The figure is striking. At level 2, which PISA considers the baseline level of proficiency, students can at least identify the main idea of a moderately long text, find information under certain conditions and reflect on a text's purpose or form when asked.

Only 41% of Vietnamese students reached level 2 or above, compared with an average of 69% across participating countries. Vietnam's average reading score was 392, below the OECD average of 461, which itself fell from the previous assessment.

At the same time, PISA found another trend: more students are becoming "hasty readers." They move through texts so quickly that they may not finish reading them before giving an answer, often getting it wrong.

This is a broader trend. Across 35 OECD countries, the share of hasty readers nearly doubled from 6.6% in 2018 to 11.4% in 2025.

These two figures measure different things. The share of students who fail to reach basic reading proficiency tells us one thing. The share of hasty readers tells us another. We cannot conclude that skimming is the cause of weaker reading skills. But put the two figures side by side and a bigger question emerges: How do we teach students to read accurately and get them to spend enough time and attention to understand what they are reading?

Children read text books and reference books displayed in a school yard in Vietnam. Photo by Ho De

People are "reading" more than ever: Text messages, emails, Facebook posts, TikTok videos, news headlines, comments, and app notifications. But much of this reading happens at a very different pace. Our eyes move very quickly, we look for keywords, pick up a few points and move on to the next thing.

One U.S. student interviewed by The New York Times said finishing a book has become much harder because they constantly lose focus and want to check their phone. Other students said short videos, social media, and screens have made it harder to stay focused on long texts.

Some admitted that AI gives them answers without requiring them to understand how those answers were reached.

Another OECD report found that Covid-19 played a significant role in this trend. Online learning put one generation in front of screens for more time than before and disrupted the development of foundational skills.

But it is hard to blame the pandemic for everything. Reading performance had already begun to decline before Covid.

And now we have AI.

The better machines become at reading, the less people may feel they need to read themselves.

A 100-page report can be summarized in seconds. A long research paper can become five bullet points. Students can ask AI, "What is this passage saying?" without having to work through the passage themselves.

Technology can save people time. These tools are expected to become "a natural part" of learning and work. The problem begins when AI moves from supporting reading to replacing it.

If students read only an AI-generated summary, they may know what the author concluded without knowing how strong the evidence is, what assumptions underpin the argument or where its weaknesses lie.

If the AI gets the summary wrong, someone who never read the original will have little chance of spotting the error.

I saw this happen just two weeks ago when a PhD student used AI to summarize a reference paper. A colleague of mine, who happened to be the paper's author, immediately pointed out that "the AI's conclusion was completely opposite to the paper's message and findings." The PhD student then received a 30-minute lecture from the professors.

That creates a paradox. The smarter AI becomes, the more people need strong reading comprehension, the ability to verify information and the ability to think critically.

Following the results, Vietnam's Ministry of Education and Training said: "This is not just a matter for literature classes. To solve a word problem in math, understand a scientific phenomenon, read a map or chart or follow instructions on the Internet, students need to know how to read, identify important information, compare it and draw conclusions."

So, what is the solution?

The first is very traditional: Read more and read longer.

Schools do not have to force students to start with books that run hundreds of pages. They can set aside regular periods for uninterrupted reading, gradually increase the length and complexity of texts and ask students to explain the evidence behind their answers rather than simply choose one.

Second, teachers need to rethink assignments in the age of AI. Instead of asking only, "What did the article conclude?" they can ask: "Which passage supports that conclusion?" "Where do these two sources contradict each other?" "What did AI leave out of its summary?"

Students can still use AI, but they have to return to the original text to check it. That is the approach the University of Bristol, where I teach, has adopted in some courses.

In that case, AI does not replace reading. It gives students another reason to read closely.

Finally, adults cannot stand outside this discussion when it comes to education. We cannot ask children to read slowly and think carefully if we habitually read only headlines before sharing them, watch a few dozen seconds of a video before forming an opinion or argue about an article we have not finished reading.

In an age when machines can read millions of pages in seconds, the human advantage does not lie in reading faster than a machine. It lies in reading slowly enough to know what can be trusted, deeply enough to understand what is being said, and independently enough to keep thinking and questioning rather than hand those tasks over to a machine.