The New Education Divide: Why Knowing How to Use Learning Tools Matters

 

---

For years, the digital divide in education was largely understood as a question of access: Who has a laptop, a reliable internet connection, educational software or a quiet place to study? That divide has not disappeared. But artificial intelligence is creating another layer to it.

The more difficult question now is whether students know how to use learning tools well.

The distinction matters. A student who asks an AI chatbot to provide an answer is using technology differently from a student who asks it to explain a difficult concept, challenges its reasoning, checks its sources, compares alternative explanations and then solves the problem independently. Both may have access to exactly the same tool. Their learning experience can be very different.

New international evidence makes this distinction increasingly important. OECD data from PISA 2025 show that AI use among students is widespread, but its relationship with academic performance is complex. Students who learn at school how to assess AI-generated information tend to show somewhat better science performance when they also use AI for learning. At the same time, disadvantaged students report fewer opportunities to develop this kind of critical AI literacy.

The emerging divide, therefore, may be less about who can reach the technology and more about who has learned to think with it without allowing it to think for them.

Key Takeaways

• Access to AI does not automatically translate into better learning; how students use it matters considerably.
• OECD data show that AI use and academic performance have a complex relationship rather than a simple positive or negative one.
• Students taught to evaluate AI-generated information appear better positioned to use these tools productively.
• Socio-economic differences can influence who receives opportunities to develop critical AI skills at school.
• Research increasingly points toward AI literacy, metacognition and purposeful use as important educational capabilities.
• The strongest learning model treats AI as a tool for thinking, feedback and exploration—not as a replacement for intellectual effort.

From the Digital Divide to the Learning Divide

The first generation of digital inequality focused on access.

Students without computers, broadband connections or suitable digital resources could be disadvantaged before a lesson even began. Schools and governments responded by expanding infrastructure, distributing devices and increasing access to digital learning platforms.

Generative AI changes the equation because the same tool can produce very different outcomes depending on the student’s knowledge and judgment.

Consider two students researching the same scientific concept.

One asks an AI system for a ready-made explanation, copies the response and moves on. The other asks for an explanation at different levels of difficulty, requests examples, identifies assumptions, challenges an uncertain statement, checks important claims against reliable sources and then attempts the problem without assistance.

The difference is not access.

It is learning strategy.

That distinction is increasingly visible in international data. The OECD’s PISA 2025 results found that students’ relationship with AI and science performance varies according to how AI is used. Students using AI for specific tasks such as summarising, drafting or preliminary research generally scored lower in science than students who did not use AI for those tasks. But students who used AI more generally to help them learn on a weekly basis showed science performance similar to non-users after accounting for socio-economic status.

That finding should not be interpreted as proof that AI lowers achievement. PISA is largely observational in this area, so it cannot establish that AI use caused the differences. The important point is that simply measuring whether a student uses AI tells us very little about whether the technology is helping that student learn.

The Skill That Matters: Knowing What to Ask and What to Doubt

AI literacy is broader than prompt writing.

UNESCO’s AI Competency Framework for Students identifies 12 competencies across four dimensions: a human-centred mindset, ethics of AI, AI techniques and applications, and AI system design. The framework progresses from understanding to applying and eventually creating. It also emphasizes critical judgment, responsible participation and human agency.

For students, practical AI literacy can involve several different capabilities:

• Knowing when an AI tool is appropriate for a learning task.
• Formulating questions that encourage explanation rather than answer extraction.
• Recognising that AI-generated information can be inaccurate or incomplete.
• Checking important claims against reliable sources.
• Comparing competing explanations instead of accepting the first response.
• Using AI for feedback, brainstorming or practice without outsourcing the entire intellectual task.
• Recognising when independent reasoning is more valuable than assistance.
• Understanding academic, ethical and privacy implications of AI use.

This changes the definition of being a “good technology user.”

Technical fluency alone is not enough. A student may know dozens of AI features and still use them poorly. Conversely, a student who understands the limitations of a single tool may extract substantially more educational value from it.

Research is beginning to reflect this distinction. A 2025 study following university students over 16 weeks found that sustained interaction with generative AI was associated with the development of prompt-engineering skills, metacognitive awareness, critical evaluation of AI outputs and personalised workflows. The researchers described AI literacy as something that develops through practice, reflection and experience rather than simple exposure to a tool.

More AI Does Not Necessarily Mean More Learning

One of the easiest mistakes in the AI debate is to treat usage frequency as a measure of educational benefit.

It isn’t.

A student who uses AI every day to avoid difficult thinking may be learning less effectively than a student who uses it twice a week as a carefully controlled tutor or feedback mechanism.

The PISA 2025 findings reinforce this caution. Across OECD countries, 46% of students reported using AI chatbots weekly or more to help them learn. Yet the OECD found that the relationship between AI use and performance depended strongly on the purpose and context of use.

There is also evidence that the educational value of generative AI can be positive under appropriate conditions. A 2025 systematic review and meta-analysis of experimental studies covering 68 studies found a moderate overall positive effect of GenAI-supported interventions on learning outcomes, while also reporting substantial variation between contexts.

Another 2025 meta-analysis covering 57 studies and more than 5,300 participants found positive effects across several university learning outcomes, including academic achievement and higher-order thinking, while finding no statistically significant effect on metacognition.

The message from this research is therefore more nuanced than either “AI improves education” or “AI is ruining education.”

AI can help learning, but the educational design around the tool matters.

The New Inequality May Begin Inside the Classroom

This is where the issue becomes more consequential.

If AI literacy becomes an important learning skill, schools are not simply deciding whether students should have access to AI. They are also determining which students receive structured opportunities to learn how to use it critically.

The OECD found that around six in ten students across OECD countries reported that they had been asked in school lessons to assess the quality of information generated by AI. But these opportunities were not evenly distributed. Socio-economically disadvantaged students were less likely to report being given such tasks.

That creates a potentially important second-order divide.

Students from more advantaged backgrounds may have access not only to devices and AI services, but also to teachers, parents, schools and learning environments that help them understand how to question, verify and strategically use those tools.

Students with fewer educational resources could instead encounter AI primarily as an answer generator.

The difference can compound over time.

One group learns how to interrogate a tool.

Another learns how to consume its output.

Both may technically have access to the same AI system.

What Schools Should Teach Instead of Simply Banning or Promoting AI

The educational response does not have to be a choice between unrestricted AI use and blanket prohibition.

A more useful approach is to teach students how to use AI in ways that preserve the learning objective.

For example, instead of asking an AI system to write an essay, a teacher could ask students to:

1. Develop their own argument first.
2. Use AI to challenge or critique that argument.
3. Investigate the AI’s claims independently.
4. Identify inaccuracies or missing perspectives.
5. Revise their work based on evidence.
6. Explain which parts of the final work reflect their own reasoning.

The same principle can apply to mathematics, science, programming, languages and research.

AI can become a practice partner, explainer, critic, simulator or source of alternative perspectives rather than simply a machine for producing finished assignments.

UNESCO’s student competency framework similarly places emphasis on human agency, ethics, critical judgment and moving from understanding AI toward applying and creating with it.

Teachers matter just as much.

UNESCO’s separate AI competency framework for teachers identifies competencies spanning human-centred thinking, AI ethics, AI foundations and applications, AI pedagogy and professional learning.

That suggests a crucial principle: students cannot be expected to develop sophisticated AI literacy if educational systems do not give teachers the knowledge and support to teach it.

The Real Advantage May Be Metacognition

The deepest educational issue is not whether students know a particular chatbot.

Tools change.

Today’s leading model may be replaced by another system tomorrow. Interfaces will change. Features will disappear. New capabilities will emerge.

The durable skill is knowing how to learn when powerful tools are available.

That requires metacognition: understanding what you know, what you do not know, what kind of help you need and whether the assistance is actually improving your understanding.

This is particularly important because AI can make intellectual tasks feel easier without necessarily making the learner more capable.

A generated explanation may remove the frustration of solving a difficult problem. But some frustration is part of learning. Struggling to formulate an argument, retrieve information, test an assumption or debug a program can build knowledge that disappears if the tool performs every difficult step.

This is why the distinction between assistance and substitution matters.

If AI helps a student identify a misconception, learning may become stronger.

If AI removes the need to identify the misconception, the student may simply finish the task faster.

Those are not equivalent educational outcomes.

What the New Education Divide Could Look Like

The emerging divide is unlikely to be a simple line between students who use AI and those who do not.

It may instead look more like a spectrum:

Student approach	Typical use of learning tools	Potential educational value
Passive	Requests finished answers	Limited if understanding is bypassed
Convenience-focused	Uses AI to summarise or complete routine tasks	Useful for efficiency, but risk of shallow engagement
Guided	Uses AI for explanations, examples and feedback	Greater potential for understanding
Critical	Questions, verifies and compares AI outputs	Builds information and AI literacy
Strategic	Chooses when to use AI and when to work independently	Supports autonomy and metacognition
Creative	Uses multiple tools to explore, test and develop ideas	Potentially expands problem-solving and creation

This is not a scientific classification of students; it is a practical way of understanding the central distinction emerging from the evidence.

The important question is therefore not “Does this student use AI?”

It is:

“What does the student do with the AI, and what remains the student’s own intellectual work?”

The Advantage Will Belong to Students Who Can Use Tools Without Depending on Them

The education systems that respond well to AI will probably not be those that simply provide the most technology.

They will be the ones that teach students when technology helps, when it interferes and how to distinguish between the two.

That means AI literacy should not become another isolated technology lesson. It belongs inside reading, writing, mathematics, science, research, media literacy and problem-solving.

The OECD’s PISA findings offer an important warning here: access alone does not guarantee better learning, and the opportunities to develop critical AI skills are themselves unevenly distributed.

For students, the practical lesson is straightforward. Learn the tools but also learn their weaknesses. Use AI to ask better questions, expose gaps in your understanding, test ideas and receive feedback. Verify important information. And regularly work without assistance so that the underlying skill remains yours.

For schools, the challenge is larger.

The goal should not be to produce students who are dependent on AI, nor students who are afraid to use it.

It should be to produce students who can think independently in a world where intelligent tools are always available.

Conclusion

The first digital education divide was about who could get online. The emerging divide is more subtle: who has learned to turn powerful digital tools into genuine learning advantages.

That distinction matters because the same AI system can either strengthen a student’s thinking or quietly replace it.

The evidence does not justify declaring AI either a cure for educational inequality or a threat to learning. What it does suggest is more practical: purpose, guidance, critical evaluation and human judgment determine much of the value students get from these tools.

The most valuable student in the AI-enabled classroom will not necessarily be the one who knows the most tools.

It may be the one who knows when to use them, how to question them and when to put them away and think alone.