AI Use by Teachers in India: What the Data Actually Says

The three numbers that matter

CENTA, the Centre for Teacher Accreditation, surveyed more than 5,000 Indian educators in 2025 as part of the NEP@5 series. Three findings describe the situation more precisely than any anecdote.

Over 70 percent are already using AI in their classrooms. Around 60 percent use it for lesson planning, and 26 percent for activity ideas.

Sixty seven percent rate their own AI skills as above average.

And 57 percent could correctly identify a basic AI misconception.

Two of those numbers measure confidence. One measures capability. The distance between them is the entire story of AI in Indian schools right now.

Adoption already happened, without a policy

Most conversations in school leadership still treat AI adoption as a decision to be made. It is not. It was made some time ago, quietly, one teacher at a time, in the gap between a Sunday evening and a Monday lesson.

The same survey found adoption rising to around 75 percent among teachers with more than three years of experience. This is not a story about early career staff experimenting with new software. The teachers using AI most are the ones who have been in classrooms longest.

That matters because it changes what a school leader should be asking. Not whether teachers should use AI, but whether anyone would be able to tell how well they are using it.

There is a cheap way to find out. Ask three teachers what they used AI for most recently, and what they did with the output before it reached a child. Those two questions produce more useful information than any staff survey.

What teachers actually use AI for

The CENTA data tells us which tasks: lesson planning leads at around 60 percent, activity ideas at 26 percent.

What it does not tell us is how those tasks are approached, and that turns out to be the more important question.

A systematic review published in Humanities and Social Sciences Communications in 2026, covering 28 empirical studies of teacher and AI collaboration in learning task design, found that AI is "most often serving as an assistant or content generator, and less frequently as a co-designer or dialogic partner."

In plain terms: teachers mostly ask AI to make things. Far fewer ask AI to think with them.

That distinction is where most of the available value is being left behind.

The difference one question makes

Consider a teacher preparing a Grade 4 worksheet on comparing fractions. She asks AI to make one, and receives twelve competent questions. Nothing is wrong with any of them.

A second teacher asks something first: what do Grade 4 students typically get wrong when comparing fractions?

The answer changes her lesson. Children routinely believe one eighth is larger than one quarter, because eight is a larger number than four. This is among the best documented misconceptions in primary mathematics.

She no longer wants twelve questions. She wants three that catch precisely that error.

Same tool. Same ten minutes. One teacher produced a worksheet. The other produced a diagnostic, and learned something about her own class in the process.

That single question, asked before anything is generated, is the step called Mine in the IMPACTS framework, the method I teach through Credo Learnings. It is the smallest change a teacher can make and the one that most reliably improves what comes back.

Why capability is not the same as confidence

The gap between 67 percent confidence and 57 percent capability is not a criticism of teachers. It reflects how AI training has been delivered.

Almost all school AI training teaches people to operate the tool, and stops there. Very little teaches them to question what it produces. That distinction, between functional and critical AI literacy, is drawn from an integrative review of 124 studies presented at ACM ICER in 2025, and it explains most of what goes wrong. I have written about why AI training in schools fails separately, because it deserves its own treatment.

The short version is that operational skill alone changes very little. A teacher who can run ChatGPT but has never been shown how to interrogate it is faster at producing something that looks finished. That is not the same as producing something better.

What this means for the CBSE mandate

CBSE Circular Acad-15/2026, issued on 1 April 2026, makes computational thinking and AI a mandatory curricular area for Classes 3 to 8 from the 2026-27 session.

The curriculum asks for structured thinking, problem solving and ethical awareness around AI. Those are dispositions, not topics, and a teacher can only build in children what she has built in herself.

Which places staff capability ahead of vendor selection in the sequence, not after it. The implementation detail, hours, class bands and what schools should do first, is covered in the CBSE CT and AI curriculum guide.

The encouraging part of the data

There is a reading of these numbers that is often missed.

The gap is not a knowledge gap. It is a transfer gap.

An experienced teacher already defines outcomes before planning. She already anticipates what children will misunderstand. She already evaluates her own materials against a standard, and revises them. Those are precisely the judgments that AI use requires.

Nobody told her they applied here.

That is a far easier problem to solve than building professional judgment from nothing, and it is why AI training that starts from existing teacher expertise produces different results from training that starts from the software.