PISA 2025 Shows Learning Declines as Students Increase AI Use

The release of the international PISA 2025 survey has laid bare a complex dual reality in modern classrooms. While learning outcomes in core academic subjects have largely stagnated or worsened since the drops recorded in 2022, students have rapidly adopted generative artificial intelligence tools for schoolwork according to reporting from Infobae. In Argentina, for instance, 52 percent of fifteen-year-old students report using AI chatbots at least once a week to support their learning.

Yet this technological integration unfolds against a backdrop of severe academic strain. Only 24 percent of students in Argentina reach the minimum expected proficiency level in mathematics, while reading and science proficiency stand at 40 percent and 41 percent respectively. These figures place the country well below the averages established by the Organisation for Economic Co-operation and Development.

Global Reading Deficits and the Rise of Superficial Digital Habits

The broader international picture outlined by PISA 2025 highlights reading literacy as an area of acute vulnerability. OECD Secretary-General Mathias Cormann noted that average scores in reading literacy fell by 28 points in OECD participating areas and by 16 points in non-OECD countries between 2015 and 2025. Mathematics scores across the OECD also dropped by an average of 22 points over the same decade, while science performance remained comparatively stable as detailed by Heise.

Education experts connect this reading crisis directly to transformed digital consumption habits. Andreas Schleicher, the OECD Director for Education and Skills, pointed out that the proportion of superficial readers—students who skim texts hastily and jump to incorrect conclusions—has nearly doubled globally between 2018 and 2025. Schleicher observed in a commentary for a German excerpt of the results that decreasing frequency of reading for pleasure and a shift toward rapid social media feeds impair students’ ability to engage with complex texts.

Schleicher also warned that excessive reliance on automated tools risks eroding independent thought, noting that the widespread use of artificial intelligence should not lead to people unlearning how to think for themselves according to the same international analysis.

Stagnation and Demographic Divides Across Latin America

Regional findings underscore how wider performance trends intersect with local socioeconomic factors. In the Dominican Republic, PISA 2025 data marks a sharp halt to the upward trajectory the country demonstrated between 2018 and 2022. Achieving 361 points in science, 339 in mathematics, and 346 in reading, the nation registered virtually no statistically significant change compared to the previous survey cycle as reported by Acento.

Only about 9 percent of Dominican students reach the minimum expected level in mathematics, alongside 23 percent in reading and 26 percent in science. Furthermore, socioeconomic background continues to exert a powerful influence on educational outcomes. A notable 69-point gap separates students of higher and lower socioeconomic conditions in science.

Gender performance patterns also emerge clearly in the regional data. Dominican adolescent girls outperformed boys in science, scoring 364 to 357, and maintained a substantial 30-point advantage in reading at 360 versus 330. Boys retained a smaller lead in mathematics, scoring 342 compared to 336 for girls according to regional commentary.

Classroom Integration of Artificial Intelligence and Distraction

The survey reveals that students are integrating artificial intelligence into multiple stages of their academic routines. Beyond the 52 percent who use chatbots weekly to learn in Argentina, 41 percent use them to conduct preliminary research, 41 percent to summarize required readings, and 32 percent to draft written assignments. Only one in ten students reports never using these tools for schoolwork.

Photo: acento.com.do

However, researchers emphasize that the educational impact of technology depends entirely on how it is deployed. Asking an AI model to generate a summary requires very little cognitive effort compared to comparing that summary against an original text, evaluating its accuracy, or scrutinizing its arguments. When digital devices function primarily as shortcuts rather than tools for critical engagement, they may bypass the cognitive processes essential for deep learning.

Photo: Infobae

The boundary between academic utility and classroom distraction remains porous. Argentine students report spending an average of 1.7 hours daily on digital devices for learning at school—matching the OECD average—while also logging 1.6 hours daily on recreational digital activities inside the classroom, compared to an OECD average of 1.1 hours. Moreover, 55 percent of Argentine students state that their peers become distracted by digital devices during most or all science lessons, well above the 28 percent OECD average noted in educational reporting.

As educational systems grapple with these shifting dynamics, further insights are anticipated. The PISA 2025 survey introduced a dedicated testing category examining self-regulated learning and algorithmic problem-solving titled Learning in the digital world, with detailed results scheduled for publication in autumn 2027 according to international survey disclosures.