Author: wadiagounine

Research update — 23 October 2026. Adaptive learning in primary maths is often promoted as a way to personalise practice for every child. A research update from Radboud University on 2 October reported a large, multi-year study of 7,885 Dutch primary-school pupils: children using an adaptive learning system showed slightly more positive growth in mathematics performance than peers in classes that did not use it. The result is encouraging, but its limits matter just as much as the headline.

The key lesson is not that software can replace a teacher. It is that adaptive practice may help teachers respond to differences between learners—when it is connected to a clear curriculum, interpreted by educators, and evaluated through learning rather than screen activity.

What did the 7,885-pupil study find?

According to Radboud University’s research summary, researchers examined the development of numeracy skills among 7,885 Dutch primary-school pupils over a three-year period. The data covered the years 2014–2017, when classes using an adaptive learning system were compared with classes teaching mathematics without that system.

The software adjusted the difficulty of exercises in response to pupils’ performance: learners doing well received more challenging tasks, while those struggling received simpler ones. The university reported a small, generally positive trend in maths growth. The effects appeared more evident in larger schools and in schools serving more socio-economically disadvantaged pupils, where differences in prior attainment may be harder for one teacher to address simultaneously.

The study is reported as Arithmetic Development: A Longitudinal Observational Study into the Use of Adaptive Technology, listed by the university as forthcoming in Learning and Instruction. The important distinction is methodological: this was a longitudinal observational study, not a randomised trial. It can identify a pattern across a large sample, but it cannot by itself establish that the software caused every difference observed.

Why “adaptive learning” is not the same as a chatbot

Searches for AI learning tools often group very different technologies together. An adaptive maths platform typically selects or adjusts practice tasks using a learner’s answers and progress data. A generative AI chatbot produces new explanations or responses in natural language. Some products combine both, but the evidence for one type should not automatically be treated as evidence for the other.

The Dutch study examined adaptive learning technology used in 2014–2017—well before today’s generative AI chatbots became widely available. It therefore offers useful evidence about personalised practice and learner progression, but it does not demonstrate that a conversational AI tutor improves maths learning. For a separate discussion of how to judge education tools, see our guide to evaluating AI learning tools in schools.

What the findings mean for primary-school teachers

1. Use software to target practice, not to define the curriculum

Adaptive systems can vary the pace and difficulty of practice, but teachers still need to decide which mathematical ideas pupils should understand. A platform’s next-task recommendation is not a substitute for curriculum sequencing, classroom discussion, or professional judgement. Before adopting a tool, map its activities to the learning objectives and identify which concepts require direct explanation or concrete materials.

2. Ask pupils to explain their reasoning

Correct answers alone do not reveal whether a child understands place value, has memorised a procedure, or guessed successfully. Pair digital practice with brief explanations: “How did you work it out?”, “Why does this answer make sense?”, or “Can you show another method?” Teachers can use recurring errors as prompts for a small-group lesson instead of assuming that a progress dashboard tells the whole story.

3. Check transfer away from the platform

Improvement inside an app is not necessarily the same as durable learning. Periodically give pupils a short, curriculum-aligned task on paper or at the board, without hints from the software. Include a new problem that requires the same underlying idea in a different form. If performance drops sharply when the tool is removed, the next step may be more explanation and guided practice—not simply more app time.

4. Keep the teacher in the feedback loop

The Radboud report explicitly stresses the continuing role of teachers. A useful workflow is to review the platform’s error patterns, compare them with classwork, select one misconception to address, and then check whether pupils can apply the correction independently. Automated feedback can make practice more responsive, but it may not reveal language difficulties, anxiety, gaps in prior knowledge, or the reason a child chose an answer.

5. Test whether the tool is equitable in your setting

The reported stronger pattern in larger and more disadvantaged schools is worth investigating, not a guarantee that every under-resourced school will benefit. Schools should check device availability, connectivity, accessibility, language support, and whether the tasks reflect pupils’ curriculum and context. Ask which learner data are collected, how long they are retained, who can access them, and whether teachers can export or delete them under the school’s policies.

How schools can evaluate an adaptive maths tool

Before expanding a platform, schools can run a small, time-limited evaluation. Agree on the learning objective first, record a baseline measure, and decide what evidence would count as improvement. Where feasible, compare results with a similar class or with pupils’ performance before implementation, while acknowledging that simple before-and-after comparisons cannot rule out other explanations.

• Learning: Do pupils improve on curriculum-aligned questions, including questions they have not practised in the platform?
• Retention: Can they still solve comparable problems after a delay?
• Reasoning: Can they explain a method and detect an implausible answer?
• Equity: Are gains and access similar across groups, or are some pupils receiving less useful practice?
• Teaching: Does the system give teachers actionable information and free time for instruction, or add monitoring work?
• Privacy and reliability: Are data use, security, accessibility and failure procedures documented?

These checks complement, rather than replace, the research literature. A 2025 open-access meta-analysis of 34 empirical studies on learning-analytics interventions reported an overall positive effect on learning outcomes, while also finding substantial variation across studies. That broader evidence concerns learning analytics interventions generally—not only AI systems or primary mathematics—so it supports careful evaluation rather than a universal claim of effectiveness. See the published meta-analysis for methods and limitations.

What the evidence does not yet tell us

Three cautions should shape decision-making. First, a small average benefit can conceal variation between pupils, schools, and implementation models. Second, data from one national context and a particular generation of adaptive software may not transfer unchanged to schools with different curricula, languages, resources, or teaching practices. Third, evidence about adaptive exercise selection should not be used to make stronger claims about generative AI tutors, which introduce different questions about accuracy, explanation quality, privacy, and student dependence.

Research on AI-supported learning more broadly similarly points to implementation as a central issue: a recent review in the Italian Economic Journal argues that the impact of computer-assisted learning and AI-guided tutors depends on curriculum alignment, teacher preparation, interactive design, and sustained human support. Read the open-access review of AI-guided tutors and computer-assisted learning for a broader synthesis.

A practical conclusion for schools

The 7,885-pupil study offers a useful reminder that adaptive learning in primary maths should be judged by the quality of learning it supports, not by the novelty of the technology. Its reported gains were small, the design was observational, and the data predate the current generative-AI wave. The most defensible response is neither blanket adoption nor blanket rejection: choose tools that fit the curriculum, use their data to inform—not automate—teacher decisions, and verify progress with independent tasks and transparent evaluation.

For a related framework on implementation planning, read our article on a five-step framework for integrating AI in education. For subject-specific design principles, see our guide to AI chatbots and higher-order thinking in STEM.

Further reading

• Radboud University: Learning maths goes (slightly) better with AI, but teacher plays a key role (2 October 2026).
• The Effectiveness of Learning Analytics-Based Interventions in Enhancing Students’ Learning Effect (2025).
• Artificial Intelligence in Education: Computer-Assisted Learning and AI-guided Tutors (published online 2025).

Generative AI is often introduced into STEM lessons as a faster way to explain a concept, solve a problem, or produce an answer. A different design goal is emerging: an AI chatbot can be built to make students explain, compare, and justify their own reasoning instead of doing the thinking for them.

A study published online on 14 October 2026 in the Journal of Microbiology & Biology Education offers a timely proof of concept. Researchers developed the UCI Math ThinkBot to prompt fifth- and sixth-grade students to reflect on proportional-reasoning strategies. The study is promising, but it is not evidence that chatbots automatically improve learning. Its value lies in the instructional design choices it makes visible.

This article examines what the study found, what it cannot yet establish, and how teachers and learning designers can apply the principles responsibly.

What the October 2026 study actually tested

Richland and colleagues studied 33 students in grades five and six. After a short lesson on ways to compare ratios, students completed an assessment and then spoke with Math ThinkBot about how they had solved the problems. Rather than delivering a worked solution, the chatbot asked learners to describe their strategy, explain why it worked, compare alternative methods, and reflect on which approach was useful in different circumstances.

The researchers examined the dialogue for “higher-order thinking talk” (HOTT): language that signals comparison, relationships, inference, justification, or generalisation. Their qualitative and frequency-based analyses found HOTT cues in both the chatbot’s prompts and students’ responses. The amount of HOTT language produced by the bot and by children was positively associated.

That is a useful finding about the interaction: carefully designed dialogue can create more opportunities for students to articulate reasoning. But the study was a proof of concept with a small sample, and the chatbot conversation occurred after the lesson and assessment. The researchers did not measure whether the chatbot caused learning gains. Cue-word counts are also only a rough proxy for the quality of thinking; saying “compare” is not the same as making a sound comparison.

Primary source: Richland et al. (2026), “Developing a generative AI chatbot to support students’ higher-order thinking in STEM education,” Journal of Microbiology & Biology Education.

Why the design matters more than the chatbot label

“AI tutor” can describe very different experiences. One system may provide a complete answer after a short prompt; another may ask the learner to explain a hypothesis, inspect evidence, compare strategies, and revise an argument. Both use generative AI, but they create different demands on the learner.

Math ThinkBot was designed to avoid simply supplying the answer. Its prompts were grounded in the relevant mathematics, sample dialogues, common misconceptions, and alternative solution strategies. The researchers also constrained the chatbot to use short, student-friendly turns. These are not minor interface details: they define the pedagogical role the model is being asked to play.

The distinction connects with broader teacher-development work. A 2026 UNESCO IITE analysis of 36 national, institutional, and academic teacher AI competency frameworks stresses that AI competence should be pedagogically grounded and aligned with curriculum, implementation, and assessment—not reduced to knowing how to operate a tool. See the UNESCO IITE report on Teacher AI Competency Frameworks.

Six design principles for AI chatbots that support higher-order thinking

1. Ask for reasoning before offering an explanation

Start with prompts such as “How did you approach the problem?”, “What makes you think that?”, or “Which evidence supports your claim?” The aim is not to withhold help indefinitely. It is to make the learner’s current thinking visible before the system supplies additional information.

For example, when a student solves a ratio problem, the chatbot might ask the student to describe the relationship between the two quantities before showing another method. Teachers can then distinguish a correct answer supported by understanding from one reached by guessing or copying.

2. Make comparison a deliberate learning move

Comparison can reveal the structure of a concept. Ask students to compare two solution methods, models, scientific explanations, or interpretations. Useful prompts include “What is similar?”, “Where do the methods differ?”, “Which approach is more efficient here, and why?”, and “Would your reasoning still work if the numbers changed?”

These questions encourage learners to connect procedures with underlying relationships. They also help teachers see whether a student can transfer an idea beyond the exact example that was practised.

3. Use scaffolding without taking over

A productive chatbot should respond to the learner’s answer with a next step matched to the apparent difficulty. If a response is vague, it can request one specific reason. If the learner is stuck, it can offer a hint, a representation, or a smaller subproblem. It should not immediately replace the learner’s attempt with a polished solution.

Scaffolding needs boundaries. A model can misread a student’s intent, provide incorrect feedback, or keep asking questions that do not fit the curriculum. Teachers should be able to inspect the prompts, correct misconceptions, and intervene when the dialogue stops helping.

4. Ground the chatbot in the curriculum and likely misconceptions

General-purpose chatbots may generate plausible but irrelevant explanations. A subject-specific learning design should specify the target concept, grade level, permitted methods, common misconceptions, expected representations, and the kinds of help the system must not provide. These materials should be reviewed by a teacher or subject expert.

Grounding does not eliminate error. The chatbot should be tested against a set of realistic student responses, including incomplete reasoning, incorrect answers, ambiguous language, and attempts to get the final answer without explanation.

5. Keep each turn short and cognitively focused

Long, multi-part responses can overload learners and make it difficult to know what to do next. One focused question at a time is easier to answer and easier for a teacher to review. A useful turn should serve a clear purpose: elicit a reason, invite a comparison, test a generalisation, or help the learner reflect.

Age, reading ability, language proficiency, disability, and accessibility needs matter. Plain language, flexible response formats, and a way to request a hint can make dialogue more usable without lowering the intellectual goal.

6. End with independent transfer, not a chatbot transcript

A strong interaction should lead to a task the student completes without AI assistance. Ask the learner to solve a new problem, explain a concept to a peer, critique a new claim, or apply the same principle in a different context. This makes it possible to test whether the learner can use the idea independently.

The chatbot conversation itself is not proof of learning. Students may echo the system’s vocabulary, and a fluent explanation produced with support may not be reproducible later. Independent work is therefore essential when the educational goal is durable understanding.

A practical classroom workflow

1. Set the learning objective. Name the concept or reasoning practice students should demonstrate, such as explaining proportional relationships or evaluating evidence for a scientific claim.
2. Collect an initial attempt. Have students work independently first. Preserve their reasoning, not only the final answer.
3. Run a bounded dialogue. Use a teacher-reviewed chatbot prompt that asks students to explain, compare, and reflect. Make clear that the system is a thinking partner, not an answer key.
4. Review the interaction. Sample transcripts for conceptual errors, unhelpful loops, overconfident claims, and whether students did the reasoning themselves.
5. Assess transfer without AI. Give students a parallel problem or short explanation task. Compare the quality of reasoning with the initial attempt using a consistent rubric.
6. Revise before scaling. Discuss results with students and teachers, adjust the prompt or task, and repeat on a limited scale before broader adoption.

How to evaluate whether the chatbot is helping

Schools should measure learning outcomes, not merely how often students use a tool or how satisfied they report feeling. A small pilot can track several complementary indicators:

• Reasoning quality: Does the student justify claims, connect representations, identify assumptions, or compare methods accurately?
• Independent transfer: Can the student solve a new task without the chatbot?
• Misconception handling: Does the system surface and address errors without reinforcing them?
• Participation and access: Who can use the system effectively, and who encounters language, disability, device, or connectivity barriers?
• Teacher workload: Does the design save time overall, or does reviewing transcripts and correcting errors add more work?
• Safety and privacy: Are student data minimised, access controlled, and retention practices explained to families and staff?

Where feasible, compare the AI-supported activity with a credible alternative, such as teacher-led reflection or peer discussion. Use the same learning objective and comparable assessments. If sample sizes are small or conditions differ, report results as exploratory rather than as causal proof.

For a wider institutional checklist, see Prowad’s guides to evaluating AI learning tools with evidence and building an AI integration framework for education. They address adoption and governance; this article focuses on the interaction design that can support students’ reasoning.

Risks and open research questions

Several questions remain unresolved. The Math ThinkBot study involved 33 children, one mathematical context, and reflective dialogue after instruction. It did not establish durable learning gains, effectiveness across other STEM subjects, or superiority over teacher-led discussion. Future studies need more diverse samples, direct measures of learning, independent transfer tasks, and comparisons with alternative instructional approaches.

There is also a measurement challenge. Counting words such as “why,” “same,” or “different” can indicate that a conversation contains reasoning prompts, but it cannot fully capture whether the reasoning is correct, original, or conceptually deep. Human review or validated semantic coding is needed to judge the substance of students’ explanations.

Finally, schools should avoid assuming that more AI interaction is always better. Prowad has previously examined evidence that an AI study assistant can backfire when its design changes how students engage with learning. The relevant question is not whether AI is present, but whether the task preserves meaningful student effort and produces evidence of independent understanding.

Conclusion

AI chatbots for higher-order thinking should be designed to elicit reasoning rather than replace it. The October 2026 Math ThinkBot study offers a concrete example: ask students to explain strategies, compare alternatives, and reflect on conceptual relationships. The result is promising evidence about the potential of structured dialogue, not proof that a chatbot improves achievement.

For teachers and institutions, the most defensible next step is a bounded, curriculum-aligned pilot with teacher oversight and an independent transfer assessment. If students can explain and apply the concept after the chatbot is gone, the design may be supporting learning. If the main output is simply a fluent conversation, more evidence is needed.

References

• Richland, L. E., et al. (2026). Developing a generative AI chatbot to support students’ higher-order thinking in STEM education. Journal of Microbiology & Biology Education. Published online 14 October 2026.
• UNESCO Institute for Information Technologies in Education. (2026). Analytical Report: Teacher AI Competency Frameworks.
• Yang, Y., Chen, L., Sun, D., Zhu, W., & Pang, S. (2026). Fostering preservice teachers’ design thinking by integrating generative AI and analytics-supported reflective assessment in a collaborative inquiry environment. Computers & Education, 249, 105602.

AI integration in education is no longer simply a question of whether teachers and students can access a chatbot. The harder question is whether schools and universities can turn informal experimentation into learning practices that are safe, equitable, measurable and aligned with curriculum goals.

Two studies published in Morocco in October 2026 illustrate why this distinction matters. One mixed-methods paper reports widespread informal generative-AI use among a regional sample of educators and students, alongside much less institutionally guided use. A separate qualitative study of 15 postgraduate students describes both practical benefits and concerns about confidence and intellectual independence. Neither study represents every Moroccan institution, let alone the world. Together, however, they provide timely case evidence for a broader implementation problem: access can spread faster than pedagogical guidance.

This article translates that problem into a practical five-step framework for education leaders. It combines the recent studies with international competency guidance and identifies what institutions should measure before expanding AI use.

Why AI integration in education needs an implementation plan

Schools and universities often begin with individual experiments: a teacher generates quiz questions, a student asks a chatbot to explain a difficult concept, or a department uses AI to draft learning materials. These activities may be useful, but scattered experimentation does not automatically create a coherent educational strategy.

A functioning implementation plan connects five elements: a defined learning problem, teacher and learner competence, rules for data and academic integrity, an instructional design, and evidence that the approach is helping. Without those links, institutions may measure tool adoption rather than learning.

The distinction is increasingly reflected in formal guidance. The European Commission and OECD presented an AI Literacy Framework for primary and secondary education on 18 June 2026. It describes four dimensions of engagement with AI and 19 competences, and it is intended to help educators turn AI literacy into classroom learning rather than treating it as a narrow prompt-writing skill. UNESCO’s AI competency framework for teachers similarly emphasizes human-centred practice, ethics, AI foundations, AI pedagogy and professional learning.

These frameworks are not plug-and-play implementation plans for every institution. They are reference points that schools must adapt to local curriculum, infrastructure, languages, learner needs and legal obligations.

What two recent Moroccan studies can—and cannot—tell us

In an article published on 14 October 2026, Mohsine Khrou examined generative-AI use in secondary and higher education in Morocco. The study describes a mixed-methods design involving 320 participants in the Oriental region: 80 teachers and 240 students. It reports 74% informal use, 12% guided or institutional use, and 14% non-use or rejection by caution. The article also describes concerns about AI literacy, infrastructure, critical reflection and institutional alignment. Read the original study in African Scientific Journal.

Those figures should be interpreted carefully. The sample is regional and combines distinct groups; the reported percentages are not a nationally representative estimate of AI use across Morocco. The article is useful as a local signal of a possible gap between informal uptake and structured guidance, not as proof that every school faces the same pattern.

A second paper, published in Discover Education on 17 October 2026, explored the experiences of 15 postgraduate students from two Moroccan public universities through qualitative interviews. Participants described using AI for research support, explanation, language editing and lesson preparation, while some also reported unease about depending on AI validation. Because the sample is small and purposively selected, the findings are exploratory rather than causal or nationally generalizable. The full open-access article offers a closer view of the tensions that implementation policies need to address.

The practical conclusion is not that institutions should ban AI or adopt it everywhere. It is that they should define the intended educational purpose, make expectations explicit, prepare educators and evaluate outcomes before scaling.

A five-step framework for responsible AI integration

Step 1: Start with a learning problem, not a product

Before selecting a platform, specify the problem that needs solving. Examples include helping multilingual learners rehearse academic vocabulary, giving students formative practice between lessons, reducing repetitive teacher preparation, or supporting accessible explanations of complex content.

Write a short problem statement that names the learners, the curriculum objective, the current difficulty and the desired improvement. Then ask whether AI is actually needed. A carefully designed worksheet, peer discussion, teacher conference or existing accessibility feature may be simpler, cheaper and more reliable.

For each proposed use, state what the AI may do and what remains a human responsibility. For example, AI may suggest feedback questions, but the teacher checks their accuracy and decides how they fit the learner’s needs.

Step 2: Build role-specific AI literacy

Training should go beyond demonstrations of popular tools. Teachers need to judge generated content, verify claims, protect learner information, design meaningful activities and recognize when an AI system is inappropriate. Students need to know how to question outputs, compare them with credible sources, disclose use where required and preserve their own reasoning.

Use different learning outcomes for different roles. A school leader may need to evaluate procurement and governance; a teacher may need to redesign an assessment; a student may need to distinguish a plausible answer from a supported explanation. A single introductory workshop cannot establish competence across all these tasks.

For a deeper discussion of teacher preparation, see Prowad’s analysis of AI teacher-training competency frameworks. For practical models of teacher–AI collaboration, see our guide to four teacher–AI collaboration models. For learner-facing instruction, the article on measuring AI literacy beyond prompting explores why performance and verification matter.

Step 3: Publish clear rules before high-stakes use

An institutional AI policy should answer practical questions in plain language:

• Which uses are encouraged, restricted or prohibited in each course or activity?
• When must students disclose AI assistance, and what evidence of their own work should they retain?
• Which information must never be entered into public AI services?
• Who checks accuracy, bias, accessibility and age suitability before a tool is adopted?
• How can a learner or teacher challenge an AI-supported decision?
• Who is accountable when the system produces an error or an unfair result?

Rules should distinguish low-risk preparation from consequential decisions. Drafting practice questions is not equivalent to using AI to grade a high-stakes assessment, determine admission, or infer a learner’s emotional state. The more consequential the use, the stronger the required human oversight, documentation and review should be.

Policies also need to respect applicable privacy, copyright, child-protection, accessibility and data-protection requirements. This article is an educational planning guide, not legal advice; institutions should have their own legal and data-protection specialists review deployments where required.

Step 4: Pilot one bounded use case and compare it with a baseline

Do not scale a tool because staff like the demonstration or because it produces polished outputs. Run a limited pilot with a defined learning objective, a realistic comparison and an agreed stopping rule.

A pilot plan should record:

• Baseline: what learners can do before the intervention, using an assessment aligned with the objective.
• Comparison: where feasible, compare the AI-supported approach with existing practice or another appropriate instructional design.
• Process: how often learners use the tool, what support they receive, and how much teacher time is required.
• Outcomes: performance on the target skill, retention after a delay, transfer to a new task and learner access across groups.
• Safeguards: errors, privacy incidents, accessibility barriers, inappropriate outputs and complaints.

When random assignment is impractical, institutions can still collect useful evidence through pre/post measures, matched groups, structured observations and learner interviews. They should describe the limits of the design and avoid claiming that AI caused an improvement when other explanations remain plausible.

Schools evaluating a specific platform can use Prowad’s seven evidence checks for AI learning tools as a companion resource.

Step 5: Scale only after reviewing evidence, equity and workload

At the end of a pilot, convene teachers, learners, technical staff and leadership to review the evidence. Ask not only whether average performance changed, but also who benefited, who struggled, which tasks became easier, and whether the new workflow introduced hidden work or dependence on a vendor.

Scale in stages. A successful use case in one subject or age group may not transfer to another. Recheck accuracy, cost, data handling, accessibility and staff workload when the model or product changes. Maintain a non-AI alternative where appropriate, particularly when access, disability, language or connectivity could otherwise exclude learners.

For institutions serving vocational learners, implementation should be tied to authentic workplace competencies rather than novelty. Prowad’s guide to AI in vocational education and training discusses this curriculum-to-skills connection.

What should institutions measure?

A useful dashboard separates adoption from educational value. The following measures can help, but each should be selected according to the use case rather than collected simply because it is easy to count.

Dimension	Possible measure	Question it answers
Learning	Aligned pre/post assessment; delayed retention; transfer task	Did learners gain and retain the intended skill?
Quality	Sampled accuracy checks; teacher review; error severity	Are outputs reliable enough for this purpose?
Agency	Unaided performance; learner explanation; source verification	Can learners reason without simply repeating AI output?
Equity and access	Participation and outcomes by relevant learner groups; accessibility testing	Who is helped, excluded or disadvantaged?
Workload	Teacher preparation time, review time and support requests	Does the tool reduce work or shift it elsewhere?
Governance	Incidents, disclosure compliance, data-handling review	Are safeguards working in practice?

Institutions should set success thresholds before a pilot begins. If the goal is better formative feedback, for example, a success criterion might require improved revision quality without increased teacher correction time or a meaningful rise in factual errors. The precise threshold should be agreed locally; there is no universal percentage that guarantees an AI tool is educationally effective.

Common implementation mistakes

• Buying before defining the need: a platform can be technically impressive yet poorly matched to curriculum or learner context.
• Training only on prompts: learners and teachers also need verification, ethics, pedagogy and self-regulation skills.
• Measuring clicks instead of learning: frequent use does not prove comprehension, retention or transfer.
• Writing a policy without classroom input: rules that ignore real assignments and constraints are difficult to apply consistently.
• Scaling from testimonials alone: satisfaction is useful evidence, but it cannot replace learning measures and review of unintended effects.
• Assuming one pilot proves universal effectiveness: results depend on age, subject, language, task design, teacher support and access.

Conclusion: institutional readiness is a learning capability

The emerging evidence from Morocco is a useful reminder that people can begin using generative AI long before institutions have agreed on its educational purpose. The appropriate response is neither automatic adoption nor blanket rejection. It is structured experimentation with clear responsibilities, role-specific AI literacy, privacy and integrity safeguards, and evidence of learning.

For education leaders, the next step is concrete: choose one well-defined learning problem, document the current baseline, prepare the people involved, publish rules, and run a bounded pilot with a review date. Continue only when the evidence supports the educational value and the risks can be managed.

AI integration in education is successful when it improves the conditions for learning while preserving learner agency, teacher judgment and equitable access—not merely when the technology is available.

References and further reading

1. Khrou, M. (2026, 14 October). The didactic integration of generative AI tools in secondary and higher education in Morocco. African Scientific Journal, 3(38). DOI: 10.5281/zenodo.23373372.
2. Mellouk, F., & Ouelfatmi, M. (2026, 17 October). The impact of AI on students’ achievement, confidence, and critical thinking in education in Morocco. Discover Education, 5, 1191. DOI: 10.1007/s44217-026-02332-w.
3. European Commission (2026, 18 June). New AI Literacy Framework helps schools prepare learners for the age of artificial intelligence.
4. UNESCO (2024). AI competency framework for teachers.

Evidence note: The Moroccan studies provide contextual, exploratory evidence with regional or small purposive samples. Their findings should not be treated as national prevalence estimates or causal proof. The five-step framework and suggested measures in this article are implementation recommendations, not a validated intervention tested as a package.

AI tools are increasingly presented as a route to more personalised and inclusive learning. For students with disabilities, however, a tool that can generate an explanation or simplify a text is not automatically accessible, safe, or educationally useful. A student may be able to receive an answer but still be unable to navigate the interface, formulate a prompt, interpret the response, or use it independently. Schools therefore need to evaluate not only what an AI system can produce, but whether a learner can meaningfully access, question, and benefit from it.

This question has become more urgent as new research has begun to focus specifically on inclusive AI. On 28 September 2026, UNESCO’s Institute for Information Technologies in Education highlighted a global study on AI, ICT, and inclusion for children with autism, ADHD, and learning disabilities. In a peer-reviewed study published in July 2026, special education teachers described both useful applications and barriers involving accessibility, privacy, bias, and training. A separate 2026 classroom study examined the support educators provided when students with disabilities used ChatGPT in science learning. Taken together, these sources suggest a practical principle: AI inclusion depends on the fit between the learner, the task, the interface, and the human support around the technology.

Why accessibility is more than an AI feature

In ordinary product demonstrations, accessibility may appear as a list of features: speech input, audio output, summarisation, image description, or simplified text. These can be valuable, but a feature is not the same as an accessible learning experience. A voice interface may not work reliably for a learner with a speech disability. A long AI-generated answer may overwhelm a student who needs information broken into smaller steps. A visual explanation may remain inaccessible to a learner who uses a screen reader. A chatbot may be technically usable while producing language that is difficult to interpret or that misrepresents a learner’s communication style.

The UNESCO IITE report draws on a literature review, expert interviews, and case studies from 22 countries. It proposes a system-level framework organised around six domains—People, Products, Pedagogy, Policy, Place, and Provision—and four cross-cutting conditions: Capacity, Co-creation, Consistency, and Conscience. Its central implication is that inclusion cannot be delivered by software alone. Educator preparation, learner participation, reliable provision, institutional policy, and the learning environment all influence whether technology supports meaningful participation.

This matters because the phrase “AI for special education” covers very different tools and needs. A speech-to-text tool, a generative chatbot, an adaptive practice system, an augmentative and alternative communication (AAC) aid, and an AI-powered image description service do not have the same purpose or risk profile. Evaluation should begin with a specific learner goal rather than a general assumption that AI will personalise learning.

What recent research actually tells us

A qualitative study by Marie Sakowicz and Foad Hamidi, published in Universal Access in the Information Society on 28 July 2026, interviewed seven special education teachers in public schools in the eastern United States. Teachers described using AI-enabled technologies for personalised learning and engagement, but also reported accessibility problems—especially for some students with speech and communication disabilities—as well as concerns about data privacy, algorithmic bias, and insufficient professional development. Because the sample was small and qualitative, the study does not estimate how common these experiences are across all schools. Its value is in making implementation barriers visible and identifying questions that broader research should test.

A peer-reviewed study by Kok-Sing Tang and colleagues, first published online on 12 May 2026, analysed educator support during ChatGPT use in an inclusive science classroom. The authors organised observed support into four domains: instructional, literacy, emotional-behavioural, and technological. In practice, this meant helping learners understand the purpose of an AI activity, formulate and interpret questions, remain engaged, and operate the technology—including using audio output or receiving help with typing where needed. The authors describe a situated classroom study, not a universal recipe; the taxonomy is best treated as a useful lens for planning and reflection, not proof that one approach works for every disability or age group.

These studies do not establish that generative AI improves academic outcomes for all learners with disabilities. They instead show why the quality of implementation deserves attention. UNESCO’s report likewise combines evidence and practice examples to identify opportunities and systemic barriers, rather than presenting a single controlled trial that proves the effectiveness of every tool. Schools should preserve this distinction between promising use cases, practitioner experience, and causal evidence of learning gains.

Eight checks before adopting AI for inclusive learning

1. Start with a learner-defined goal

Define the barrier the learner is trying to overcome. Is the goal to access written instructions, practise a concept, communicate an idea, plan a multi-step task, or demonstrate knowledge in a different format? Involve the student, family where appropriate, special educators, classroom teachers, and relevant support professionals. Avoid treating a diagnosis as a complete description of a learner’s preferences or abilities. Two students with the same diagnosis may need different supports, and the same student may prefer different modes for different tasks.

2. Test the complete interaction, not just the feature list

Ask the learner to complete a real classroom task using their usual access method. Can they sign in, navigate controls, enter a question, understand the response, correct an error, and save or share the result? Test keyboard-only navigation, screen readers, magnification, speech input, switch access, captions, and audio playback where relevant. The W3C’s Web Content Accessibility Guidelines (WCAG) 2.2 provide a useful baseline for digital interface accessibility, including keyboard operation, but technical conformance alone does not guarantee that a learning activity is pedagogically accessible.

3. Reduce language and cognitive friction without removing the learning goal

Check whether learners can ask questions in the mode they can use comfortably and whether responses can be adjusted for length, vocabulary, structure, and modality. A useful tool might offer one instruction at a time, a worked example, audio playback, a visual sequence, or a choice between a short and detailed explanation. But simplification should not silently remove essential concepts or lower the intellectual demand of the task. Accessibility means removing unnecessary barriers while preserving meaningful opportunities to learn.

4. Keep educators in the loop

Use AI as a supported resource, not an automatic replacement for teacher judgement or established accommodations. Model how to ask a question, check an answer, request clarification, and recognise when the system has misunderstood. The four support domains identified by Tang and colleagues offer a practical planning prompt: What instructional, literacy, emotional-behavioural, and technological support might this activity require? The answer may change during the lesson, so staff should be able to adapt support in real time.

5. Protect learner agency and provide an alternative

Students should be able to decline a tool that does not work for them and use an appropriate alternative without being penalised. Explain what the AI is doing, allow learners to correct or reject its output, and avoid making a chatbot the sole route to an essential learning resource. Where possible, include learners with disabilities in tool selection and testing before procurement. This aligns with the Universal Design for Learning (UDL) principle of designing multiple ways to engage with information and express learning, rather than expecting every learner to use the same interface.

6. Review privacy, safety, and representation

Before using a system with students, document what data it collects, who can access it, how long it is retained, whether prompts or uploads may be used to improve the service, and how records can be deleted. Do not enter identifiable student information into unapproved services. Review whether the tool produces stereotyped or demeaning descriptions of disability, misreads non-standard language, or treats a communication difference as a lack of understanding. Keep a clear route for students and staff to report harmful outputs and pause use while serious issues are investigated.

7. Measure participation and learning separately

Record more than whether the learner opened the tool or completed an activity. Useful indicators might include whether the learner could begin the task independently, how often assistance was needed, whether they could explain or apply the target concept, whether they felt in control of the interaction, and whether the accommodation reduced an existing barrier. Use a baseline or comparison with the current approach where feasible. Collect feedback from the learner and educator, and avoid treating a small pilot as proof of broad effectiveness.

8. Check who is excluded by the rollout

Consider connectivity, device availability, licensing costs, language support, compatibility with existing assistive technology, and the time required for staff training. If a school adopts a tool that only works well on recent devices or for learners who can comfortably use text-based interfaces, it may widen the very gaps it aims to close. UNESCO’s emphasis on People, Products, Pedagogy, Policy, Place, and Provision is useful here: assess the whole environment, not just the product.

A six-week, low-risk pilot for schools

The following is a proposed implementation sequence, not a validated intervention protocol. It can help a school learn whether a particular AI tool is suitable before wider adoption.

• Week 1 — Define the goal: Select one learning barrier and one specific task. Agree on the learner’s preferred access methods and what success would look like.
• Week 2 — Co-test accessibility: Have the learner and staff test the actual workflow, including assistive technologies, alternative input methods, and failure cases.
• Weeks 3–4 — Run a limited pilot: Use the tool for a defined set of activities. Record support required, technical failures, learner feedback, and any safety or privacy concerns.
• Week 5 — Evaluate independently: Check whether the learner can demonstrate the intended knowledge or skill, with appropriate accommodations but without relying on the AI to supply the answer.
• Week 6 — Decide and document: Continue, modify, or stop the pilot based on predefined criteria. Record which learners benefited, what remained inaccessible, and what additional training or procurement conditions are needed.

Set stop conditions in advance. A privacy incident, repeated inaccessible interaction, harmful or discriminatory output, or a pattern of replacing rather than supporting the learner’s own participation should trigger review. If the pilot is small, describe its findings as local feasibility evidence. Do not present them as proof of effectiveness across disabilities, subjects, or school systems.

The strategic shift: from AI availability to inclusive capability

The emerging research points away from a narrow question—“Does this platform have AI accessibility features?”—toward a more demanding one: “Can this learner use this system to participate in this learning activity, with appropriate agency, support, and protection?” That shift makes accessibility a shared responsibility across procurement, pedagogy, professional development, governance, and product design.

For school leaders, the practical priority is not to buy the most sophisticated tool first. It is to identify the learner’s goal, co-test the real interaction, preserve alternatives, train staff, and measure meaningful participation and learning. AI may help remove barriers in some contexts; in others, it may introduce new ones. An inclusive strategy makes room for both possibilities and lets evidence—not marketing claims—guide the decision.

Sources and further reading

• Agius, M. (2026 release page; report text dated 2025). Report on the Use of AI and ICT to Support the Inclusion of Children with Autism, ADHD and Learning Disabilities. UNESCO Institute for Information Technologies in Education. The report draws on literature review, expert interviews, and case studies from 22 countries.
• Sakowicz, M., & Hamidi, F. (2026). Perspectives of special education teachers on AI-enabled technologies: accessibility, inclusion, and professional development needs. Universal Access in the Information Society. Published 28 July 2026.
• Tang, K.-S., Rappa, N. A., Nonis, K., & Ilyas, K. (2026). Supporting Students With Disabilities to Use Generative AI: A Classroom-Based Taxonomy of Educator Scaffolding Practices. Journal of Special Education Technology. First published online 12 May 2026.
• CAST. (2024). Universal Design for Learning Guidelines, Version 3.0. A research-informed framework for designing learning environments that reduce barriers and support learner agency.
• World Wide Web Consortium (W3C). (2024). Web Content Accessibility Guidelines (WCAG) 2.2. Technical guidance for making web content more accessible; it should complement, not replace, learner-centred pedagogical evaluation.

Evidence limitations: The cited studies include a global practice-oriented report and small qualitative research projects. They help identify design and implementation questions but do not establish that all AI tools improve learning for all students with disabilities. The six-week pilot above is a proposed local evaluation method, not a validated programme.

Schools are being asked to make decisions about AI learning tools while the evidence is still developing. A platform may produce fluent explanations, adapt questions to a learner, or keep students engaged; none of those features, by itself, proves that students learn more or retain knowledge longer. The practical question for school leaders is not simply whether an AI tool works, but what it improves, for whom, compared with what, and under which conditions.

Two European Commission reviews published in September 2026 sharpen this question. One synthesises causal evidence on intelligent tutoring systems in school education; the other reviews research on generative AI among higher-education students. Separately, an October 2026 expert-consensus project from researchers at the University of Minnesota, the University of North Carolina at Chapel Hill, and the University of California, Irvine maps where generative AI may support or disrupt 65 learning processes. Together, these sources point toward a more disciplined way to evaluate AI learning tools: judge educational outcomes, not product demonstrations.

The evidence is becoming more specific—but not interchangeable

The European Commission’s systematic review of intelligent tutoring systems (ITS) in primary and secondary education concludes that these systems can contribute meaningfully to learning and skill development. The evidence for learning outcomes is more extensive and consistently positive than evidence for motivation, engagement, well-being, or socio-emotional outcomes. Findings in those latter areas are more mixed or still indicative rather than conclusive. The review also stresses that value depends on whether learner information is translated into appropriate instructional support and how the system is embedded in teaching.

That is a useful signal, but it should not be generalised to every product marketed as “AI for education.” An intelligent tutoring system with a defined subject domain, sequenced practice, and feedback is not the same intervention as an open-ended chatbot that writes essays, summarises readings, or answers any question. Different systems use different mechanisms, and they should be evaluated against different claims.

The European Commission’s separate review of generative AI in higher education, based on international literature published between January 2022 and April 2026, examines student engagement, academic performance, learning outcomes, and academic integrity while highlighting limitations in the evidence base. Meanwhile, the October 2026 expert-consensus report asked 30 learning scientists and educational researchers to assess 65 learning processes; consensus was reached on 42. The researchers suggest that generative AI may be most useful during practice and consolidation, while unreflective use during initial learning or independent application can encourage cognitive offloading. This is expert synthesis, not proof that every learner or AI system will respond in the same way.

For school decision-makers, the lesson is simple: avoid both extremes. The evidence does not justify assuming that AI automatically improves learning, nor does it justify treating all educational AI as one uniform intervention. The task is to match the tool to a learning problem and test the claim carefully.

Seven evidence checks before adopting an AI learning tool

1. Define the learning problem before evaluating the product

Start with a specific educational need: for example, learners are struggling to retrieve key vocabulary, identify errors in algebra, practise a second language, or receive timely feedback on formative exercises. “We need AI” is not a learning objective. Write down the intended change in learner performance and the group expected to benefit.

This step prevents a common procurement error: selecting a tool because its features look impressive and then searching for a problem that justifies the purchase. If the actual need is more frequent low-stakes practice, a simpler non-AI intervention may be cheaper, easier to evaluate, and just as effective.

2. Identify what kind of system is being evaluated

Describe the intervention precisely. Is it a domain-specific intelligent tutor, an adaptive practice engine, a generative chatbot, an automated feedback system, or a learning analytics dashboard? Does it give hints, reveal full answers, diagnose misconceptions, or merely generate content? Does the teacher control the sequence, or does the system make decisions autonomously?

These differences matter because evidence for one design cannot simply be transferred to another. A product should not be allowed to borrow the credibility of research on structured tutoring if its own design is a general-purpose chatbot with a different instructional model.

3. Ask for evidence on learning—not just usage

Request studies that measure outcomes aligned with the product’s claim. Useful outcomes may include accuracy on curriculum-aligned tasks, delayed retention, ability to solve unfamiliar problems, quality of explanations, or transfer to new contexts. Login counts, time spent in the platform, student satisfaction, and the speed of producing an assignment are implementation indicators; they are not substitutes for learning outcomes.

Give greater weight to well-designed comparisons with an appropriate control or alternative practice. Randomised studies can help estimate causal effects when implemented well, but a small, short trial in one subject or institution may not predict results across age groups, languages, curricula, or school systems. Observational studies and user surveys can still be informative, but they answer different questions.

4. Test unaided performance and retention

A learner may complete a task more successfully with AI assistance without being able to perform it independently afterwards. A credible pilot should therefore separate supported performance from learning that persists when the tool is removed. Include an immediate assessment without AI and, where feasible, a delayed check days or weeks later. Add a transfer task that requires learners to apply the idea to a new example.

This is particularly important for generative tools that can produce polished answers. The school should ask whether the learner has developed the underlying knowledge and reasoning—not only whether the final product looks better. Where appropriate, ask students to explain their decisions, identify an AI error, or solve a parallel problem independently.

5. Evaluate the instructional design, not just the model

Educational outcomes depend on how the technology is used. Does the tool encourage retrieval practice, give graduated hints, ask learners to explain their reasoning, and correct misconceptions? Or does it quickly reveal complete solutions? Can teachers configure it to fit the curriculum and inspect its feedback? Is AI use concentrated in a learning phase where assistance is helpful, while students still practise independent thinking and application?

The October 2026 expert-consensus report is especially relevant here: its authors distinguish acquiring information, practising to consolidate it, and applying knowledge independently. They argue that careless assistance can interfere with some of the mental work learning requires, while structured practice may offer a more promising use. Because this is consensus-based guidance rather than a trial of every product, schools should treat it as a design hypothesis to test—not a universal law.

6. Check who benefits, who may be disadvantaged, and what data are collected

Average results can hide important differences. Analyse outcomes, where sample sizes permit, by prior attainment, language background, disability or accessibility needs, and access to devices or reliable connectivity. Ask whether the system supports multilingual learners and assistive technologies, and whether its feedback is understandable and appropriate for the age group. Do not infer fairness from a vendor’s general statement that a product is inclusive.

Before a pilot, document what learner data the system collects, where they are stored, how long they are retained, whether they are used to train models, who can access them, and how deletion works. Avoid entering identifiable student information into tools that have not been approved under the school’s data-protection and safeguarding requirements. A tool that improves a narrow outcome but introduces unacceptable privacy or safety risks may not be suitable for adoption.

7. Keep teachers responsible for interpretation and decisions

AI-generated feedback, predictions, and recommendations should be treated as inputs for professional judgement, not as unquestionable decisions. Teachers need to know the system’s limits, inspect questionable outputs, correct errors, and decide when a human explanation or relationship is more appropriate. Students should also understand when AI is being used, what it can get wrong, and how to question its answers.

Schools should measure teacher workload as well as student outcomes. A tool may save time on one task but create extra work through checking, troubleshooting, or adapting generic output. The relevant comparison is the total effect on teaching and learning—not a vendor’s demonstration of how quickly a single task can be completed.

A practical pilot schools can run in eight weeks

The following is a proposed implementation model, not a protocol validated by the cited reports. It gives schools a manageable way to generate local evidence before committing to a wider rollout.

• Weeks 1–2 — Define and baseline: Choose one subject, one learning problem, and one primary outcome. Record baseline performance and agree on privacy, safeguarding, accessibility, and teacher-oversight requirements.
• Weeks 3–4 — Design the comparison: Specify exactly how the AI will be used and what the comparison group or existing practice will be. Train participating teachers and prepare equivalent assessment tasks.
• Weeks 5–6 — Run the pilot: Record implementation, teacher workload, student participation, technical problems, and any unintended effects. Do not change the intervention repeatedly without documenting the changes.
• Week 7 — Assess independently: Use an assessment completed without AI assistance. Where possible, include a delayed retention or transfer task and examine differences among learner groups.
• Week 8 — Decide and report: Compare results with the baseline and comparison condition, document uncertainty and missing data, and decide whether to stop, redesign, extend the pilot, or scale cautiously.

Before the pilot begins, set decision rules. For example, define what level of improvement would be educationally meaningful, what workload increase is acceptable, and which safety or equity problems would halt the trial. This reduces the risk of interpreting any positive result as success after the fact. If the pilot is too small to detect reliable differences, report it as a feasibility test rather than evidence of effectiveness.

What school leaders should conclude from the latest evidence

The emerging research supports a more precise approach to AI adoption. Structured intelligent tutoring systems have a promising evidence base for some learning and skill outcomes, but results vary by outcome and context. Generative AI creates a different set of opportunities and risks, and current evidence is not strong enough to support sweeping claims about its long-term effects on learning. Expert consensus can help identify mechanisms and sensible design principles, but it does not replace controlled evaluation.

The strongest procurement question is therefore not “Which AI platform is best?” It is: “Which intervention is most likely to solve this learning problem, what credible evidence supports that claim, and how will we know whether our own students have actually learned?” Schools that ask those questions can move beyond technology adoption as a goal and treat AI as a testable part of instructional design.

Sources and further reading

• Theis, D. M., Thomsen, M., Vogel, C., & Wasserfuhr, V. (2026). The impact of intelligent tutoring systems in general education: A systematic review of causal evidence on learning outcomes, motivation, and socio-emotional effects. European Commission, ENESET. Published 23 September 2026.
• Pinto, M., & Ferreira, S. (2026). Impact of generative AI on higher education students. European Commission, ENESET. Review of international literature from January 2022 to April 2026; published 25 September 2026.
• Kendeou, P., Greene, J. A., Nixon, N., et al. (2026). Ways Generative AI Can Support and Threaten Learning in K–20 U.S. Education. Expert Consensus Report, University of Minnesota College of Education and Human Development, with collaborating researchers. Published October 2026.
• European Commission. (2026). New reports examine how AI and digital technologies are shaping education in Europe. European Education Area, 22 September 2026.

Evidence note: This article synthesises recent reviews and expert consensus to propose an evaluation process. It does not claim that the proposed eight-week pilot has itself been experimentally validated, and it does not assume that findings for one type of AI system generalise to all educational tools.

Vocational education and training (VET) must respond to changing occupational standards, digitalisation, the green transition and the growing use of artificial intelligence. Yet updating curricula and qualifications can be slow, resource-intensive work involving employers, teachers, public authorities and social partners. AI may help teams analyse evidence and prepare revisions more efficiently—but it cannot decide alone what a qualification should mean or what competent performance looks like.

The OECD’s June 2026 report, Developing Vocational Education and Training with Artificial Intelligence, examines current uses and pilots across 25 countries and case studies from ten countries. Its central message is balanced: AI can support the development of curricula, qualifications and occupational standards, but its value depends on human oversight, reliable data, quality assurance and collaborative governance.

What the OECD review tells us

The report finds that AI use in VET development remains uneven and exploratory. Public VET agencies and industry bodies appear more active than many individual providers and teachers, and most current applications support established processes rather than replace them. Examples include labour-market analysis, skills anticipation, competency mapping, synthesis of consultation responses, drafting learning outcomes and checking consistency with qualification frameworks.

This distinction matters. The report is about developing and updating VET systems—not proof that every AI tutor improves the learning of every apprentice. Curriculum development, training delivery and assessment are related but different activities. Each requires its own evidence and safeguards.

Five useful applications of AI in vocational education and training

1. Detecting emerging skill needs

Labour-market information is abundant but fragmented. AI-assisted analysis can help teams compare job vacancies, skills forecasts, occupational standards and qualification registers to identify potential changes in demand. This can help experts decide which skills deserve further investigation before a curriculum review begins.

Good practice: treat automated signals as prompts for investigation, not as a direct instruction to rewrite a programme. Vacancy data can overrepresent some employers, occupations or regions and may not reveal the full range of skills required for safe, high-quality work.

2. Mapping competencies across standards

AI can compare learning outcomes, occupational standards and qualifications to highlight overlapping, missing or inconsistently described competencies. This is especially useful where multiple frameworks describe similar skills in different language or where programmes are being modularised.

Good practice: ask subject experts to verify every proposed match. A phrase that looks similar in two documents may represent a different level of autonomy, risk or practical responsibility.

3. Drafting and revising curriculum material

Generative AI can prepare a first draft of learning outcomes, explanatory notes, assessment criteria or alternative wording. It may reduce repetitive drafting work and allow curriculum specialists to spend more time discussing priorities and validating content.

Good practice: use AI for an editable first draft, never as the final authority. Each outcome should remain observable, assessable and aligned with the actual occupational standard. Remove vague statements that sound impressive but cannot be assessed in practice.

4. Synthesising stakeholder consultation

VET development depends on input from employers, teachers, learners, regulators and other social partners. AI may help organise consultation responses, group recurring themes and prepare evidence briefs, potentially making a large volume of feedback easier to review.

Good practice: preserve minority and conflicting views rather than flattening them into a false consensus. Keep a traceable record of how stakeholder input was interpreted, and allow contributors to correct summaries that misrepresent their position.

5. Checking consistency and compliance

AI-assisted tools can flag inconsistent terminology, missing fields, duplicated outcomes or possible misalignment with a qualification framework. These checks can help quality-assurance teams identify documents that need closer review.

Good practice: distinguish mechanical checks from substantive judgement. A document can be internally consistent and still describe the wrong skills, omit an important safety procedure or fail to reflect current workplace practice.

Examples cited in the OECD report

The OECD describes several country and sector examples. England’s SkillsCompass uses natural-language processing and generative AI to analyse occupational standards, labour-market data and foresight trends. The Netherlands’ SBB-Chat supports qualification drafting and related processes. Switzerland’s ICT sector has developed occupation-specific custom GPT models for modular curriculum design and other VET tasks.

These examples show possible directions, not a ranking of products or proof that the same approach will work everywhere. The report notes that fully integrated, system-wide solutions remain rare. Countries with stronger data infrastructures and skills-intelligence systems are better positioned to move beyond small pilots.

Four risks that VET leaders should manage

• Unreliable or biased data: incomplete labour-market records and unevenly represented occupations can lead to misleading skill forecasts.
• Loss of human accountability: responsibility becomes unclear if institutions accept AI-generated standards or curriculum changes without named reviewers and documented approval.
• Weaker stakeholder participation: a polished AI summary should not replace discussion among employers, teachers, learners and public authorities, especially when interests conflict.
• Privacy and security exposure: unpublished qualification material, employer data and sensitive consultation responses may require restricted access, retention limits and approved tools.

These are governance problems as much as technical ones. Providers also need staff with sufficient AI, digital and data literacy, secure tools suited to the task, and guidance that reflects the realities of vocational education. Smaller institutions may face particular constraints in expertise, infrastructure and funding.

A practical 90-day roadmap for responsible adoption

The following roadmap is a proposed implementation approach informed by the OECD’s findings; it is not a timetable prescribed by the report.

Days 1–30: Choose one bounded problem

Select a specific task, such as comparing a set of occupational standards or synthesising consultation feedback. Establish a baseline: how long does the current process take, what errors recur, and who must approve the result? Map the data involved and identify privacy, security and legal requirements before staff upload any documents.

Days 31–60: Run a supervised pilot

Use an approved tool with a small team of curriculum specialists and subject experts. Require reviewers to check factual claims, source coverage, missing perspectives and alignment with the relevant standard. Record corrections and the time spent reviewing AI output as well as the time saved during drafting.

Days 61–90: Evaluate and decide

Compare the pilot with the baseline. Possible measures include review time, factual error rate, missing competency flags, consistency across documents, stakeholder satisfaction and the proportion of AI suggestions accepted after expert review. Do not use speed alone as the success criterion. If the tool saves time but introduces errors that are difficult to detect, the apparent gain may be illusory.

At the end of the pilot, document what the tool may be used for, what it must never decide independently, who signs off on outputs, how errors are reported and when the experiment should stop. Scale only when the evidence supports it and the institution can maintain the safeguards.

How to connect curriculum reform with training outcomes

Updating a curriculum is only one part of the system. Institutions also need to know whether training helps learners acquire and retain the intended competencies. That requires separate evaluation of teaching, practice, assessment and transfer to real tasks. Our related article on AI training at work and a 2,574-employee field experiment discusses why learning outcomes and later application should be measured rather than assuming that a technology feature guarantees success.

Providers evaluating digital tools can also use Prowad’s evidence checklist for AI learning tools as a starting point, adapting it to occupational standards, practical assessment and workplace risk.

Conclusion

AI offers VET systems useful support for labour-market analysis, competency mapping, drafting, consultation and quality checks. The OECD’s 2026 review shows that these applications are emerging across countries, but adoption remains uneven and integrated solutions are still uncommon. The responsible approach is not to automate curriculum decisions wholesale. It is to use AI for bounded analytical and drafting tasks, keep expert judgement and stakeholder participation central, protect sensitive data, and evaluate whether the resulting qualifications remain accurate, relevant and trusted.

Reference

OECD (2026). Developing Vocational Education and Training with Artificial Intelligence. OECD Reviews of Vocational Education and Training. OECD Publishing, Paris. Published 23 June 2026.