As AI permeates classrooms around the world, Google realized that the toughest lessons about how the technology could actually scale were coming not from Silicon Valley but from schools in India.
India has become a testing ground for Google’s education AI as competition intensifies from rivals such as OpenAI and Microsoft. The country, with more than 1 billion internet users and an education system shaped by state-level curriculum, strong government involvement, and unequal access to devices and connectivity, currently has the highest rate of Gemini learning in the world, according to Chris Phillips, vice president and general manager of education at Google.
Phillips spoke on the sidelines of Google’s AI for Learning forum in New Delhi this week, where he met with industry stakeholders, including K-12 school administrators and educators, to gather feedback on how AI tools are being used in the classroom.
The size of India’s education system helps explain why the country has become such a significant testing ground. According to the Government of India’s Economic Survey 2025-26, the country’s school education system serves approximately 247 million students in approximately 1.47 million schools, supported by 10.1 million teachers. Its higher education system is also one of the largest in the world, enrolling more than 43 million students in 2021-22, an increase of 26.5% from 2014-2015, complicating efforts to deploy AI tools across a vast, decentralized and unevenly resourced system.
One of the most obvious lessons for Google is that AI in education cannot be deployed as a single, centrally defined product. While curriculum decisions in India are made at the state level, with ministries playing an active role, Phillips said Google needs to design its educational AI so that schools and administrators, not companies, can decide where and how to use it. This marks a shift for Google, which, like most Silicon Valley companies, has traditionally built products that scale globally rather than following the preferences of individual institutions.
“We’re not offering a one-size-fits-all offering,” Phillips told TechCrunch. “There are very diverse environments around the world.”
Beyond governance, that diversity is also reshaping how Google thinks about AI-driven learning itself. Phillips said the company is accelerating the adoption of multimodal learning in India that combines video, audio and images with text, reflecting the need to reach students across different languages, learning styles and access levels, especially in classrooms not built around text-centric instruction.
Maintaining teacher-student relationships
A related change is Google’s decision to design its educational AI around teachers rather than students as its primary point of control. Rather than bypassing AI experiences that bring educators directly to students, the company is focused on tools that help with planning, assessment, and classroom management, Phillips said.
“The teacher-student relationship is very important,” he said. “We’re here to help it grow and thrive, not replace it.”
In parts of India, AI in education is being brought to classrooms that previously lacked per-student devices and reliable internet access. Google is encountering schools where devices are shared, connections are spotty, and learning moves directly from pen and paper to AI tools, Phillips said.
“Access is universally important, but when and how it happens varies widely,” he added, pointing to environments where schools rely on shared or teacher-directed devices rather than one-on-one access.
Meanwhile, Google is rolling out early learning in India, including AI-powered JEE Main preparation through Gemini, a nationwide teacher training program for 40,000 Kendriya Vidyalaya educators, and partnerships with government agencies for vocational and higher education, including India’s first AI-enabled state university.
For Google, India’s experience serves as a harbinger of challenges that may surface in other regions as AI penetrates deeper into the public education system. The company expects that issues around control, access and localization (now evident in India) will increasingly impact the global expansion of AI in education.
From entertainment to learning as top AI use cases
Google’s efforts also reflect broader changes in how people use GenAI. Phillips said entertainment dominated AI use cases last year, adding that learning is now emerging as one of the most popular ways people engage with technology, especially among younger users. Education has become more immediate and important to Google as students increasingly rely on AI to learn, prepare for exams, and improve their skills.
India’s complex education system has also attracted attention from Google’s rivals. OpenAI has begun building regional leadership focused on education, hiring former Coursera APAC managing director Raghav Gupta as head of education for India and APAC and launching its Learning Accelerator program last year. Meanwhile, Microsoft is expanding its partnerships with Indian institutions, government agencies, and edtech companies, including Physics Wallah, to support AI-based learning and teacher training, highlighting how education has become a key battleground as AI companies seek to incorporate their tools into public systems.
At the same time, India’s latest economic survey points to risks to students from the use of non-trivial AI, including over-reliance on automated tools and potential impact on learning outcomes. Citing research from MIT and Microsoft, the study found that “reliance on AI for creative work and writing contributes to cognitive atrophy and reduced critical thinking abilities.” It’s a reminder that AI is in a race to enter the classroom amid growing concerns about how it will shape learning itself.
Whether Google’s India Handbook will serve as a model for AI in education in other regions remains an open question. But as GenAI moves deeper into the public education system, the pressures now being seen in India are likely to surface in other countries, making it difficult for the industry to ignore the lessons Google is learning there.
