AI Leap: Estonia's nationwide programme to rethink education in the age of AI

The problem

Artificial Intelligence had become widely used by students across Estonia, with between 64% and 90% of students drawing on commercial tools such as ChatGPT to help with their schoolwork. Much of that use was about arriving at an answer quickly, prompting concern that heavy reliance on tools not designed for education would hinder the development of critical thinking, focus, creativity, and self-management.

Part of the difficulty lay in the tools themselves. Commercial AI models are built for productivity, to produce a polished answer as fast as possible, which can work against students' learning. When a finished answer was just a prompt away, students were left taking the result rather than working through tasks that would build their underlying skills and knowledge.

There was a further concern beyond individual learning: that unguided use could widen the gaps between students. Those who already had strong learning skills were better positioned to use these tools to push ahead, while those who lacked them might lean on AI instead of their own thinking and fall further behind.

Given how widely students were already using AI, banning it was not a credible response, and ignoring it would have left that gap to widen unchecked. The realistic course was to shape how AI was used in learning rather than to resist it. The underlying challenge, students adopting commercial AI faster than schools can guide them, is not unique to Estonia.

The solution

AI Leap, launched in 2025, is Estonia's nationwide response. It takes its name and its approach from Tiger Leap, the 1990s programme that put computers and internet access into every Estonian school and helped set the country on its path to a digital state. Where Tiger Leap supplied the hardware, AI Leap supplies the tools and the know-how to use AI in a way that supports learning rather than shortcuts it.

The programme has three parts: a custom learning application for students, a training programme for teachers, and a research study to measure the effectiveness of these interventions.

At the centre of the programme is an application built in collaboration with OpenAI and researchers at the University of Tartu. It is based on ChatGPT but behaves very differently from the commercial version students had previously defaulted to.

When a student asks the regular ChatGPT to write an essay about Greek gods, it produces a polished essay. When a student asks the AI Leap application the same thing, it starts by asking questions: what is this essay for? Are you pressed for time? The student has to think about what they are doing and why before the tool helps them do it.

The design principle is to steer students toward reaching an answer themselves rather than handing it over. To hold the app to that, it is tested continuously against a set of 19 behavioural criteria developed up by Estonian educational psychologists, which check whether it is doing what a good tutor would: helping a student plan, prompting them to catch their own mistakes, asking them to explain their reasoning, reassuring them that confusion is part of learning, and pointing them toward ways to practise in a new setting. Where it falls short, it is adjusted.

The app replies only in Estonian, except in language lessons, and keeps its answers short so students are not overloaded. Two conditions were treated as non-negotiable for its use in schools: student conversations are not used to train the underlying model, and the app can recognise warning signs in a chat and steer a student toward Estonian support services.

Teachers were prepared before students received access
AI Leap reached teachers well before it reached students. From August 2025, every upper secondary teacher in Estonia was given free access to premium versions of ChatGPT and Gemini, alongside a structured programme of support. Schools set up Professional Learning Communities, in which teachers work together to determine how to use AI, compare what is working in their classrooms, and develop approaches suited to their own subjects.

By March 2026, 71% of teachers had activated the premium ChatGPT and Gemini accounts provided to them, up from 59% in December 2025. Across the year as a whole, the share of teachers using AI tools rose by 20 percentage points.

There was also a harder constraint behind the sequencing. Estonian schools did not yet have a legal basis to use AI applications in formal instruction, and the rules protecting minors' data made this a real barrier to giving students access. A legislative amendment in December 2025 closed that gap, and students gained access to the learning application from late January 2026, several months after their teachers.

Every upper secondary school in the country, from the start
From the outset, AI Leap covered the whole country. Its pilot year included all 154 of Estonia's upper secondary schools, roughly 20,000 students in grades 10 and 11, and about 4,900 teachers. Estonia opted for national coverage from day one, rather than a small set of volunteer schools, so that every student had the same access regardless of which school they attended.

How much and in what ways the app is used are left to individual teachers and students; the programme is not trying to increase usage for its own sake. As Visak has put it, the aim is to give students “a pedagogically guided alternative to widely used commercial AI models.”

The pilot year runs on a €4 million budget, divided equally between the Ministry of Education and Research and private-sector partners, among them Telia, Skaala and Smart Future Fond.

Measuring whether it works
Estonia built evaluation into the programme from the start, rather than treating it as something to do later. The University of Tartu, working with Stanford University and OpenAI, is running a large-scale randomised controlled trial to examine how, and under what conditions, the app affects students' learning, drawing on students' own chat logs alongside standardised testing and tracking changes in motivation, persistence, learning strategies, and self-regulation.

The impact

1. Teacher take-up has grown steadily
Premium AI accounts were made available to 99% of upper secondary schools. The share of teachers who had activated theirs rose from 59% in December 2025 to 71% by March 2026, and the number of weekly active teacher users more than doubled, from 1,456 to 3,523. By the end of 2025, nine in ten school leaders had taken part in the foundation's training.

2. Students are taking it up, with a positive early response
The learning app became available to all students in late January 2026. By March 2026, 7,700 had activated accounts, with just under half using it each week. Before that national release, the app had been tested with around 900 students between October and December 2025, and the reaction was encouraging: they liked being asked questions rather than being handed answers, valued that their conversations were not used to train the model, and some raised their own concerns about the environmental cost of AI.

3. Learning outcomes are not yet known
AI Leap is in its first year, so the evidence so far is about reach and take-up rather than results. Whether the app improves how students actually learn is the question the trial is built to answer, with findings due later this year.

Lessons learned

• Prepare teachers before reaching students. Two things had to come before students were given the app: teachers confident enough to guide its use, and a legal basis for using AI in formal instruction, which only arrived with the December 2025 amendment. For any government putting AI in front of
• The hardest problem is a motivational one. The programme is open about this. Visak calls it the “core pedagogical dilemma” and the “elephant in the room”: why a student would choose a tool that makes them think when another will simply generate the answer. The app is an attempt to change the terms of that choice. Visak describes it as a tool that “supports the growth mindset and planning skills,” and AI Leap's own account of the design is that prompting the student with questions before it does anything is meant to shift them from passively collecting a response to actively working through the task.
• The programme was built to keep improving and to share what it learns. Jaan Aru of the University of Tartu describes Estonian education as a system that the country “continuously improve[s],” and says the team is “excited to contribute methods that other education systems can reuse and build on.” That intent is built into the plan. AI Leap runs over three years, with later phases set to extend it to vocational education and to lower secondary teachers, and to fold the approaches that prove themselves into the national education system. Its findings, due later this year, are meant to be useful well beyond Estonia's own schools.

This case study was written with assistance from artificial intelligence.