PISA 2025 and AI in the classroom: chatbot use with no simple pattern per task

Data profile

PISA 2025 asks students how often they use artificial intelligence chatbots (ChatGPT-type) for four school tasks: summarising a text, searching for information, drafting an assignment, and “help me learn” in general.

• Source: PISA 2025 public microdata (OECD), a file downloadable by anyone.
• Sample: 29,966 Spanish PUF records, out of an official total of 29,967 students in 956 schools.
• Analysis: independent, by labs.tever.es (Tever), with open code.

The finding: no simple pattern per task

Unlike distraction from digital devices in Science class — which follows a clear monotonic pattern, more perceived distraction, worse score —, none of the four AI chatbot use tasks shows a simple relationship with the score. Each task draws its own shape: for “summarising a text,” the highest average belongs to those who never use the chatbot and the lowest to those who use it daily, but with a dip at occasional use and a partial rebound at weekly use. For “searching for information,” the highest average doesn’t belong to those who never use it, but to those who use it once or twice a week.

Why it matters

The OECD does compare the average difference of around 20 points between non-users and users in some specific tasks with the approximate reference of one year of schooling — but that benchmark is approximate, varies between countries, and should not be mechanically turned into “years lost” from using AI: it’s an observational association, not a causal effect.

Related to PISA 2025 report — Spain and the role of digital devices in the classroom: same report, same microdata file, two different findings — digital distraction does follow a simple pattern, AI use does not.