Reading performance is one of the starkest findings in PISA 2025. It's not as simple as stating that students are reading less effectively. Many of the skills showing the greatest decline are the skills students need most to use AI well, such as evaluating credibility, identifying bias, distinguishing evidence from opinion, and constructing arguments from complex information.

‍

These are precisely the abilities needed to navigate a world saturated with AI-generated content. This changes what product teams should be designing. Many educational technologies have been built around reducing the reading burden through automated summaries, simplification, and content compression.

‍

Yet the strategic direction of education policy is moving the other way. Governments need students who can critically engage with complex information, not simply consume simplified versions of it. My advice to product teams is to prepare for growing demand around solutions that develop: deep reading, sustained attention, critical evaluation, information literacy, and source validation.

‍

The next generation of literacy products must look very different from the last. Product strategy should be moving from how quickly learners can access information to whether they can meaningfully interrogate it.

For years, my advisory conversations around educational equity focused on basic access. Did learners have devices? Did schools have bandwidth? Did students have internet access at home?

‍

Those questions are still relevant, but PISA 2025 shows us how AI is changing education. While access to AI tools is becoming widespread, effective use is not. In my research and field observations, students from different socioeconomic backgrounds appear increasingly likely to have very different levels of AI literacy.

‍

This creates a new form of educational inequality. Some learners will understand how to prompt, evaluate, verify, critique, and collaborate with AI systems. Others will simply consume AI outputs without understanding their limitations.

‍

When advising EdTech providers seeking expansion in emerging markets across MENA, Africa, Latin America, and Southeast Asia, I highlight this as a critical factor. Many products currently assume a baseline level of AI literacy that simply does not exist yet, creating adoption and learning risks. The most successful global products will explicitly teach AI literacy as part of the core experience. In upcoming procurement cycles, equity will be evaluated as a design choice.

‍

In other words: can your product help close the AI capability gap, or does it unintentionally widen it?

PISA 2025 is unlikely to end the screen time debate, but it will sharpen it. More technology is not automatically better. Digital devices support learning up to a point, countries where students used devices for two to three hours per day saw higher scores. Beyond that, scores fell.

‍

The schools and universities I work with are becoming far more demanding as a result. They want answers to questions such as:

‍

• How much time does your product require?
• What activities does it replace?
• How does usage correlate with outcomes?
• What level of teacher oversight exists?
• How do you prevent misuse?

‍

I regularly remind sales and product leads that reporting engagement metrics is no longer enough. Procurement teams are starting to view high screen-time usage with scepticism unless it is directly linked to measurable learning gains.

‍

ISTE+ASCD's From Screentime to Learning Time offers a useful evidence-informed framework for this discussion, distinguishing the quality and educational purpose of screen use from its duration, and emphasising the importance of instructional design and implementation. These questions are being explored in practice too.

‍

International Schools Partnership has spent several years developing its approach to Screen Value through its LabSchools, while Connecticut's guidance on purposeful technology use and CoSN's guidance for school districts seek to distinguish educational value from screen exposure alone. These initiatives represent different approaches, and do not establish that any single screen-value metric has been validated.

‍

When guiding product roadmaps, I recommend incorporating teacher controls, time-on-task analytics, usage transparency, learning impact dashboards, and evidence linking use patterns to outcomes.

‍

The future conversation will be about productive screen time, not screen time alone.

In England, the Department for Education's EdTech Evidence Board pilot findings, published in September 2026, describe work to test independent assessment of evidence supporting EdTech effectiveness. The report finds this kind of evaluation feasible and valued, while highlighting the need for stronger, more accessible evidence. EDT&Partners is involved through VP of Strategy Laurie Forcier's participation on the Board. This is a useful example of efforts to improve decisions about educational technology, rather than evidence acting solely as a procurement hurdle.

‍

And PISA 2025 has given every ministry a 91-country dataset suggesting that AI tools, when unstructured, correlate with worse outcomes.

‍

For years, EdTech companies could differentiate primarily through feature lists. Today, feature parity arrives rapidly, accelerated by AI. What remains scarce is credible evidence that a solution genuinely improves learning. I coach teams to treat evidence as a core strategic asset.

‍

The organisations I work with build evaluation into the product architecture from day one, knowing:

‍

• What outcomes matter
• How those outcomes are measured
• What usage patterns drive success
• Which learner groups benefit most
• Where limitations exist

‍

Evidence has become a primary growth strategy because buyers now have access to dozens of competing products that look functionally identical on paper. The companies that can show independent outcomes data will outcompete those that can only show feature lists.

PISA 2025 introduced a new assessment domain: Learning in the Digital World, measuring students' capacity to use modelling and programming tools, conduct experiments, and design digital products. Almost two-thirds of OECD students reached the proficiency level, with Macao, Singapore, Japan, and Estonia among the top performers.

‍

This is early data, but the direction is clear: computational and AI competencies are moving from enrichment topics to core assessment categories.

‍

In my strategy workshops, I emphasise the core components of this trend: coding, data literacy, systems thinking, algorithmic reasoning and problem decomposition.

‍

For EdTech companies building in this space, the product roadmap is now aligned with where assessment frameworks are heading. Unlike areas where EdTech demand has stabilised, these capabilities connect directly to workforce development priorities and national AI strategies. The full Learning in the Digital World results are due in May 2027, and the OECD's Media and AI Literacy (MAIL) assessment is planned for 2029. EdTech leaders should act on what this opens up.

In my recent work with Tajikistan's Ministry of Education, serving as an assessment specialist on a three-year national STEM reform programme funded by the Asian Development Bank, I have watched a country enter the PISA framework for the first time, and seen how global benchmarks meet local classroom realities.

‍

For emerging education systems, entering PISA is a major milestone, but it also creates an immediate operational challenge. While international policy debates focus on AI integration, many project schools on the ground are working through a more foundational transition: moving from recall-based testing to competency-based, classroom-level assessment in mathematics, physics, chemistry, and biology.

‍

When advising EdTech founders and product leaders looking at growth markets across Central Asia and beyond, I use the Tajik context to highlight that global readiness is not uniform:

‍

• Meeting realistic infrastructure baselines: In many regions, schools still rely primarily on physical STEM textbooks. Advanced digital tools must be designed to complement traditional learning materials, not assume their immediate replacement.
• Focusing on teacher capacity over feature complexity: The immediate bottleneck is rarely software capability, but teacher workload and familiarity with skills-based rubrics. Practical, lightweight assessment tools that support teacher professional development build far stronger adoption than complex feature suites.
• Prioritising classroom-level assessment: National exam reform is important, but meaningful performance gains happen when formative assessment tools directly help teachers evaluate problem-solving and critical thinking in everyday lessons.

‍

Markets like Tajikistan need products flexible enough to bridge high-level PISA performance goals with the practical, step-by-step needs of developing school systems.

For EdTech leaders, these findings offer guidance on where schools need support. How can technology help students become stronger readers and critical thinkers? How can AI strengthen rather than replace the cognitive work of learning? How can products help teachers recognise progress, respond to different needs and develop students' capacity to use technology well? These are questions worth building around, whether or not they become explicit procurement requirements.

‍

In mature markets, some patterns keep recurring. Companies optimise for efficiency over cognitive growth. The AI tools boost short-term metrics but compress the productive struggle learners need. In institutional procurement, that is becoming a dealbreaker. And companies still rely on feature lists where procurement teams now expect outcomes evidence.

‍

In emerging markets, companies deploy complex prompt-based interfaces that assume a level of AI literacy that does not exist yet. They design cloud-heavy, high-bandwidth products for schools still working with physical textbooks. They overload teachers with feature-rich software when what is needed are practical, lightweight tools that match existing capacity.

‍

The findings point toward a future where:

‍

• AI in education will be judged on learning outcomes, not productivity gains.
• Reading comprehension and critical thinking are rising as procurement priorities.
• AI literacy is becoming an equity issue, not just an access issue.
• Screen time faces scrutiny tied to measurable learning outcomes.
• Evidence of EdTech impact is replacing feature lists in procurement decisions.
• Computational thinking is moving from enrichment to core assessment.

‍

For EdTech executives, the challenge is anticipating how ministries, districts, publishers, and investors will act on them. It comes back to one question, the one I hear from school leaders, the one procurement teams are building into their frameworks, and the one every EdTech board should be testing their product against:

‍

How does this technology improve learning?

‍

That is the conversation PISA 2025 has started, and it is the conversation every EdTech boardroom should be having now, internally, and with their educator partners.

Does AI help students learn? PISA 2025 suggests the answer depends on how it is used: unstructured use for drafting and summarising correlates with lower scores, while guided use paired with AI literacy instruction correlates with slightly higher ones.

‍

The headline finding from PISA 2025 is that students who use AI chatbots for specific school tasks scored lower in science than those who did not. Some commentators have rushed to conclude that AI itself is the problem. The data tells a more interesting story: students who use AI for general learning purposes and receive structured AI literacy instruction at school scored slightly higher than both heavy users and non-users. The OECD was careful to frame these findings as observational, not causal — students who struggle may be more likely to turn to AI for help, a selection effect rather than a treatment effect.

‍

What matters is how students use AI. Across the education systems I advise, two models are emerging.

‍

The shortcut AI model

In this model, AI performs the cognitive work on behalf of the learner. Students ask for answers rather than explanations. They receive completed essays, not structured guidance. The technology reduces effort and compresses productive struggle. From a product perspective, shortcut AI in learning often looks like:

‍

• Automatic answer generation
• Assignment completion tools
• Instant content summaries
• Automated writing that replaces original thinking
• Assessment support that obscures student reasoning
• No teacher oversight, analytics or controls

‍

These products may create efficiency and boost short-term engagement metrics. However, they risk working against the fundamental capabilities education systems are trying to develop.

‍

The scaffold AI model

In contrast, scaffold AI supports thinking without replacing it. The technology guides, questions, prompts, and coaches. Cognitive effort remains with the learner. In practice, scaffold AI might include:

‍

• Socratic questioning systems
• Writing feedback without providing complete answers
• Guided problem-solving environments
• Metacognitive reflection tools
• Teacher dashboards that make student thinking visible

‍

The difference between shortcut and scaffold AI models will become the defining distinction for policymakers and procurement teams. This is a core product strategy decision. The companies that grow over the next decade will be the ones that build AI to increase learning rather than decrease cognitive effort.

Reading performance is one of the starkest findings in PISA 2025. It's not as simple as stating that students are reading less effectively. Many of the skills showing the greatest decline are the skills students need most to use AI well, such as evaluating credibility, identifying bias, distinguishing evidence from opinion, and constructing arguments from complex information.

‍

These are precisely the abilities needed to navigate a world saturated with AI-generated content. This changes what product teams should be designing. Many educational technologies have been built around reducing the reading burden through automated summaries, simplification, and content compression.

‍

Yet the strategic direction of education policy is moving the other way. Governments need students who can critically engage with complex information, not simply consume simplified versions of it. My advice to product teams is to prepare for growing demand around solutions that develop: deep reading, sustained attention, critical evaluation, information literacy, and source validation.

‍

The next generation of literacy products must look very different from the last. Product strategy should be moving from how quickly learners can access information to whether they can meaningfully interrogate it.

For years, my advisory conversations around educational equity focused on basic access. Did learners have devices? Did schools have bandwidth? Did students have internet access at home?

‍

Those questions are still relevant, but PISA 2025 shows us how AI is changing education. While access to AI tools is becoming widespread, effective use is not. In my research and field observations, students from different socioeconomic backgrounds appear increasingly likely to have very different levels of AI literacy.

‍

This creates a new form of educational inequality. Some learners will understand how to prompt, evaluate, verify, critique, and collaborate with AI systems. Others will simply consume AI outputs without understanding their limitations.

‍

When advising EdTech providers seeking expansion in emerging markets across MENA, Africa, Latin America, and Southeast Asia, I highlight this as a critical factor. Many products currently assume a baseline level of AI literacy that simply does not exist yet, creating adoption and learning risks. The most successful global products will explicitly teach AI literacy as part of the core experience. In upcoming procurement cycles, equity will be evaluated as a design choice.

‍

In other words: can your product help close the AI capability gap, or does it unintentionally widen it?

PISA 2025 is unlikely to end the screen time debate, but it will sharpen it. More technology is not automatically better. Digital devices support learning up to a point, countries where students used devices for two to three hours per day saw higher scores. Beyond that, scores fell.

‍

The schools and universities I work with are becoming far more demanding as a result. They want answers to questions such as:

‍

• How much time does your product require?
• What activities does it replace?
• How does usage correlate with outcomes?
• What level of teacher oversight exists?
• How do you prevent misuse?

‍

I regularly remind sales and product leads that reporting engagement metrics is no longer enough. Procurement teams are starting to view high screen-time usage with scepticism unless it is directly linked to measurable learning gains.

‍

ISTE+ASCD's From Screentime to Learning Time offers a useful evidence-informed framework for this discussion, distinguishing the quality and educational purpose of screen use from its duration, and emphasising the importance of instructional design and implementation. These questions are being explored in practice too.

‍

International Schools Partnership has spent several years developing its approach to Screen Value through its LabSchools, while Connecticut's guidance on purposeful technology use and CoSN's guidance for school districts seek to distinguish educational value from screen exposure alone. These initiatives represent different approaches, and do not establish that any single screen-value metric has been validated.

‍

When guiding product roadmaps, I recommend incorporating teacher controls, time-on-task analytics, usage transparency, learning impact dashboards, and evidence linking use patterns to outcomes.

‍

The future conversation will be about productive screen time, not screen time alone.

In England, the Department for Education's EdTech Evidence Board pilot findings, published in September 2026, describe work to test independent assessment of evidence supporting EdTech effectiveness. The report finds this kind of evaluation feasible and valued, while highlighting the need for stronger, more accessible evidence. EDT&Partners is involved through VP of Strategy Laurie Forcier's participation on the Board. This is a useful example of efforts to improve decisions about educational technology, rather than evidence acting solely as a procurement hurdle.

‍

And PISA 2025 has given every ministry a 91-country dataset suggesting that AI tools, when unstructured, correlate with worse outcomes.

‍

For years, EdTech companies could differentiate primarily through feature lists. Today, feature parity arrives rapidly, accelerated by AI. What remains scarce is credible evidence that a solution genuinely improves learning. I coach teams to treat evidence as a core strategic asset.

‍

The organisations I work with build evaluation into the product architecture from day one, knowing:

‍

• What outcomes matter
• How those outcomes are measured
• What usage patterns drive success
• Which learner groups benefit most
• Where limitations exist

‍

Evidence has become a primary growth strategy because buyers now have access to dozens of competing products that look functionally identical on paper. The companies that can show independent outcomes data will outcompete those that can only show feature lists.

PISA 2025 introduced a new assessment domain: Learning in the Digital World, measuring students' capacity to use modelling and programming tools, conduct experiments, and design digital products. Almost two-thirds of OECD students reached the proficiency level, with Macao, Singapore, Japan, and Estonia among the top performers.

‍

This is early data, but the direction is clear: computational and AI competencies are moving from enrichment topics to core assessment categories.

‍

In my strategy workshops, I emphasise the core components of this trend: coding, data literacy, systems thinking, algorithmic reasoning and problem decomposition.

‍

For EdTech companies building in this space, the product roadmap is now aligned with where assessment frameworks are heading. Unlike areas where EdTech demand has stabilised, these capabilities connect directly to workforce development priorities and national AI strategies. The full Learning in the Digital World results are due in May 2027, and the OECD's Media and AI Literacy (MAIL) assessment is planned for 2029. EdTech leaders should act on what this opens up.

In my recent work with Tajikistan's Ministry of Education, serving as an assessment specialist on a three-year national STEM reform programme funded by the Asian Development Bank, I have watched a country enter the PISA framework for the first time, and seen how global benchmarks meet local classroom realities.

‍

For emerging education systems, entering PISA is a major milestone, but it also creates an immediate operational challenge. While international policy debates focus on AI integration, many project schools on the ground are working through a more foundational transition: moving from recall-based testing to competency-based, classroom-level assessment in mathematics, physics, chemistry, and biology.

‍

When advising EdTech founders and product leaders looking at growth markets across Central Asia and beyond, I use the Tajik context to highlight that global readiness is not uniform:

‍

• Meeting realistic infrastructure baselines: In many regions, schools still rely primarily on physical STEM textbooks. Advanced digital tools must be designed to complement traditional learning materials, not assume their immediate replacement.
• Focusing on teacher capacity over feature complexity: The immediate bottleneck is rarely software capability, but teacher workload and familiarity with skills-based rubrics. Practical, lightweight assessment tools that support teacher professional development build far stronger adoption than complex feature suites.
• Prioritising classroom-level assessment: National exam reform is important, but meaningful performance gains happen when formative assessment tools directly help teachers evaluate problem-solving and critical thinking in everyday lessons.

‍

Markets like Tajikistan need products flexible enough to bridge high-level PISA performance goals with the practical, step-by-step needs of developing school systems.

For EdTech leaders, these findings offer guidance on where schools need support. How can technology help students become stronger readers and critical thinkers? How can AI strengthen rather than replace the cognitive work of learning? How can products help teachers recognise progress, respond to different needs and develop students' capacity to use technology well? These are questions worth building around, whether or not they become explicit procurement requirements.

‍

In mature markets, some patterns keep recurring. Companies optimise for efficiency over cognitive growth. The AI tools boost short-term metrics but compress the productive struggle learners need. In institutional procurement, that is becoming a dealbreaker. And companies still rely on feature lists where procurement teams now expect outcomes evidence.

‍

In emerging markets, companies deploy complex prompt-based interfaces that assume a level of AI literacy that does not exist yet. They design cloud-heavy, high-bandwidth products for schools still working with physical textbooks. They overload teachers with feature-rich software when what is needed are practical, lightweight tools that match existing capacity.

‍

The findings point toward a future where:

‍

• AI in education will be judged on learning outcomes, not productivity gains.
• Reading comprehension and critical thinking are rising as procurement priorities.
• AI literacy is becoming an equity issue, not just an access issue.
• Screen time faces scrutiny tied to measurable learning outcomes.
• Evidence of EdTech impact is replacing feature lists in procurement decisions.
• Computational thinking is moving from enrichment to core assessment.

‍

For EdTech executives, the challenge is anticipating how ministries, districts, publishers, and investors will act on them. It comes back to one question, the one I hear from school leaders, the one procurement teams are building into their frameworks, and the one every EdTech board should be testing their product against:

‍

How does this technology improve learning?

‍

That is the conversation PISA 2025 has started, and it is the conversation every EdTech boardroom should be having now, internally, and with their educator partners.