PISA 2025: Colombia hits rock bottom in education, and AI is not the explanation it seems

By Zendoric · September 12, 2026.

The results of PISA 2025, the test the OECD administers every three years to 15-year-old students in reading, mathematics and science, confirm an unprecedented collapse. According to the Programme for International Student Assessment (PISA) itself, this edition recorded "the lowest average performance observed so far in the OECD" in science, reading and mathematics, with a sample of more than 760,000 students tested across 91 countries and economies, representative of some 33 million 15-year-olds worldwide.

Colombia illustrates that decline starkly. The report covered by Las2orillas places the country's performance well below the OECD average (461 points) and below the best regional result, Chile's (436). In reading, the gap with the world leader, Singapore, is vast: 399 points against 535. Only 1% of Colombian students reach high performance levels, while 43.9% fail to reach basic levels. Overall, seven out of ten students assessed did not attain the minimum competencies expected for their age: barely 29% master basic competencies in mathematics (against 65% in the OECD), 46% in reading (against 69%) and 50% in science (against 74%). Within the region, Colombia falls below Chile and Uruguay, though above Argentina, the Dominican Republic and Peru.

The report adds two pieces of context that rarely make the headlines: 20.7% of Colombian students report having suffered bullying, above the OECD average (19.6%); and 63.5% say they use digital devices for leisure more than an hour a day, while 25.9% admit to being distracted by digital resources in most of their classes.

This is where the original article poses its question: did artificial intelligence have anything to do with this collapse? The figure it offers is suggestive: Colombian students are among those who turn to these tools most often to do their homework, even more than Singapore, the country with the best global results. But that same figure, read carefully, points in the opposite direction to the one the question implies: if Singapore uses AI for schoolwork at a lower rate than Colombia and still leads the world ranking, AI use on its own does not explain the performance gap.

There is an even more decisive argument in the report itself: the decline did not begin with ChatGPT. According to PISA, reading peaked around 2012, science has been declining gradually for a decade and mathematics has suffered its sharpest drop since 2018, years before generative AI reached classrooms in late 2022. Blaming AI for a trend that predates it is, at best, a simplification; at worst, a convenient alibi for not looking at the structural problems.

Our reading is that the debate is badly framed. The variable separating Singapore from Colombia is not whether students use AI, but how and with what support network they use it. A system with well-trained teachers, demanding curricula and solid infrastructure can turn AI into a tutor that amplifies learning —what in our analysis of the education sector we have described as the teacher who orchestrates the tool rather than merely transmitting content. A system with overcrowded classrooms, connectivity gaps and structural poverty, by contrast, runs the risk of AI being used as a shortcut to avoid cognitive effort. That is the real, short-term risk that deserves attention in Colombia: not the tool itself, but the absence of a pedagogical framework to accompany it. The 25.9% of students who say they are distracted in class by digital resources is a symptom of that missing framework, not proof that AI ruins learning by definition.

In the long run, that same technology is the one with the greatest potential to reverse precisely the kind of educational inequality these figures reflect. A personalized AI tutor, available outside school hours and adapted to each student's pace, could reach where teachers, materials or stable connectivity are lacking today, provided it is deployed with the right pedagogical support and not as a cheap substitute for investment in education. The abundance AI promises is not only economic: it can also be an abundance of personalized educational attention, today a luxury reserved for those who can afford a private tutor.

Colombia does not need a technological scapegoat; it needs to confront an education crisis that the PISA tests have been documenting since before ChatGPT existed: poverty, territorial inequality, teacher training and bullying —the last of these is the only one of those factors the source backs with data, in the same report that mentions AI in passing. Pointing at the tool of the moment is easier than funding the education reform the country needs, but it does not solve it.