This is not a single experiment but a huge collection of descriptive data. Here are three findings we think are worth a closer look.

In short: teacher shortages show up more in under-qualified staff than in empty posts; falling pupil numbers will not automatically fix them; and a new pattern in graduate unemployment is worth watching, but it is far too early to blame AI.

1. Shortages show up in qualifications, not just vacancies

When we picture a teacher shortage, we tend to imagine unfilled posts. In most countries with data, unfilled vacancies are actually below 3% of positions, and often below 1%. Latvia and Luxembourg are exceptions, at around 4% in secondary education.

More often, shortages are hidden in who fills the posts. On average across OECD countries, 9.1% of primary and secondary teachers were not fully qualified in 2024/25, as schools recruited people without standard teaching qualifications to cover gaps. Long absences add pressure too: in Austria, 11% of teachers were off for more than two months during the year.

Shortages are also uneven. Maths and natural sciences are hit hardest almost everywhere, while arts, social studies and PE are less affected. In some countries the problem is concentrated in remote rural areas; in others, in individual schools, often in disadvantaged communities.

2. Falling pupil numbers will not automatically solve the problem

The number of children aged 5–14 across the OECD is projected to fall by 9% between 2024 and 2033. It is tempting to assume that fewer pupils means fewer teachers needed. The OECD is sceptical: over the past decade, many countries kept or even increased teacher and class numbers as rolls fell.

At the same time, 25% of secondary teachers are aged 55 or older and close to retirement. Without careful planning, the report warns, retirements and shrinking rolls could deepen shortages in some schools and regions while leaving others with classes too small to be financially sustainable.

3. A new pattern in graduate unemployment, but AI is not proven to be the cause

Between 2023 and 2025, unemployment among tertiary-educated 25–34-year-olds rose by 0.4 percentage points on average across the OECD, while among young adults with upper-secondary qualifications it fell by 0.2 points. The gap between the two narrowed from 2.4 to 1.8 points.

According to the OECD, this is the first period since its data series began in 2000 in which unemployment rose for young graduates while falling for those with upper-secondary qualifications. Some have linked this to AI reducing demand for graduate skills. The report is careful here: degrees still bring substantial labour-market benefits, and evidence for a lasting AI-driven shift in demand for skills remains limited. Its advice is to monitor the trend closely, not to jump to conclusions.

What else the report says about teachers

• Job satisfaction is high on average. Around 85% of teachers are satisfied with their jobs, similar to other professions, but this varies hugely between countries.
• Career changers can help. In Australia and the Netherlands, more than 15% of lower-secondary teachers came to teaching from another career, supported by flexible, employment-based training routes. Many countries have no structured route at all.
• Teaching competes with the whole economy. Vacancy rates in the wider economy are higher than in teaching in more than half the countries with data. While skills shortages persist elsewhere, schools will keep competing for a limited pool of workers.

Why it matters

Reports like this help separate problems within education from the conditions around it. Teacher shortages, demographic change, labour-market pressures and funding all shape what schools can realistically deliver, whatever happens in the classroom.

It is also a useful reminder that a trend and its explanation are two different things. A change in graduate unemployment does not, on its own, show that AI is responsible.

A note of caution

International comparisons need care. Countries differ in how their education systems and labour markets work, how they define terms such as “fully qualified”, and what data they collect. OECD averages can hide very large differences between countries, and the figures for any one country, including the UK, may look quite different.

The report is also mainly descriptive. It is excellent for spotting patterns and raising questions, but the statistics alone cannot tell us what causes those patterns.

Source

OECD (2026). Education at a Glance 2026: OECD Indicators. Paris: OECD Publishing. Published 29 September 2026. The report and supporting materials are free to read.

The post Education at a Glance 2026: teacher shortages, falling rolls and a puzzle about graduates appeared first on My Tutor Hour.

In short: students who learned about organic solar cells with an interactive AR app gained more conceptual understanding than students taught conventionally, and held fewer misconceptions. The effect reported is unusually large, so it needs confirming by other researchers.

What the researchers looked at

Organic solar cells rely on molecular structures and physical processes, such as how charges are generated and move through the material, that are abstract and difficult to visualise. Students often come away with misconceptions.

The team, Muhammad Naeem Sarwar, Abdullah Muzil Alharbi and Muhammad Faizan Nazar, first consulted chemistry educators about where students struggle. They then built an immersive AR app with:

• interactive 3D visualisations;
• explanations at both the molecular level and the level of the whole device; and
• simulations of key processes, including how the cell operates and how charge is transported.

To test it, they compared students who learned with the app against students who received conventional instruction, using tests before and after. They then interviewed students about their experience. Based on the reported statistics, roughly 120 students took part, though the paper’s summary does not give the exact number per group.

What they found

• Greater understanding. After accounting for differences in pre-test scores, the AR group improved significantly more than the conventionally taught group, particularly on the structure of the cells, how they work, molecular interactions and charge transport.
• Fewer misconceptions. The authors report that misconceptions were markedly reduced in the AR group.
• Positive feedback. In interviews, students said the visualisations made the processes more accessible and engaging.

Why it matters

The study illustrates a well-established idea in educational psychology: how something is represented can affect how easily it is understood. When a topic involves processes that are invisible, abstract or happen at a scale we cannot see, a well-designed visualisation may help learners build a more coherent mental model.

But adding technology does not automatically improve learning. The useful question is not “Does AR work?” but “Does this particular representation help students grasp this particular concept?” A clear animation or a good physical model might sometimes do the same job at far lower cost.

A note of caution

The study was quasi-experimental rather than a randomised trial, so pre-existing differences between the groups, or differences in how they were taught, may have contributed to the results. Novelty can also boost engagement in the short term.

The effect size was exceptionally large. The researchers report that the teaching method accounted for over 80% of the variation in post-test scores once prior scores were taken into account (partial η² = 0.82). Effects that big are rare in education research. That does not mean the result is wrong, but it does make independent replication especially important before treating AR as a reliable route to similar gains.

Finally, this was one app, one specialised chemistry topic and one group of students. The findings cannot simply be extended to other subjects, age groups or kinds of AR. The article is also an early version that may be edited before final publication.

Source

Sarwar, M. N., Alharbi, A. M. & Nazar, M. F. (2026). Fostering conceptual understanding of organic solar cells through immersive augmented reality: a quasi-experimental study in chemistry education. Scientific Reports. Published 8 October 2026. Open access.

The post Seeing the invisible: can augmented reality help students understand solar cells? appeared first on My Tutor Hour.

In short: among 390 university students in Bangladesh, those who found AI tools useful, and those who used them more often, reported better mental health and less cognitive overload. The study is a one-off survey, so it shows associations, not cause and effect.

What the researchers looked at

The team, led by Mohammad Nurul Alam, surveyed 390 undergraduate and postgraduate students at public and private universities in Bangladesh. They asked about:

• how useful students thought AI tools were for their learning;
• how often they used AI for academic work;
• their sense of cognitive overload, the feeling of having more information and demands than they can process;
• their mental health; and
• their digital self-efficacy, meaning their confidence in using digital tools well.

Using a statistical technique called structural equation modelling, they tested how these factors related to one another, and whether cognitive overload helped explain any link between AI use and mental health.

What they found

• Usefulness and mental health. Students who saw AI tools as useful reported better mental health.
• Frequency and mental health. Students who used AI more often for their studies also reported better mental health.
• The role of overload. Both perceived usefulness and frequency of use were linked to lower cognitive overload, and higher overload was linked to poorer mental health. Overload partly explained the connection between AI use and well-being.
• Confidence matters, sometimes. Digital self-efficacy strengthened the link between finding AI useful and mental health, but did not significantly change the link between how often students used AI and their mental health.

Why it matters

The findings fit with the idea that AI tools can help students organise information, work through difficult material and manage their workload, and that reducing unnecessary mental burden may support well-being.

But it is worth being clear about what the study did not measure. Despite the paper’s title, it did not assess learning or achievement. Feeling less overwhelmed and learning more are related but different things. A student might feel calmer while learning no more effectively. Equally, a tool could make tasks feel easier precisely because it is doing some of the thinking that would otherwise build understanding.

For universities, the study suggests that how students experience AI tools, and how confident they feel using them, deserves attention alongside questions of academic integrity and learning quality.

A note of caution

This was a cross-sectional survey: students were asked about their experiences at a single point in time. That means it cannot show that AI use caused better mental health or lower overload. The relationship could easily run the other way, or be driven by something else. Students who are already confident, less stressed or better supported may simply be more likely to use AI tools, and to use them well.

All the measures were self-reported, and all the students came from one country. The results may not apply to students with different educational experiences or different access to technology. The article is also an early version that may be edited before final publication.

Source

Alam, M. N., Alharbi, T. F., Islam, M. A., Amin, M. B., Mustafa, Z., Ali, Z., Zygiaris, S. & Azad, M. A. K. (2026). Integrating artificial intelligence into higher education balances student learning benefits and cognitive overload. Scientific Reports. Published 8 October 2026. Open access.

The post Does AI lighten the load? University students, cognitive overload and well-being appeared first on My Tutor Hour.

In short: during group work, students switched between being on task, off task and idle roughly every 10–30 seconds, and the total time they spent on task predicted how well they did on a later test.

What the researchers looked at

The team, including Ido Davidesco, David Poeppel and Suzanne Dikker, worked with nine groups of four high-school students. Each group worked together to build a model of a cell.

The researchers used several kinds of evidence at once:

• Video of each session, which was transcribed and coded moment by moment.
• EEG (electroencephalography), which recorded each student’s brain activity during the task.
• A test to measure what students had learned.

Combining what students were visibly doing with what their brains were doing allowed the team to track how engagement changed over the course of the activity, rather than giving each student a single “engaged” score.

What they found

• Engagement was constantly shifting. On average, students moved between on-task, off-task and idle states every 10 to 30 seconds.
• Time on task mattered. The total time a student spent on task significantly predicted their test performance.
• “Idle” looked like “off task” in the brain. During idle moments, when a student was neither clearly working nor clearly distracted, activity in the theta (4–7 Hz) and alpha (8–12 Hz) frequency bands resembled off-task periods and differed from on-task periods.

That last point is interesting. A quiet student who seems to be listening may look engaged from across the room. The EEG data hint that, at least in this study, those idle moments were closer to switching off than to thinking hard.

Why it matters

For teachers, the study is a reminder that a group can look productive overall while individual students drift in and out of meaningful participation. A busy table is not the same as four engaged learners.

It also highlights a measurement problem. Observation alone may miss some of what is happening cognitively, and brain data alone cannot tell us what a student is actually thinking about. The authors argue that combining the two gives a more informative picture than either on its own.

Some practical questions follow for anyone who uses group work:

• Are roles and tasks structured so that every student has something to do, not just the most confident one?
• How do you check individual understanding, rather than judging the group’s product?
• Would shorter, more clearly defined steps help students stay with the task?

A note of caution

This is a small study: nine groups, or 36 students in total. That is enough to reveal an interesting pattern, but not enough to draw broad conclusions about classroom learning.

The task was also very specific. Building a cell model together may not resemble a maths lesson, independent reading or a university seminar, so we should not assume the same rhythms of engagement apply everywhere.

Finally, the link between time on task and test scores is a relationship, not proof of cause. The study does not show that simply pushing students to spend more time on task will improve learning in every setting. The article is also an early version that may be edited before final publication.

Source

Davidesco, I., Liu, Y., Chaloner, K., Laurent, E., Ali, G. A., Creider, S. C., Hughes, S., Noejovich, L., Valk, H., Bevilacqua, D., Poeppel, D. & Dikker, S. (2026). The dynamics of student engagement in cooperative science learning: a multimodal approach. npj Science of Learning. Published 30 September 2026. Open access.

The post On, off, idle: how student engagement flickers during group work appeared first on My Tutor Hour.

In short: loading delays of around one to two seconds were linked to less engagement and lower achievement, across two very different groups of online learners.

What the researchers looked at

The team, led by Asaf Mazar and including Angela Duckworth, analysed two large datasets:

• 26,707 primary and secondary school students in the United States completing online maths assignments.
• 55,184 aspiring coders, mostly in Latin America, using an online learning platform.

Rather than simply comparing learners with fast connections to learners with slow ones, who might differ in many other ways, the researchers used two longitudinal, quasi-experimental field studies. They looked at how the same learner behaved and performed when their pages happened to load more slowly than usual. That within-person comparison helps rule out explanations such as household income or prior attainment.

What they found

Even short “microdelays” of about one to two seconds were associated with worse outcomes. When school students experienced longer loading times, they:

• completed fewer maths assignments;
• took longer to finish the ones they did;
• were more likely to get distracted mid-task; and
• scored lower.

The coders showed a similar pattern. With more delay, they spent less time actively engaged, multitasked more, completed fewer lessons and were less likely to reach a key course milestone.

Why it matters

It is tempting to assume that if the learning material is good, small technical hiccups do not matter much. This study suggests they may matter a great deal. Learning depends on sustained attention, and every pause is an invitation for the mind, or the phone, to wander. A few seconds lost here and there can add up to fewer tasks attempted and less practice completed.

There is also an equity issue. Learners on slow connections or older devices meet more of these delays, which could put them at a disadvantage that has nothing to do with their ability or motivation.

For schools, tutors and EdTech designers, some practical implications follow:

• Treat page speed as part of learning design, not just an IT concern.
• Test platforms on the slower devices and connections that students actually use at home.
• Where possible, choose tools that pre-load the next question or work well offline.

A note of caution

These were quasi-experimental field studies, not randomised trials in which students were deliberately assigned different internet speeds. The within-person design strengthens the case that delays play a causal role, and the authors report that the results held across several analyses. Even so, the findings are not proof that each extra second of loading costs a fixed amount of learning. The two settings, US school maths and adult coding courses, also differ, so we should be careful about assuming the effects carry over unchanged to every classroom or platform.

Source

Mazar, A., Tomaino, G., Siedahmed, A., Abdolsaheb, A., Heffernan, N. T., Carmon, Z., Duckworth, A. L. et al. (2026). Microdelays disrupt online learning. Nature Human Behaviour. Published 8 October 2026. The abstract is free to read; the full text may require a subscription. The research data and code are available through OSF.

The post Waiting a second: how tiny loading delays undermine online learning appeared first on My Tutor Hour.