against the student’s actual IEP goals.
• A department chair and teacher co-create a prompt protocol for using AI in lesson planning, testing it on real upcoming lessons and revising based on results.

You Do: The Teacher Integrates Independently

In this phase, the teacher uses AI independently in their practice, applying the frameworks and judgment they developed during modeling and co-planning. The coach remains available for consultation but is no longer co-constructing every AI interaction.

The cognitive science rationale: This is the independent practice phase, where retrieval from long-term memory is strengthened through application without external supports. The gradual release of responsibility model (Pearson & Gallagher, 1983) predicts that learners who have moved through structured modeling and guided practice are more likely to transfer skills to new contexts than learners who jump directly to independent use.

What independence looks like in practice:

• A teacher routinely uses AI to generate rough drafts of instructional materials, then applies their own quality filter before using them in class—and can articulate what that filter includes.
• A teacher adapts their AI workflow for new content areas or student needs without waiting for a coach to show them how.
• A teacher begins innovating—discovering new AI applications for their specific classroom that the coach never modeled—because they understand the underlying principles, not just the steps.

The full cycle takes time. It is not a single workshop or a one-hour coaching session. It is a process that unfolds over weeks and months, with each phase building capacity for the next. Schools that try to skip to the You Do phase—handing teachers a tool and asking them to innovate—typically see shallow adoption and quick abandonment. Schools that invest in the full cycle build durable AI practices that persist after the initial enthusiasm fades.

Related reading: Boost Student Learning with Interactive Worked Examples · Autonomous AI and the Possible Future of AI in Schools · Harnessing Generative AI for a Comprehensive WASC Accreditation Visit · The Death of the LMS in Higher Ed

Starting Points: Where Is Your School?

There is no single right entry point for AI in education. Where you start depends on where you are. These self-assessment questions can help you locate your school or district on the implementation continuum—and identify the guide that addresses your most pressing needs.

If you are a district leader:

• Does your district have a written AI policy that specifies what teachers can do, not just what they cannot?
• Have you identified a cross-functional AI implementation team with clear decision-making authority?
• Can you name three measurable outcomes you expect from your AI investments this year?
• Do you have a process for evaluating AI tools before pilots launch, or are pilots the evaluation?

If these questions highlight gaps, start with the AI School Leadership Guide.

If you are an instructional coach or teacher leader:

• Can you model an AI workflow for a teacher and articulate why you made each decision in the process?
• Do you have a framework for co-planning with AI that goes beyond “try this prompt”?
• Are you integrating AI into the coaching cycles you already use, or are you treating AI coaching as a separate initiative?

If these questions reveal that your coaching practice needs structure, start with the AI Instructional Coaching Guide.

If you are a teacher evaluating AI tools:

• Do you have criteria beyond “it seems cool” or “a colleague recommended it”?
• Can you assess an AI tool’s privacy practices, accessibility features, and evidence base?
• Does the tool connect to your existing workflow, or does it create a new one?

If these questions suggest your evaluation process is informal, start with the AI Tools Evaluation Framework.

If you work in special education, co-teaching, or with multilingual learners:

• Have you considered the ethical implications of AI-generated IEP content or automated accommodations?
• Can you use AI to differentiate within your co-teaching model without disrupting the partnership?
• Are you evaluating AI tools for bias and cultural sensitivity before using them with vulnerable populations?

If these questions identify blind spots, start with the AI Special Education & Co-Teaching Guide.

Getting Support

Implementing AI in education is not a solo project, and the challenges are often more organizational than technical. Dr. Matt Rhoads works with schools and districts at every stage of this process.

• Consulting services — Strategic planning for AI implementation, including policy development, team structures, coaching frameworks, and tool evaluation.
• Speaking engagements — Keynotes and workshops on AI in education, instructional coaching, co-teaching, and data-driven decision-making.
• 25 Tips for Instructional Coaches and Leaders — Matt’s book provides actionable strategies for coaching and leadership, including integrating technology into instructional practice.

Further Reading

The following posts explore specific topics covered in this guide in greater depth:

• 5 Ways Agentic AI Can Transform Your Teaching Workflow
• Amplifying Multilingual Learners with the Support of AI Tools
• Amplifying Special Education Teachers’ IEP Workflow with AI Tools
• Autonomous AI and the Possible Future of AI in Schools
• Boost Student Learning with Interactive Worked Examples
• Creating AI Agent Safeguards
• Harnessing Generative AI for a Comprehensive WASC Accreditation Visit
• How Data Decision-Making and AI Can Future-Proof Your Organization
• The Best AI Tools for Teachers
• The Death of the LMS in Higher Ed
• The Rise of In-House AI Built Learning Applications
• Using AI to Build Powerful Retrieval Practice Activities
• Using AI to Support Interleaving, Spaced Practice, and Retrieval
• Using Generative AI to Support Teachers’ Planning Workflow
• Week of AI Conference: Revolutionize Data-Driven Decision-Making
• What’s All the Talk About ChatGPT
• #WhyKnowledgeMatters Podcast: Co-Teaching Evolved

AI in education will not be defined by the technology itself. It will be defined by the systems schools build to use it well—the policies that guide it, the coaching that develops it, the evaluation criteria that filter it, and the ethical commitments that protect the students it serves. The tools will change. The need for implementation that is intentional, evidence-grounded, and centered on teaching quality will not.

Common Barriers and Challenges

The lesson: Starting with a tool instead of a problem is how districts end up with expensive subscriptions nobody uses. The readiness assessment exists because “we need AI” is not a problem statement.

The lesson: A pilot that expands before evidence is not a pilot — it’s a purchase. This pattern plays out in district after district, which is why effective AI policy must explicitly define what constitutes a pilot — and what evidence is required before any expansion can occur.