General AI vs Domain AI: Why Specialized Models Are Winning in Education

In education, the question is no longer whether AI belongs in the classroom. It is which kind of AI belongs there, and on whose terms.

When ChatGPT launched, schools scrambled to respond. Some banned it. Some embraced it. Most did both at different times. The first wave of AI in education was defined by general-purpose tools — ChatGPT, Google Gemini, Claude — being used by students and teachers for everything from essay writing to lesson planning. These tools were powerful, versatile, and completely undesigned for the educational context.

That is changing. Google launched Gemini for Education with over 30 new classroom capabilities, enterprise-grade data protection, and teacher controls — now available free of charge across all Google Workspace for Education editions. By 2025, Google had integrated Gemini for Education into over 1,000 US higher education institutions, and its education AI tools reached 10 million students. Pearson partnered with Google to create personalized learning tools powered by AI models embedded in Pearson's curriculum products, especially for K-12. The OECD's Digital Education Outlook 2026 calls for generative AI systems designed with teachers so they can monitor student interactions and learning outcomes. Khan Academy's Khanmigo, Duolingo's AI tutor, and a growing number of EdTech platforms are building AI that is purpose-built for learning.

The pattern is clear: education is moving from general AI to domain AI. Not because general-purpose models are bad, but because education has specific requirements — safety, privacy, pedagogical alignment, teacher oversight, curriculum standards, and age-appropriate behavior — that general tools were never designed to meet.

This does not mean general AI disappears from schools. It means the market is splitting. General AI will remain widely used by individual teachers and students. But the institutional purchasing decisions — what districts procure, what gets integrated into learning management systems, what meets compliance requirements — will increasingly favor specialized tools built for education.

Why General AI Spread First in Education

General-purpose AI tools dominated the first wave of classroom adoption for a simple reason: they were available first.

When large language models became publicly accessible, they were immediately useful for education-adjacent tasks. Students could get help with homework, explore topics conversationally, generate study materials, and practice writing. Teachers could draft lesson plans, create rubrics, differentiate assignments, and generate quiz questions. The versatility of general AI made it useful across every subject and grade level.

The adoption was bottom-up — and staggering in its speed. Nearly two in three K-12 teachers (63 percent, up 12 percent year-over-year) now say they have incorporated generative AI into their teaching process. High school teachers are the most active adopters at 69 percent, compared with 42 percent of elementary teachers and 33 percent of pre-K teachers. Three in 10 teachers use AI weekly, saving an estimated six weeks per year on tasks like lesson planning, grading, and drafting parent communications.

Students adopted even faster. Sixty-three percent of US teens use AI tools like ChatGPT for schoolwork. But there is a confidence gap: only 30 percent of teachers report feeling confident using the same AI tools their students use daily.

The problem was that these tools were not designed for education. They had no concept of grade level. They could not align outputs to curriculum standards. They had no teacher dashboard. They had no age-appropriate content filters beyond basic safety guardrails. They could not distinguish between a student seeking understanding and a student seeking to cheat. They treated a 10-year-old and a graduate student identically.

General AI proved the demand. Domain AI is now meeting the requirements.

Where Generic Tools Fall Short in Schools

The gap between what general AI can do and what education requires is structural, not cosmetic.

Safety and age-appropriateness. General AI models are trained on the open internet, which includes content that is inappropriate for children. While safety filters exist, they are designed for a general audience, not for the specific context of a K-12 classroom. A tool that is safe enough for an adult user may still surface content that is inappropriate for a 12-year-old — not because it is generating harmful content intentionally, but because it lacks the context to modulate its responses based on the age, maturity, and educational setting of the user.

Pedagogical alignment. Teaching is not just providing correct answers. Good pedagogy involves scaffolding understanding, asking probing questions, encouraging productive struggle, and knowing when to provide information and when to withhold it so the student discovers the answer. General AI tools are optimized for helpfulness — they give the best answer as quickly as possible. This is the opposite of what a good tutor does. A domain-specific education AI can be designed to guide rather than tell, to ask "what do you think?" before providing the answer, and to adjust its scaffolding based on the student's demonstrated understanding.

Curriculum standards. Education systems operate on structured curriculum frameworks — Common Core, state standards, national curricula. Teachers need AI that aligns its content and assessments to these standards. General AI has no concept of scope and sequence, grade-level expectations, or the specific learning objectives that a lesson is designed to address.

Teacher oversight. Schools need AI that gives teachers visibility into how students are using it — what questions they are asking, where they are struggling, and whether they are engaging productively or simply copying outputs. General AI tools provide no teacher dashboard, no usage analytics, and no way for an educator to monitor or guide student interactions.

Privacy and compliance. Schools operate under strict privacy regulations — FERPA in the United States, GDPR in Europe, and various state-level laws governing student data. General AI tools designed for consumers may not meet these requirements. Student data — including the questions students ask, the topics they explore, and the errors they make — is sensitive educational information that requires specific protections.

These are not nice-to-have features. They are requirements that determine whether a school district can legally and ethically deploy an AI tool at scale.

What Makes Education AI "Domain-Specific"

Domain-specific education AI is not just general AI with an education prompt. It is AI that has been architecturally designed for the educational context.

Guardrails and content policy. Gemini for Education operates as a "school official" under FERPA, meaning student data is contractually protected and never used for model training. Google consulted with child safety and development experts to shape content policies, partnered with learning science experts, tested with youth advisory panels, and added extra data protection for all education users. The platform is both FERPA and COPPA compliant, with parental controls available through Google Family Link for children under 13.

Pedagogical design. The interaction model is different. Instead of optimizing for helpfulness, education AI optimizes for learning. This means the AI asks questions before giving answers, provides hints rather than solutions, explains concepts at the appropriate level, and adapts its approach based on the student's responses. The Socratic method — guiding through questions — is fundamentally different from the information-retrieval model of general AI.

Curriculum integration. Domain-specific tools align with specific curriculum standards and can generate content, assessments, and activities that map to learning objectives. Gemini in Classroom lets teachers generate vocabulary lists with definitions and example sentences, create quizzes and rubrics, and build differentiated assignments tailored to actual class context. Teachers can share custom AI experts called "Gems" with their colleagues. A teacher can specify "I need a formative assessment on fractions for fourth graders aligned to Common Core standard 4.NF.A.1" and get a pedagogically appropriate result.

Teacher controls and dashboards. Education AI provides teachers with visibility into student usage. Gemini for Education includes class tools for managed Chromebooks that give teachers real-time control over student devices — switching between explore, focus, and locked modes. Educators will soon see actionable insights providing opportunities for a transparent, reflective approach to AI learning experiences. This turns the AI from a black box into a teaching tool that the educator can monitor and adjust.

Privacy by design. Education AI tools are built with FERPA, COPPA, and other regulations as architectural requirements, not afterthoughts. Student data is handled according to specific governance frameworks. Conversations are not used for model training. Data retention policies align with institutional requirements.

The Evidence Gap: What We Know and Do Not Know About AI Tutoring

The promise of AI tutoring is immense — but the evidence base is thinner than the marketing suggests, and the distinction between types of AI tools matters enormously.

Khan Academy's broader platform — its structured exercises, videos, and mastery-based learning system — has demonstrated measurable learning gains. Internal studies show that students using the Khan Academy platform for 30 or more minutes per week achieve approximately 20 percent greater learning gains than peers who do not. But it is important to distinguish between the platform's established adaptive learning system and Khanmigo, its newer AI chatbot tutor. As of early 2026, Khanmigo specifically lacks published randomized controlled trial (RCT) evidence. One independent study found no significant difference in learning outcomes between students using Khanmigo and students using Google search for homework help.

This does not mean AI tutoring is ineffective. It means the evidence is still emerging and the design of the tutoring interaction matters enormously. Adaptive learning platforms that have been refined over years — like Khan Academy's core system, DreamBox, IXL, and ALEKS — have stronger evidence bases than newer chatbot-style tutors. The key variable is not the sophistication of the underlying AI model but the quality of the pedagogical design: how the system diagnoses misconceptions, sequences content, provides feedback, and scaffolds understanding.

Schools evaluating AI tutoring tools should ask for evidence that is specific to the tool being purchased, not extrapolated from general AI capabilities or from a different product by the same company. A platform's track record with structured adaptive learning does not automatically validate its chatbot tutor, and vice versa.

Safety, Privacy, and Teacher Controls

The safety and privacy requirements of education AI are not optional — they are the gatekeepers of institutional adoption.

FERPA (Family Educational Rights and Privacy Act) protects student education records and limits how schools can share student data. Any AI tool that processes student interactions is handling educational records. FERPA requires that schools maintain control over this data and that vendors handle it according to specific agreements.

COPPA (Children's Online Privacy Protection Act) applies to children under 13 and imposes strict requirements on the collection, use, and disclosure of personal information. In K-12 education, this means that AI tools used by elementary students must meet COPPA requirements — including parental consent mechanisms and data minimization principles.

State-level regulations add additional complexity. During the 2026 legislative session, 49 bills across 23 states address artificial intelligence in classroom instruction. Ohio passed a law requiring all K-12 public schools to adopt AI policies by mid-2026. South Carolina's H.B. 5253 would establish some of the strongest statewide guardrails, requiring written parental opt-in consent, annual public disclosure of AI tools, and prohibiting AI from replacing licensed teachers in core instruction. California, Connecticut, and Texas introduced bills creating oversight boards and regulatory sandboxes for AI in education.

Teacher controls are the practical interface through which these requirements are met. A well-designed education AI tool gives teachers the ability to set boundaries on what the AI can discuss, monitor student interactions in real time or through reports, adjust the level of scaffolding the AI provides, and ensure that the AI's behavior aligns with the classroom's learning goals.

General AI tools have none of these capabilities. This is not a criticism of general AI — these tools were designed for a different purpose. But it explains why institutional procurement is moving toward specialized alternatives.

Personalized Learning Without Losing Pedagogy

The promise of AI in education has always been personalization — meeting each student where they are and adapting instruction to their individual needs. This promise is real, but it comes with a pedagogical tension that domain-specific AI must navigate.

Personalization without pedagogy is just customized content delivery. A system that adjusts difficulty based on student performance is useful, but it is not teaching. Teaching involves understanding why a student is struggling — whether the issue is conceptual, procedural, motivational, or related to prior knowledge gaps — and adapting the approach accordingly.

Domain-specific education AI attempts to bridge this gap. Adaptive learning platforms like DreamBox, IXL, and ALEKS use AI to diagnose student misconceptions, not just measure performance. Tutoring systems built on pedagogical models can adjust not just what content they present, but how they present it — using different representations, analogies, or scaffolding strategies based on the specific nature of the student's difficulty.

The evidence on AI tutoring effectiveness is growing but uneven. Fifty-nine percent of teachers say AI has enabled more personalized instruction. Structured adaptive learning platforms with years of refinement show the strongest results. Newer chatbot-style AI tutors show promise but lack the rigorous trial evidence that schools should demand before large-scale deployment. The key variable is not the sophistication of the AI model — it is the quality of the pedagogical design that shapes how the AI interacts with learners.

Why Procurement Favors Specialized Systems

The shift from general to domain AI in education is driven not just by pedagogy but by procurement.

School districts do not buy AI the way individuals do. They go through formal procurement processes that evaluate products against specific criteria: compliance with privacy regulations, alignment with curriculum standards, evidence of effectiveness, accessibility requirements, integration with existing learning management systems, and total cost of ownership.

General AI tools struggle to meet these criteria because they were not designed for them. A district evaluating AI tools will ask: Does this tool meet FERPA requirements? Is it COPPA-compliant? Does it provide teacher dashboards? Can it align outputs to our state standards? Has it been validated in educational settings? Does it integrate with our LMS?

Domain-specific education AI tools are built to answer "yes" to every one of these questions. They are designed from the ground up to pass through the procurement filter that general AI cannot navigate.

This creates a market dynamic where general AI tools remain widely used by individual teachers and students — adopted bottom-up through personal choice — while domain-specific tools win the institutional purchasing decisions that drive scale. The budget goes to the specialized tool. The informal usage stays with the general tool. Both coexist, but the institutional investment flows toward specialization.

The Likely Split Between General and Domain AI

The future of AI in education is not general versus domain. It is general and domain — with each serving different roles and different users.

General AI will continue to be used by teachers for lesson planning, content creation, and professional development. It will continue to be used by older students for research, writing support, and exploration. Individual adoption will remain high — 83 percent of K-12 teachers already use generative AI tools for either personal or school-related purposes — and will continue to outpace institutional adoption.

Domain AI will be what schools buy, integrate, and govern. It will be embedded in learning management systems, curriculum platforms, and assessment tools. It will be the AI that meets compliance requirements, provides teacher oversight, and delivers personalized learning at institutional scale.

The coexistence model is already visible. Schools use Google Workspace or Microsoft 365 for general productivity — and they use specialized platforms like Khanmigo, Duolingo, or district-approved adaptive learning systems for structured instruction. The same split will apply to AI: general tools for general use, specialized tools for institutional deployment.

What Schools Should Actually Evaluate Before Buying

For schools and districts navigating the education AI market, the evaluation criteria should be specific and pragmatic.

Compliance first. Does the tool meet FERPA requirements? Is it COPPA-compliant for the age groups that will use it? Does it align with your state's data privacy laws? If the answer to any of these is unclear, the tool is not ready for institutional deployment.

Pedagogical design. Is the AI designed to teach, or just to answer? Does it scaffold understanding rather than providing answers? Has it been developed with input from educators and learning scientists? Is the interaction model appropriate for the intended age group?

Teacher controls. Can teachers monitor student usage? Can they set boundaries on what the AI can discuss? Do they get actionable insights about student learning? Is the AI a tool that the teacher controls, or a black box that the teacher cannot see into?

Curriculum alignment. Can the tool align its outputs to your curriculum standards? Can it generate assessments, activities, and content that map to specific learning objectives?

Evidence of effectiveness. Has the tool been studied in educational settings? Is there evidence — from randomized controlled trials, not just case studies or engagement metrics — that it improves learning outcomes? Are the studies independent or vendor-funded? Be wary of vendors who cite evidence for a different product or a general platform when selling a specific AI feature.

Integration. Does it work with your existing LMS, SIS, and classroom tools? Or does it require a separate login, a separate workflow, and a separate training program?

The schools that ask these questions — and hold vendors accountable for the answers — will build AI programs that serve students and teachers. The schools that adopt AI based on hype, demos, or vendor promises will find themselves managing tools that do not fit the educational context.

---

At AIReady.fit↗, we help professionals and teams build productive AI workflows. Our AI Foundations track covers how AI is reshaping education and other professional domains — practical skills for anyone adapting to the next generation of workplace AI tools.