Average reading scores across the more than 90 countries the OECD tests have fallen 28 points in a decade, which the organization puts at roughly a year and a half of schooling.
The straightforward conclusion is that a less literate workforce produces a slower-growing economy. John Burn-Murdoch, who ran the numbers on adult literacy and earnings for the Financial Times, said the economy may be the last place the damage becomes visible.
"The extreme version of this is that among the subset of the American adult population who are functionally illiterate, unable to read or understand even quite basic texts, they earn on average the same amount as the average Brit."
Burn-Murdoch is the Financial Times' chief data reporter and a columnist there, and the analysis he brought to the segment is his own: American adult literacy against British adult literacy, set against what each economy actually produces.
I listened to the full segment so you can skip it.
Here are the 7 arguments that matter.
👤 Guest: John Burn-Murdoch, Chief Data Reporter and columnist at the Financial Times, who writes the paper's data-led column on demography, health and economics
🎙️ Host: Ed, who presents Prof G Markets for Prof G Media
📰 Published: 9 September 2026 on YouTube (Prof G Markets)
🔴 YouTube | ⏱️ 37 min
Key Takeaways
The pandemic explanation for falling test scores no longer works, because the follow-up shows more decline rather than recovery
Three years ago the drop was read as school closures; three years on there is no rebound
The OECD found that 15-year-olds who use AI regularly are worse at summarizing educational texts
The gap held after adjusting for how educated and well-off the students' families are
Falling scores may hit the economy last rather than first
Adult literacy is far lower in America than in Britain, and the American economy performs far better
The economic returns come from the top performers, and in the United States many of them arrived from somewhere else
The bottom 10% of American students is falling away from everyone else, in this test and in others
Students who lean on AI look strong on homework and collapse in the unaided final exam
A study of Chinese high schools and universities found exactly that pattern
Burn-Murdoch expects schools to cut homework and coursework and go back to supervised final exams
AI is not the calculator, because with a calculator the learning happens first
Writing is where people work out what they think, and that step is the one being skipped
1. Not A Pandemic Blip
Ed set up the segment with the report card itself. The OECD test measures 15-year-olds in reading, math and science across more than 90 countries. Average reading scores are down 28 points over the decade, math down 22 and science roughly 10 — all three the lowest averages the OECD has recorded. US reading fell 14 points to America's lowest score ever, while math and science were flat, with math at exactly the OECD average.
Ed's own framing of it was blunt: "It seems like children, not just in America but around the world, are frankly getting stupider"
Burn-Murdoch's first point was about how rare a move this size is in data of this kind. "It's really not very common that we see such a drastic deterioration in a massive international data set like this"
The pandemic reading was tested and did not hold up. Three years ago, he said, the results were explained by school disruption. "Three years on, instead of a partial recovery or even stability, we see sustained decline"
On the scale over the full period: "The numbers you're talking about here—we're looking, really, at over the past decade, a decline equivalent to two years of lost education in mathematics and reading."
He pushed back on treating the figures as abstractions: "These are not just meaningless numbers—they reflect really measurable worse outcomes for young people and children around the world"
The breadth is what rules out a policy explanation. "This is not a clear story of a particular policy in a particular place—this is much, much, much more broad-based than that"
2. What The OECD Blames
Asked what is driving the decline, Burn-Murdoch gave three candidate causes and was careful to say the OECD's own analysts, not commentators, are the ones raising them.
He grounded the concern in who is voicing it: "So the OECD, who produced these numbers, have also put out some commentary accompanying this—this is not just sort of people looking for a negative, alarming story, these are seasoned professionals in the education space who are saying that there are real signs of problems here."
The OECD tested AI use directly this year. "And in one particular piece of analysis, they found that among children—so we're mainly talking about 15-year-olds here—among 15-year-olds who are pretty regular users of AI, their ability to summarize educational texts was significantly worse than people who don't use AI."
The obvious confounder was checked. "And even when they had tried to adjust for the kinds of things that might explain that—like maybe the people who use less AI are also from better-off, more educated backgrounds—even after controlling for that, there was a link there."
COVID has a long tail in attendance as well as in learning: "There were other disruptions that have continued from that as well—so a small number of school students stopped attending school regularly during the COVID lockdowns and never really got back into it."
The third cause is how children now read anything. "So again, this was something that the OECD picked up on in the report themselves—they found that an increasing share of young people, when they're doing these tests, are skimming through them super fast and then getting really bad results."
He tied that to media habits rather than to schools: "So there's certainly evidence that changes in how people consume information—the endless sort of stream of nuggets of information that people now encounter on social media—is very different to the longer-form pieces of text that people need to analyze when doing math, reading, science exams."
3. Why GDP May Not Show It
Ed pressed the market case: a society that cannot read, calculate or solve hard problems is an unproductive one, so where is the hit to GDP growth, and are leaders treating it seriously? Burn-Murdoch called his own answer the controversial one.
"A more, perhaps, controversial take that I could give to this is that I think some of this stuff actually may matter less than we think for the economy"
The evidence he brought is a comparison of two rich countries. "I looked at the fact that literacy rates among adults in America are much lower than they are in the UK, but the American economy performs much, much, much better than the British"
The version of that finding that lands hardest: "The extreme version of this is that among the subset of the American adult population who are functionally illiterate, unable to read or understand even quite basic texts, they earn on average the same amount as the average Brit"
His explanation is that output is driven by the tail, not the average: "So I think there is this sense now that what matters to drive the economy is how your very top performers do educationally"
He added the immigration channel: "And in a country like the U.S., where you have, in the past at least, taken in some of the most highly skilled, talented, brightest people from everywhere in the world—those are the people who start the companies that shape the economy."
He was explicit that this is not a reason to relax. "So, to be clear, I'm absolutely not saying that this is not a really important deterioration—I think it's massively important, it tells us that a lot of things are going wrong." The damage shows up first in daily life — understanding detail, weighing tradeoffs, judging whether information is any good — and in politics. "But weirdly, it may actually be that the economy may be one of the last places that we see this start to show up"
4. The Bottom 10% Falls Away
Ed took the argument to distribution: if the gains are concentrated in a small, well-educated group, GDP can rise while the spread widens. He said: "But also, more importantly, that is what we saw in this report too—I mean, the inequality of education was the number that stood out most for the U.S., nearly 300 points between the top and the bottom students in science."
Burn-Murdoch agreed that inequality is the right lens. "I think that's exactly right—I think inequality is exactly the way that I would look at this"
The pattern is not an artifact of one test. "The various other exams, assessments of reading in particular in the U.S., that are done every year and have been done for decades, show the bottom 10% really falling away"
His read of who is losing ground: "So I think that's exactly right, that what we seem to be seeing in this era of greater exposure to digital risks, for example, and disruptions—it's the bottom 10, 20% who are really falling away from the rest."
The consequence he named is civic as much as economic. "I really don't think we're headed to a good place if you end up with one in ten, one in five young people, and they become adults, and these people really just have not been equipped with the skills they need to navigate the complexity of modern life"
5. AI As A Crutch
Ed cited the report's AI finding — "The report showed that daily AI users scored 28 points below those who never use AI, meaning that if you're using AI every day, you are about a year and a half behind your peers who are not using AI" — and noted that Zohran Mamdani has opted to ban AI in New York City schools.
Burn-Murdoch said the evidence is accumulating from several directions. "So there was also a study a few weeks back out of Chinese high schools and universities that showed that students who were using AI to help with their assessments and their homework—it looked like they were doing really well on the homework and these examinations throughout the course of the year, but when it came to the final exam, where they didn't have AI, they performed really, really badly." He said the same has come out of US universities
His conclusion is about dependence rather than cheating. "So I think at this point there's a really solid case that relying on AI as part of the learning process is extremely risky, in terms of it essentially being used as a crutch that, when taken away, people just collapse at the end"
Enforcement he treats as close to hopeless: "I think it's also just incredibly hard to enforce non-use of AI in the education setting where that doesn't happen—you can talk about having lockdown versions of ChatGPT, that kind of thing that encourage study hours, but ultimately if someone wants to cheat, they're going to find a way to cheat."
He put part of the blame on who designs the tools. "I think, as well, we have this problem where a lot of the people in Silicon Valley who create these tools, they are people who have always worked really hard at school, they're always hyper-focused, they don't struggle with distraction—for them, it maybe seems obvious that, well, kids who are given these tools are going to use them to learn even more." He said that view comes from a disconnect with real classrooms: "But I think that is just born out of a real disconnect from what real classrooms are like in modern UK, modern America, where just the ability to focus is a huge challenge for a lot of people." And instant answers are "incredibly hard to resist"
The problem does not stop at the school gate. He said: "So, yeah, I think there is mounting evidence that having this available in classrooms is a problem, but even having it available outside of classrooms is a problem."
6. Exams Come Back
Out of that argument came the one forecast in the segment, about how institutions respond.
He expects assessment to move back to the exam hall. "I suspect over the next few years we'll see schools, universities, colleges shift more and more to having rigorously assessed final exams, getting rid of more homework, getting rid of coursework"
The design principle behind it is that the graded moment has to be unaided: assessment "is done away from AI," with the rest of the year building up to a final test taken with no AI
He framed the whole enforcement problem as one nobody has solved: "It's a really, really fine tightrope to try and tread, and we don't seem to be able to do that"
7. Why AI Isn't A Calculator
Ed put the strongest version of the other side to him: if AI is going to be everywhere, fluency in it matters more than performing tasks without it, and a test that measures unaided work is measuring the wrong thing.
Burn-Murdoch conceded most of the premise. "AI is going to be heavily embedded into the world of work of the future. It's already heavily embedded in the work of the present in a lot of areas—I certainly use it a lot, for example."
The objection is about sequence, not about tools. "But the problem is, school is where you learn." Calculators and spreadsheets exist in the workplace, "But we still learn to calculate mentally, to make those calculations in our heads or on paper, as we're going through the education system. It's only once you get to a certain point that the calculator comes in"
What AI removes is the stage where knowledge is built. "So the risk with AI is that the learning never happens—we go straight to the AI, we don't learn the depth of knowledge that we might then need to navigate the world later on"
The second loss is the harder one to replace. "But we don't even learn the act of how to learn—we don't learn how to spend an extended period of time with a piece of written work trying to understand what it's saying, we don't learn how to write extensively ourselves, and the act of writing is often where we actually come to understand what we think about something."
His proposed settlement: "So I think you can imagine a world where the education system is done AI-free as much as possible in terms of the actual learning experience, but there is then a sort of fast track later on in the education process where people are brought up to speed on the AI tools that they're then going to encounter in the world of work."
Bonus Insights
Ed said the OECD released the report the previous day, and that the US had one of the widest gaps between its highest- and lowest-scoring students of any country tested
Ed made the case for covering it on a markets show himself: "So the reason I really wanted to have you on to talk about this—I mean, this is not an education podcast, this is a markets podcast, we cover finance, we cover economics. But to me, this really is an economic story."
Burn-Murdoch's broader framing was that the skills at stake are not mainly economic ones: "I think this is a problem just anywhere in life, where the ability to understand details, to think about tradeoffs, to think critically about the quality of information—anywhere that comes in, this is going to be an issue. So that affects the economy, it affects the functioning of political systems, it affects how we interact with each other."
He also said the concern in the coverage is warranted rather than overblown: "I think the concerned reactions we're seeing to this are completely justified"
On educational technology generally, he noted the pattern is not new: "We've seen this with all sorts of different forms of educational technology over the years as well"
Burn-Murdoch's bottom line is that the decade-long fall in test scores is real, broad-based and no longer explainable by the pandemic, that AI use is now measurably part of it, and that the first damage will show up in politics and daily life rather than in GDP — because in an economy where output is driven by the top performers, the people falling behind are the bottom 10 to 20%, whose losses do not show in the growth rate.
Products, Companies & Tools Mentioned
OECD (Produced the report card and the commentary alongside it, testing 15-year-olds in reading, math and science across more than 90 countries)
Financial Times (Where Burn-Murdoch published the analysis comparing US and UK adult literacy against economic performance)
ChatGPT (Named as the example of the "lockdown versions of ChatGPT" schools use to restrict access; he said determined cheating gets through anyway)
Books & Resources Mentioned
The OECD's global education report card (The source of every figure in the segment: reading down 28 points in a decade, math down 22, science roughly 10, and daily AI users 28 points behind those who never use it)
A study of Chinese high schools and universities (Burn-Murdoch's example of AI dependence: strong homework and assessment scores through the year, then very poor unaided final exams. He did not name the authors or the institution)
Burn-Murdoch's Financial Times analysis of adult literacy and earnings (The basis for his claim that functionally illiterate American adults earn about what the average Briton earns)
If this was worth your time, send it to someone who has to have a view on this.
Get the latest market chatter as it happens: