EdTech: what is getting real adoption now?

What actually counts as real EdTech adoption today?

Real EdTech adoption starts when students, teachers or institutions keep using a product because it has become useful enough to stay in the workflow.

That gives us a much tougher test than account creation. A student opening ChatGPT twice for homework shows usage. A university providing ChatGPT Edu to an entire campus shows institutional commitment. A teacher returning every week because an AI tool saves an hour of preparation tells us much more. When the software is also connected to the LMS, curriculum, assessment or school administration, adoption has become difficult to reverse.

EdTech regularly produces enormous access numbers. Google says Gemini for Education has been integrated across more than 1,000 US higher-education institutions covering over 10 million students. OpenAI says hundreds of universities now work with ChatGPT Edu. Both figures are substantial. Neither tells us how many students use those products every week or how much learning behavior has changed.

For this article, we therefore put the most weight on repeated use, integration, paid or institutionally supported deployment, renewals and measurable outcomes. Those tests make it much easier to separate genuine adoption from a large pilot.

What we see	What it tells us
Trial or optional account	Interest exists
Institution-wide access	The buyer has committed
Repeated student or teacher use	Behavior is changing
LMS, curriculum or admin integration	The product is becoming embedded
Renewals plus measurable results	Adoption looks durable

Is AI actually taking over EdTech now?

AI is taking over the conversation around EdTech, while the software schools depend on every day remains much broader.

The latest Instructure data makes the gap unusually visible. Its 2026 EdTech Top 40 study looked at actual LTI launches through Canvas across 12.7 million US K–12 users, including more than 11 million students and 618,000 educators. Instead of measuring websites that schools could theoretically access, Instructure measured tools launched from inside the LMS.

The average district had 3,001 digital tools somewhere in its environment. Individual students and educators interacted with only four LTI tools on average during the measured school year.

That is a useful reality check on the AI boom. Schools may test hundreds of products, but everyday learning still concentrates around a tiny stack.

Canvas itself has tens of millions of users, more than 1,000 integrations and has handled six million concurrent users at peak. IXL says more than 18 million students use its personalized learning platform. Renaissance says more than 40% of US schools rely on its products.

AI is now being added to this installed base at high speed. The installed base itself remains the deeper adoption story.

If you want more recent data on this point, please see our latest EdTech market report.

Are teachers really using AI now?

Teacher use of generative AI has already moved well beyond the early-adopter stage, especially for lesson preparation and repetitive work.

A UK Department for Education survey published earlier this year found that 82% of primary teachers and 78% of secondary teachers had used generative AI in their role. Microsoft’s latest international education research found that 88% of educators surveyed had used AI for school-related purposes, with 76% saying their use had increased over the previous year.

The interesting part is what teachers are doing with it. In the Royal Society of Chemistry’s latest Science Teaching Survey, 79% of AI-using teachers said they created resources with AI, 66% wrote questions with it, 40% used it for administration and 35% used it for reports.

Those jobs have something in common: the teacher remains in control and can quickly inspect the result. Generating ten quiz questions is low risk because a teacher can reject the bad ones in seconds.

Recent institutional deployments point in the same direction. OpenAI initially brought ChatGPT for Teachers to nearly 150,000 US teachers and staff. Its latest expansion adds more than 100,000 educators and staff across 55 additional school systems in 20 states.

So the clearest new AI adoption in K–12 currently sits on the teacher side. Saving teachers time is proving easier to sell and easier to use repeatedly than asking AI to take over the teaching itself.

Are students really using AI for school this much?

Student AI use is already mainstream, and much of it happened before schools had time to decide how they wanted AI used.

Microsoft’s latest global education research found that 92% of surveyed students had already used AI for school-related purposes. Around a third said they used AI to learn or study in ways that worked better for them.

Faculty see the same behavior from the other side. College Board research involving more than 3,000 faculty members found that 74% believed students were using AI to write essays or papers, while 67% reported student use for paraphrasing and rewriting.

The striking part is how far student behavior has moved ahead of formal teaching. In Microsoft’s survey, 77% of students said they had received no formal AI training.

That creates an unusual EdTech adoption path. Students can adopt ChatGPT, Gemini or Copilot immediately from a personal account. Schools then have to catch up by deciding what to allow, what to teach and which institutional version they want to provide.

Consumer AI effectively skipped the traditional school procurement cycle.

Is all this student AI use actually improving learning?

Current evidence shows that AI can improve learning, but simply giving students access produces much less convincing results.

A randomized controlled trial published in Scientific Reports compared a purpose-built generative-AI tutor with an active-learning university physics class. Students using the AI tutor learned more in less time and also reported higher engagement and motivation.

Pearson has produced useful evidence at much larger scale. Its latest learning research analyzed more than 62,000 higher-education students using Study Prep. Students receiving AI-powered adaptive practice were 90% more likely to reach initial topic mastery than students receiving static practice questions while spending roughly the same amount of study time.

Another Pearson analysis covered close to 80 million interactions from roughly 400,000 students. Students using its AI study tools inside digital courseware were three times more likely to show active-reading behaviors such as testing themselves, revisiting material and taking notes.

Randomized experiments deserve more weight than vendor observational studies, but the pattern across them is useful. AI performs much better when the software keeps the student working through a learning process.

Free-form access gives students another option: ask for the finished answer.

That difference is likely to become one of the central dividing lines in EdTech. The interesting question now is less whether an AI model can explain algebra and more whether the product can keep a student thinking long enough to learn it.

If you want more recent data on this point, please see our latest EdTech market report.

Are AI tutors actually catching on?

AI tutors are reaching millions of students, but sustained tutoring behavior remains much thinner than the headline user numbers suggest.

Khan Academy says Khanmigo reached about two million students, educators and parents during one school year, an increase of more than 700% from the previous year. Around 770,000 US students also had Khanmigo available through Khan Academy district partnerships.

Pearson gives us another route to scale. More than three million higher-education students have access to its AI-powered study features inside existing Pearson products.

Then came a particularly useful reality check. A new NBER study followed Khanmigo for two years across 18 Tennessee middle schools. Assignment to Khanmigo increased math achievement by roughly 0.06 to 0.08 standard deviations over a school year, while a full year of active participation implied an effect of around 0.14 standard deviations.

The catch was usage. Although 96% of assigned students tried Khanmigo at least once, the median student messaged it on only around one-third of the days when they practiced. After making a mistake, students used the tutor in only 17% of those exercise sessions.

The researchers also found that the gains were similar to those produced by Khan Academy practice without the AI tutor.

AI tutoring has clearly reached real deployment. The next hurdle is getting students to use the tutoring part often enough for it to create an advantage over good adaptive practice.

Current example	Scale	What we actually learn
Khanmigo	About 2M users in one school year	Large real-world reach
Khanmigo district access	About 770,000 US students	Schools are deploying it
Pearson AI study tools	3M+ students with access	AI can ride existing courseware distribution
Tennessee Khanmigo trial	96% tried it, median use on about one-third of practice days	Access is much deeper than habitual tutoring use

Has personalized learning already won before generative AI?

Personalized digital practice is already one of EdTech’s most established use cases, and generative AI is now improving a model that schools have used for years.

IXL currently serves more than 18 million students and says learners have answered more than 200 billion questions on the platform. Renaissance says more than 40% of US schools rely on its learning and assessment products.

These products had personalization long before the current AI wave. Diagnostics identified gaps, algorithms adjusted difficulty, assessment data grouped students and software recommended what to practise next.

Generative AI expands what that loop can do. A system can now explain a mistake in several ways, generate another example, adapt language to the learner, summarize a student’s progress for a teacher or turn assessment data into a suggested intervention.

That makes personalized learning one of the easier places for AI to get lasting adoption. Schools already understand the workflow, students already use the software and the new AI layer can improve something that already has distribution.

We therefore see more evidence here than in visions of a completely autonomous AI school. Existing adaptive-learning products have spent years solving the harder institutional problem of getting inside classrooms.

Are schools actually cutting down their huge EdTech stacks?

Schools are increasingly concentrating usage around fewer tools, which gives LMS platforms and deeply integrated products a bigger advantage today.

Instructure’s current usage data captures the problem neatly. Districts in its sample had access to roughly 3,001 digital tools on average, yet each student and educator launched only four LTI tools on average.

Those numbers measure different things, so treating four divided by 3,001 as a utilization rate would be misleading. The useful finding is simpler: institutional software portfolios can become enormous while classroom attention stays highly concentrated.

Every extra vendor also brings work. Schools need to handle login systems, student data, privacy reviews, security, training, procurement and integration. Generative AI adds another question around what information can be sent to an external model.

Canvas benefits directly from this pressure. Instructure currently reports more than 8,000 customers across over 100 countries, more than 1,000 Canvas integrations and tens of millions of users.

The same dynamic helps Microsoft, Google and established courseware providers. Adding AI inside software that a school already approved is much easier than persuading the institution to create another standalone workflow.

For new EdTech companies, distribution through the existing stack is becoming a serious advantage in its own right.

If you want more recent data on this point, please see our latest EdTech market report.

Can consumer EdTech still get people to pay?

Consumer EdTech can still produce massive paid adoption when the product creates a habit people want to keep, and Duolingo remains the strongest example.

In its latest quarterly results, Duolingo reported 58.7 million daily active users, up 23% from 47.7 million a year earlier. Paid subscribers reached 12.7 million, compared with 10.9 million a year earlier, while quarterly revenue increased 18% to $298.5 million.

The scale is unusual even outside education. Nearly 59 million people choosing to open a learning product each day tells us much more about adoption than a large pool of registered accounts.

AI is increasingly part of Duolingo’s product, particularly in premium features, but its adoption engine was built around something older: short lessons, streaks, progression, notifications and enough gamification to make returning feel natural.

Consumer education products still have plenty of room to grow when they own the learning habit itself.

Is ChatGPT killing the old homework-help business?

General-purpose AI has already broken a large part of the old paid homework-answer model, and Chegg gives us the clearest public evidence.

Chegg’s latest quarterly revenue fell 51% year over year to $51.8 million. Its skilling revenue moved in the opposite direction, rising 2% to $17.5 million.

That split tells us quite a lot. Chegg became huge partly because students were willing to pay for fast access to explanations, worked solutions and answers. ChatGPT, Gemini and other general-purpose models made that service abundant and often free.

Compare that with Duolingo. Duolingo sells an ongoing learning experience that depends on progression and habit. Chegg’s historical homework proposition was much easier for a general AI assistant to reproduce.

As we saw above, student AI usage is already extremely high. Once the substitute sits one tab away and can answer almost any subject, a standalone homework-answer subscription becomes a much harder sell.

The sharpest disruption in EdTech so far has happened where AI can replace the old product with almost no change in student behavior.

If you want more recent data on this point, please see our latest EdTech market report.

Are universities really rolling out AI across entire campuses?

Campus-wide AI has become a real institutional market, although universities still reveal far more about access than about how deeply students use it.

Google says Gemini for Education now reaches more than 10 million college students across over 1,000 US higher-education institutions. The participating institutions include large public systems as well as universities such as MIT, Brown, Notre Dame and the University of Michigan.

OpenAI says hundreds of universities work with ChatGPT Edu. Campus-wide deployments include Arizona State University, the California State University system, Oxford, Indiana University, USC and others.

Universities have good reasons to bring AI inside an approved institutional environment. They can add single sign-on, administrative controls, contractual data protections and clearer rules around which tools students and faculty should use.

The size of these deployments means institutional adoption is already real. What remains surprisingly opaque is the next layer down. Vendors rarely disclose what percentage of provisioned students become weekly active users, how long sessions last or how often AI becomes part of coursework.

Universities are clearly buying into campus AI now. We still have much better evidence for distribution than for deep academic dependence.

Is AI creating a boom in workforce learning too?

AI is keeping workforce learning highly relevant, but current enterprise numbers look more like a mature market under pressure than a fresh spending boom.

Coursera’s combination with Udemy gives us a useful current snapshot. Under the unified reporting methodology used after the transaction, the platform reported 12,107 enterprise customers in its latest quarter, down 2% from the comparable figure a year earlier. Enterprise net retention fell from 95% to 91%.

Consumer paid subscribers moved much faster, reaching 1.655 million, up 44%.

Companies absolutely need workers to learn new AI tools, but that does not automatically translate into rapid growth for every corporate-learning vendor. Enterprises can consolidate platforms, negotiate harder and obtain similar training content from many providers.

The current winners therefore need more than an enormous course library. They need content tied closely to changing job requirements, employer workflows, assessment or credentials that companies actually value.

AI is creating plenty of learning demand. Capturing that demand as durable enterprise revenue is proving more competitive.

Can education-specific AI survive when ChatGPT and Gemini are everywhere?

Education-specific AI can still win, but the product needs to own something around the model that schools and students cannot get from a blank chatbot.

Khanmigo starts with Khan Academy’s curriculum, exercises and student progress. Pearson puts AI beside assigned textbooks, homework and courseware. MagicSchool builds around teacher workflows and district controls. Canvas can connect AI to courses, assignments and the rest of a school’s software stack.

Those advantages become more valuable as general-purpose models improve. Building another interface where a student can type a question into an LLM gives the product very little protection.

The harder assets sit around the model: curriculum context, assessment data, teacher controls, school approvals, student history, integrations and a learning flow designed around a particular subject.

General AI may actually strengthen the best specialized EdTech companies by making intelligence cheaper to add. At the same time, it puts enormous pressure on products whose entire value came from access to an answer generator.

That is why the current adoption pattern favors AI embedded inside an existing education product more than standalone “ChatGPT for schools” clones.

If you want more recent data on this point, please see our latest EdTech market report.

What part of AI education still looks mostly ahead of the market?

Fully autonomous AI teaching remains much earlier than teacher-assistance, structured practice and institution-managed AI access.

We have real evidence that well-designed AI tutors can help students. We also now have large-scale evidence showing how difficult it is to turn access into sustained tutoring behavior.

The Tennessee Khanmigo experiment is especially revealing here. Students could call on an AI tutor inside an existing math workflow, 96% tried it, and yet substantive use remained limited. That happened in a structured school environment where the tool had already cleared many of the barriers that a new consumer tutoring app would face.

Schools also face a higher risk threshold once AI starts interacting directly with children, evaluating work or making consequential recommendations. Privacy, accuracy and dependency become much more important than when a teacher uses AI privately to draft a worksheet.

For now, schools appear much more comfortable using AI around the teacher and around structured learning activities.

The vision of every student spending hours each week with an autonomous personal AI tutor is technically plausible. Current behavior still falls well short of that vision.

So what EdTech is getting real adoption now?

The EdTech products getting the deepest adoption today are established learning infrastructure, adaptive practice and assessment, while teacher AI tools are the fastest genuinely new category to break into everyday use.

Canvas, IXL, Renaissance and established digital courseware already sit inside recurring education workflows at huge scale. That layer receives less attention than generative AI, yet it remains the part schools depend on most heavily.

Teacher AI has moved much faster than the rest of the new market. UK government research puts generative-AI use around four-fifths of teachers, Microsoft finds similarly widespread adoption internationally, and OpenAI is now adding another 100,000-plus educators and staff to its district deployments. The jobs teachers choose first are practical ones: resources, questions, planning and administration.

Student AI is even more widespread, although much of that usage happens through general-purpose products outside traditional EdTech procurement. Universities are responding by rolling out Gemini and ChatGPT Edu across entire institutions. Ten million-plus students covered by Gemini for Education and hundreds of ChatGPT Edu universities make campus AI a real market already.

AI tutoring sits in a more complicated position. Khanmigo, Pearson and several controlled studies show that AI can improve learning and can reach millions of students. The latest large school experiment also shows how easily access outruns real engagement. Getting students to use an AI tutor deeply and repeatedly now looks harder than building the tutor.

Consumer EdTech gives us the clearest split of all. Duolingo’s nearly 59 million daily users and 12.7 million paid subscribers show that people will still pay for digital learning when the product creates a habit. Chegg’s 51% revenue decline shows how quickly general AI can destroy an education product when the old value proposition can be reproduced in a chat window.

Put all of that together and the current market looks much less mysterious. Real adoption is clustering around products that already own a workflow and can make that workflow better. AI works especially well when it saves teachers time, improves structured practice or appears inside software that schools already trust.

The weaker part of the market is the grander promise: autonomous AI teachers, generic education chatbots and products whose entire pitch depends on having access to an intelligent model.

As of now, the biggest EdTech shift is happening inside existing teaching and learning habits. The products gaining real adoption are the ones making those habits easier, faster or more personalized without asking schools to rebuild education from scratch.

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.