The Reading Crisis Began Before COVID — and It Didn't End When Schools Reopened
Reading has been weakening across most PISA countries since around 2012. The pandemic made it worse without having started it, and the newest results show no recovery after schools reopened. What we can say about the cause is much thinner than what we can say about the problem.
Reading difficulty is usually filed under early primary school. Someone teaches decoding, someone else teaches fluency, and by the time students reach a secondary English classroom the assumption is that reading has been handled. Our job, on this view, is to choose what they read and teach them what to do with it.
That assumption is now harder to defend, and the reason is not that classrooms feel different. Professional impressions are worth having, but they are not evidence, and they are vulnerable to nostalgia. The reason is that two international assessments, covering different ages and asking somewhat different things of readers, are now pointing the same way. And the more detailed of the two has gone unusually far in describing what, specifically, students have become worse at.
The 2025 PISA results, published on 8 September 2026, are the first full international picture of what happened after normal schooling resumed. They are not a recovery.
Part IThe pattern
The timeline does not fit the pandemic story
The tidy narrative would be: schools closed, scores fell, schools reopened, scores recover. Each part of that is testable, and the first three hold up better than the fourth.
PISA tests 15-year-olds every three years on a scale built to be comparable across cycles. The 2025 resultsPISA 2025 Results (Volume I): Future-Ready StudentsPrimary source for the OECD reading path: down 11 points from 2018 to 2022 and a further 14 to 2025, with declines across most of the achievement range and larger item-level falls on tasks tied to longer texts than on those tied to shorter ones.View the report → describe the average OECD path as an accelerating decline: small downward movements starting around 2012, a large drop between 2018 and 2022, and another large drop between 2022 and 2025. Across the 35 OECD countries with comparable data, reading fell 11 points to 2022, then a further 14 points to 2025.
OECD reading trajectory
Reading did not rebound after schools reopened
The OECD trend average for reading, across the 35 countries with comparable data on the PISA scale. It fell between 2018 and 2022, then fell again between 2022 and 2025.
Table
| Year | Reading score |
|---|---|
| 2018 | 488 |
| 2022 | 477 |
| 2025 | 463 |
Twenty-five points is a big movement on this scale. The OECD notes that around this age a year of schooling tends to be worth something in the region of twenty points, while warning against treating that as a literal conversion. Take it as a sense of proportion rather than an age equivalence. And this is not a handful of weak systems dragging an average down. Of the 74 countries and economies with comparable data, 56 recorded clear declines, and only five improved.
The pandemic did real damage, and that claim rests on better evidence than a trend line. Betthäuser, Bach-Mortensen and Engzell (2023)A systematic review and meta-analysis of the evidence on learning during the COVID-19 pandemicPandemic learning loss was real and opened early: a pooled deficit of about 0.14 of a standard deviation that had not closed by the time of review, and was larger among socioeconomically disadvantaged students.View original study → pooled 42 studies from 15 countries and found a clear average loss that opened early in the pandemic and had not closed by the time they looked. Wisenöcker and colleagues (2025)A meta-analysis of students’ academic learning losses over the course of the COVID-19 pandemicPooled loss of about 0.20 of a standard deviation, still visible more than two years on. By subject the loss was largest in mathematics (about 0.26) and smallest in reading (about 0.12). Used here only to show reading was not the hardest-hit domain, alongside the pre-2020 decline and the further fall to 2025.View original study → later pooled 103 studies from 45 countries and found the same on a larger scale, with the losses still visible more than two years in.
That second review also found that the losses were larger in mathematics than in reading.
That comparison matters, because it means reading was not uniquely or disproportionately damaged by the pandemic. Set it beside the fact that reading had already been weakening before 2020, and then fell again between 2022 and 2025, and COVID looks like an important part of the story rather than a sufficient explanation for the whole trajectory.
The PISA report makes a related point from another direction. Among 29 systems with data, school closures during the first pandemic peak ranged from none at all in Sweden to around twenty weeks in Costa Rica. Sweden, which never closed, went on to lose about 21 reading points. Costa Rica’s reading barely moved. How long a country shut its schools does not predict what happened to its readers.
It is not only fifteen-year-olds
If the problem belonged to adolescents, we could reasonably look for adolescent causes. Evidence from younger children makes that harder.
PIRLSPIRLS 2021 International Results in ReadingOf 32 countries with comparable data, 21 scored lower in 2021 than in 2016; eight were unchanged, three improved. The IEA states these declines likely reflect pandemic conditions and shows the trend lines as dotted, so PIRLS evidences weaker younger readers but not a pre-2020 fall.View the results → tests reading at the end of the fourth year of school, around age ten, and it is not a junior version of PISA. It asks ten-year-olds to find information, work out what is implied, interpret and evaluate. PISA works with older students and pushes further into weighing unfamiliar and competing sources. They are not the same test, and a fall in one does not confirm a fall in the other.
But they are pointing the same way. In PIRLS 2021, 21 of the 32 countries with comparable data scored lower than in 2016, eight were unchanged and three improved. The OECD then did something more interesting than compare averages: it lined up the cohorts. Children finishing their fourth year in 2016 were the students PISA would test around 2022, and where a country’s ten-year-olds had weakened, its 15-year-olds tended to weaken a few years later. For most systems, in other words, the 2025 reading results were already visible in the same children at age ten, which points back into earlier childhood rather than to anything that happened after 2022.
One caveat gets skipped a lot. PIRLS 2021 was run during the pandemic, and the IEA says plainly that its declines probably reflect that. So PIRLS establishes that younger readers weakened too. It does not establish that they were weakening before 2020. That half of the argument rests on PISA alone.
The bottom is growing, and the top is not holding
The most common way to misread these results is as a story about a struggling minority. The numbers do not cooperate.
PISA sets a threshold it calls Level 2, the point below which a student is unlikely to be able to use written material well enough to manage further study or ordinary working life. Across OECD countries, the share of 15-year-olds below that line in reading rose by ten percentage points between 2015 and 2025. Roughly three in ten are now below it, according to the national report for England.PISA 2025: National Report for EnglandSource here for the OECD-average share of 15-year-olds below Level 2 in reading in 2025, roughly three in ten, and for the OECD trend-average reading series: 490 in 2015, 488 in 2018, 477 in 2022, 463 in 2025.View the report → Over the same decade, the share reaching the top two levels fell by two percentage points.
Look at the whole range rather than the average and the picture sharpens. Between 2018 and 2025 the weakest readers lost around 28 points and the strongest lost around 21. The gap between them widened, but only slightly. What actually happened is that students right across the achievement range moved down together.
Then there is the result I found hardest to fit into anything I already believed. The gap between students from the most and least advantaged homes narrowed over this period, from 88 points to 77. That was not disadvantaged students catching up. Between 2022 and 2025 the least advantaged lost 4 points and the most advantaged lost 20. The OECD’s own reading is that students from comparatively comfortable homes who nonetheless struggle academically are contributing most to the decline.
A narrowing gap, in other words, can sit quite comfortably alongside a system getting worse. Norway is where that distinction bites.
Norway’s problem is bigger than its PISA rank
I teach in Norway, so this section is closer to home than the rest.
Norway scored 453 in reading in 2025, against an OECD average of 461. The rank is the least interesting fact available.
The trajectory is the story. Norwegian reading peaked at 513 in 2015, fell to 499 in 2018, to 477 in 2022 and to 453 in 2025: sixty points in a decade, roughly a quarter of that before the pandemic.
Norway's reading trajectory
Norway's reading decline began before COVID
PISA reading score for Norwegian 15-year-olds. The fall runs the length of the decade, not just the pandemic years.
Table
| Year | Reading score |
|---|---|
| 2015 | 513 |
| 2018 | 499 |
| 2022 | 477 |
| 2025 | 453 |
15%→34%
Share of Norwegian 15-year-olds reading below Level 2, 2015 to 2025. One in three now leaves compulsory school below that line.
According to the Norwegian Directorate for Education and Training,PISA 2025: Kompetanse i naturfag, lesing og matematikkNorway scored 453 in reading in 2025, down 24 points since 2022 and 60 since 2015, with the share below Level 2 rising from 15% to 34%. The OECD flags Norway’s 2025 figures for lower participation and higher exclusion rates.View the report → the share of Norwegian 15-year-olds reading below Level 2 rose from 15% in 2015 to 34% in 2025. One in three now leaves compulsory schooling below that line, and the national research team describes this as the weakest Norwegian reading performance in the history of PISA.
The younger evidence points the same way. In Norway’s PIRLS 2021 results,PIRLS 2021: Markant nedgang i norske tiåringers leseferdigheterNorway fell from 559 in 2016 to 539 in 2021, weak readers roughly doubled to about a fifth of pupils, and the share reaching the high and advanced benchmarks fell from 58% to 48%. Norway reported the lowest reading enjoyment of all 65 participating systems.View the report → ten-year-olds scored 539, down from 559 in 2016. The share of weak readers roughly doubled, the share reaching the higher benchmarks fell from 58% to 48%, and Norwegian children reported the lowest reading enjoyment of all 65 participating systems.
Now hold that against something Norway does genuinely well. Family background predicts achievement less strongly here than in most OECD countries, and Norwegian schools differ less from one another than schools almost anywhere else in PISA. Norway distributes its outcomes unusually evenly.
It is distributing a declining level unusually evenly. Fairness and competence are separate questions, and any account of Norwegian schooling that uses the fairness record to reassure itself about the achievement record has quietly swapped one for the other.
One caveat, since it will be raised. Fewer Norwegian students took part in 2025 than usual and more were excluded, so the OECD marks Norway’s figures as carrying extra uncertainty. The direction of travel across three PISA cycles is too consistent to be a sampling accident, but the individual numbers deserve a lighter grip than the trend does.
Part IIThe cause
So why is reading declining?
Here the evidence thins out considerably, and not evenly.
What we can say with confidence. The first question to ask about a decline this large is whether the test changed rather than the students. The OECD went looking, and what it found makes that unlikely. The falls appear from the opening section of the reading test, not only in the later parts, so this is not students running out of patience halfway through. And because the 2025 test reused reading tasks from 2018 and 2022, the OECD could see exactly which questions students had slipped on.
That comparison produced two patterns, and they are not the same pattern. The first concerns the kind of task a question sets. Between 2018 and 2022 the questions that suffered most asked students to weigh something up: to hold several pieces of information in mind at once and judge them. The questions that held up best asked students to find one specific piece of information, which can be done by scanning. The OECD describes that as a decline in particular on evaluating and reflecting, with the least-affected tasks being those that asked for a single piece of information to be located. Over the longer period to 2025, the OECD reports, the declines were more uniform across kinds of task, so this contrast belongs to the first period and not to the whole seven years.
The second pattern concerns the length of the text a question is attached to, which is a different feature of an item and not another kind of task. Success fell more on questions attached to longer passages than on those attached to shorter ones, and the OECD calls text length a strong driver of the overall decline. That gap was already there by 2022 and, the OECD says, did not widen afterward.
What became harder
The decline was uneven, in different ways in different periods
Held up best, 2018 to 2022
Finding one stated piece of information.
Locating information, which can be done by scanning. The OECD's least-affected kind of task in this period.
Weakened more, 2018 to 2022
Holding several pieces of information in mind at once and judging them.
Evaluating and reflecting, which the OECD singles out for 2018 to 2022. Over the full period to 2025, declines across kinds of task were more uniform.
Longer passages, 2018 to 2025
Questions attached to longer texts fell more than those attached to shorter ones.
A separate feature of the question, not a kind of task. The gap was there by 2022 and the OECD says it did not widen afterward.
The finding I keep coming back to is the one about text length, because it is the least abstract thing in the report. It is not a claim about attitudes or attention spans in general. It is a measurement that students can do less with a long text than students could do with the same long text seven years ago.
PISA’s short fluency check, where students judge quickly whether simple sentences make sense, tells the same story in miniature. The share reading both accurately and at a reasonable pace fell by about seven percentage points between 2018 and 2025, while the share answering fast and wrong nearly doubled, from about 7% of students to 11%, most of that since 2022. The same pattern runs through the whole reading test: more quick wrong answers, fewer questions left blank. Students did not give up. They answered quickly instead of carefully. The OECD’s own conclusion gives a leading role to the declining capacity of students to engage with reading tasks that require careful reading and sustained attention.
What is plausible. Reading habits have been shifting for longer than the scores have. PISA’s earlier work on reading in a digital world21st-Century Readers: Developing Literacy Skills in a Digital WorldBefore the pandemic, reading enjoyment and book, magazine and newspaper reading were already falling across OECD countries, with more students reading only when required and more reading online.View the report → found that between 2009 and 2018 more students said they read only when they had to, that enjoyment of reading fell across OECD countries, and that students were reading fewer books, magazines and newspapers while reading more online. In Norway, the share of 15-year-olds who did not read for pleasure rose from about 40% in 2009 to about 50% in 2018.
That relationship runs both ways, and anyone claiming otherwise is overreaching. Students who read well find reading rewarding and so read more, and students who read more build the vocabulary, knowledge and stamina that make reading easier. Falling enjoyment is at least as likely to be a symptom as a cause, and belongs in the picture as part of a cycle rather than as an explanation of it.
Attendance is a more tractable candidate. Across OECD countries, the share of nine and ten-year-olds missing school at least once a fortnight rose from about 11% in 2019 to about 15% in 2023, and the effect of missed teaching on learning is well established. What nobody can quantify is how much of the international decline it accounts for. Classroom conditions have tightened too. TALIS 2024Results from TALIS 2024: The State of TeachingTeachers reported spending 16% of class time keeping order in 2024, up from 13% in 2018. Self-reported, with national non-response caveats in several systems including Norway.View the report → found teachers spending 16% of lesson time keeping order, up from 13% in 2018, which is teaching time that does not come back.
Reading on paper against reading on screen belongs here too, carefully. Delgado and colleagues (2018)Don’t throw away your printed books: A meta-analysis on the effects of reading media on reading comprehensionA modest overall comprehension advantage for paper (about 0.21 of a standard deviation), concentrated under time pressure and for informational texts, and absent for narrative. The underlying studies are mostly short laboratory comparisons, so this does not explain the international decline.View original study → pooled 54 experiments involving more than 170,000 readers. Comprehension was modestly better on paper, and the detail matters more than the headline: the paper advantage was largest when readers worked against the clock, much smaller when they set their own pace, present for informational texts and absent for stories. That is a finding about particular reading conditions, not a finding that screens damage reading. Filing a long article read slowly on a tablet and a run of short videos together under “screen time” throws away the distinction that matters.
What remains guesswork. Which of these matters most. Whether digital life has reorganized attention or merely reflects a change in what people choose to attend to. Whether the pandemic worked mainly through lost teaching or by accelerating shifts that were already under way.
The trend is clearer than its cause, and both halves of that sentence are load-bearing.
The OECD’s own position, having worked through explanations at the level of the student, the school, the education system and society at large, is that these are complementary rather than competing, with several small forces lining up to produce a large effect. That is less satisfying than a villain. It is also more likely to be true.
There is a wider conclusion I think the breadth of this evidence supports. It is becoming hard to treat this as a problem belonging to particular students, particular teachers or particular schools. Reading has weakened across countries with very different curricula, across more than one age group, and across most of the achievement range rather than only at the bottom. A pattern that wide is systemic in scale, whatever turns out to be driving it. That is not the same as saying there is one system-level cause waiting to be identified, and I would resist filling the gap with one. The forces at work may well differ from one country to the next. But scale of that kind does settle one thing. The response cannot sit entirely with the English teacher who notices that a class can no longer get through a chapter.
Part IIIWhat follows
Reading is not an English department problem
One finding in the 2025 data should change how a whole school thinks about this. Countries that lost the most ground in reading also lost the most in mathematics and science, and the relationship across countries is close enough that the OECD reads it as a sign of shared causes rather than three separate problems.
Part of that is common causes. Part is more direct. PISA mathematics does not test procedures in isolation; it sets out a situation in writing and asks students to reason about it. A student who cannot hold a three-paragraph scenario in mind cannot show what they know about proportional reasoning. The same is true across the timetable. A science explanation, a historical source, a set of instructions, a word problem and a chapter of a novel each ask something different of a reader, and none of them are available to a student who reads by scanning for the answer.
This does not turn every teacher into a reading specialist. It does mean a secondary school can no longer assume that reading development finished in primary school and that our job is to hand texts to readers who are already made.
What I think teachers should do
PISA does not prescribe pedagogy, and any article claiming otherwise deserves suspicion. What follows is professional judgment, shaped by the international picture and by the wider research on how comprehension develops. None of it will reverse an international trend, and no set of classroom decisions could. What these can do is protect the specific thing the evidence says has weakened, within the part of a student’s reading life a teacher actually shapes: sustained, effortful engagement with a demanding text.
Keep students reading extended text. PISA can show that students now perform less well when a task demands sustained work across a longer text. It cannot show that keeping them in long texts will reverse the decline, and it was never designed to. The recommendation is an inference from the diagnosis. If every complex text in a course reaches students as slides, extracts, bullet summaries, teacher paraphrase or an AI summary, we have quietly removed the thing they are losing the capacity to do. Scaffolding should help a student into a demanding text, not stand in for having read it.
Build vocabulary and knowledge on purpose. Comprehension is not a free-floating skill that travels across any content. It depends on knowing enough words, and enough about the world, to make sense of what is on the page. Teaching strategies without building knowledge produces students who can name the strategy and still not understand the text.
Protect attention, and let scaffolds lead back to independence. Sustained reading needs conditions that allow it: uninterrupted stretches, clear expectations, devices used deliberately rather than ambiently. That is a claim about conditions, not about technology. And pre-teaching, modeling and guided reading, useful as they are, make it easy to lose track of what a student can do alone. Take the support away periodically and find out.
Do not simplify away the difficulty. When students struggle with a text, the reflex is to find an easier one. Sometimes that is right. Often what is needed is better preparation and a second encounter. A curriculum that lowers the demand every time students find something hard guarantees they will never build the capacity to meet it.
Make reading visible. Talk about what readers do when they are confused: reread, slow down, check a word, hold a question open, notice how a text is put together. Students who answer quickly and wrongly are usually not lazy. They do not know what careful reading looks like from the inside, because nobody has shown them.
The Reading Conditions Check
Before setting a demanding text, five questions are worth asking. They are not a lesson template. They are a way of working out which condition is missing when a text does not land.
Access · Attention · Demand · Support · Independence
1Access
Do students have enough vocabulary and background knowledge to get into this text at all?
2Attention
Have I created conditions in which sustained reading is genuinely possible?
3Demand
Is this text difficult in the way I intend, rather than trivially easy or inaccessible for reasons unrelated to the thinking I want?
4Support
Does my scaffolding help students read the text, or let them complete the task without reading it?
5Independence
Somewhere in this unit, do students have to make meaning without me?
What students should understand
Reading competence is built partly by doing something that has become very easy to avoid: staying with a text when understanding does not arrive immediately.
The distinction worth teaching explicitly is between having read the words and having made meaning from them, and the PISA results suggest a lot of students no longer reliably notice the difference. Three honest checks: Can I explain what I just read without looking back? What is the writer’s point, as opposed to what the passage was about? Which word or idea is the one blocking me?
None of that needs a worksheet. It needs students to know that the feeling of having finished a page is not the same as having understood it.
What this means for parents
Reading development does not stop when a child becomes a fluent decoder. That is roughly where the harder work begins, and it carries on through adolescence. What adolescents benefit from is unglamorous: texts they actually want to read, conversations about what they are reading, stretches of quiet long enough for a long text, and adults who treat reading as normal rather than remedial. The useful question is not how many minutes a teenager reads, but whether extended reading still has any real place in their week.
One distinction is worth carrying into any conversation about screens. Reading on a device and having your attention broken into pieces by a device are not the same thing. A teenager reading a long article on a tablet is doing something much closer to reading a book than to scrolling a feed. For the question this article is asking, that makes “screen time” too blunt a category on its own.
Part IVThe limits
What I would not conclude
I would not conclude that COVID caused the reading decline. It clearly made things worse, but the decline began before the pandemic and continued after schools reopened. I would not blame tablets either. There are real reasons to look closely at how students read on screens, and at the conditions around that reading, but the evidence is nowhere near strong enough to become “screens caused this”. It does not even show that reading on a screen is inherently worse than reading on paper: the measured advantage for print is modest and depends on particular conditions rather than being a property of the medium.
The same restraint applies elsewhere. Lower reading enjoyment does not tell us that children have simply stopped caring about books, because enjoyment and competence feed each other and we cannot say which one leads. Norway’s results are not an argument for copying whichever country currently sits at the top of the table, since a ranking shows where a system ended up, not how it got there. And PISA does not tell teachers which reading program or classroom method to use. It gives a much clearer picture of the problem than of the remedy.
Where this leaves us
Secondary education has long treated reading as a prerequisite students arrive with rather than a capacity we keep building. The evidence makes that assumption difficult to hold.
We do not know why reading has weakened across so many systems, and there may not be one reason to find. But uncertainty about the cause is not uncertainty about the problem. It is well measured, consistent across two assessments and two age groups, visible in more than fifty countries, and traceable to something specific: students are becoming less able to stay with a text that asks for careful attention over time.
That is not an argument for panic, and certainly not for pretending the last twenty years can be undone. It is an argument for taking reading seriously again in the years when we had assumed it was already taken care of. Sustained attention, vocabulary, knowledge, demanding text, productive struggle, enough independent reading to become a stronger reader: none of that is new. What is new is that we can no longer assume students pick it up somewhere else.
Research cited
Betthäuser, B. A., Bach-Mortensen, A. M., & Engzell, P. (2023). A systematic review and meta-analysis of the evidence on learning during the COVID-19 pandemic. Nature Human Behaviour, 7, 375–385. Preregistered systematic review, 42 studies across 15 countries. https://doi.org/10.1038/s41562-022-01506-4
Delgado, P., Vargas, C., Ackerman, R., & Salmerón, L. (2018). Don’t throw away your printed books: A meta-analysis on the effects of reading media on reading comprehension. Educational Research Review, 25, 23–38. Meta-analysis, 54 studies, 171,055 participants. https://doi.org/10.1016/j.edurev.2018.09.003
Department for Education, England (2026). PISA 2025: National report for England. Official national analysis; source here for OECD comparison figures. https://assets.publishing.service.gov.uk/media/6aa0073962ec7fe7bedf5160/pisa-2025-national-report-for-england.pdf
Mullis, I. V. S., von Davier, M., Foy, P., Fishbein, B., Reynolds, K. A., & Wry, E. (2023). PIRLS 2021 international results in reading. IEA, TIMSS & PIRLS International Study Center, Boston College. International assessment of reading at around age ten, data collected 2021. https://pirls2021.org/results
Norwegian Directorate for Education and Training, Udir (2023). PIRLS 2021: Markant nedgang i norske tiåringers leseferdigheter. National analysis of Norway’s PIRLS 2021 results. https://www.udir.no/tall-og-forskning/finn-forskning/rapporter/2023/pirls-2021/
Norwegian Directorate for Education and Training, Udir (2026). PISA 2025: Kompetanse i naturfag, lesing og matematikk. National analysis; 6,469 students in 281 schools. https://www.udir.no/tall-og-forskning/finn-forskning/rapporter/2026/pisa-2025-kompetanse-naturfag-lesing-matematikk/
OECD (2021). 21st-century readers: Developing literacy skills in a digital world. PISA, OECD Publishing. Secondary analysis of PISA 2009 and 2018 reading habits. https://doi.org/10.1787/a83d84cb-en
OECD (2025). Results from TALIS 2024: The state of teaching. OECD Publishing. International survey of lower secondary teachers, data collected 2024. https://doi.org/10.1787/90df6235-en
OECD (2026). PISA 2025 results (Volume I): Future-ready students. OECD Publishing. Over 760,000 students in 91 countries and economies, data collected 2025. https://doi.org/10.1787/73451bc5-en
Wisenöcker, A. S., Helm, C., Große, C. S., Hübner, N., & Zitzmann, S. (2025). A meta-analysis of students’ academic learning losses over the course of the COVID-19 pandemic. Learning and Instruction, 98, 102111. Meta-analysis, 103 studies across 45 countries. https://doi.org/10.1016/j.learninstruc.2025.102111
Each citation above carries two short judgments, visible when you open it. Study quality rates the design’s rigor: A for preregistered or peer-reviewed work with strong designs or large meta-analyses, B for solid single studies or secondary reports, C for preprints or smaller, quasi-experimental designs. Confidence in the conclusion is Northlight’s own judgment of how much weight a finding should carry for the specific point it supports here, which can be lower than the study quality alone would suggest.
Research & AI transparency
AI tools assisted with locating, organizing, comparing, and checking research, and with drafting and revising parts of this article. The tools were not treated as sources. Sources were checked against the originals. Northlight Studies determined the question, interpretation, strength of the claims, position taken, and classroom implications.