Artificial intelligence is already being used at a large proportion of the universities surveyed by UNESCO in Latin America and the Caribbean. But its integration into practice does not mean that institutions have established common rules. The gap is clear: 87 % report use in at least one area, while 26 % have a formal strategy.
These figures come from a survey of 200 institutions in 19 countries, presented by UNESCO IESALC in September 2026. They describe the responses of participants, not a representative measure of all the universities in the region. Their value lies less in ranking countries than in the question they raise: who organises the use of these tools once they have already become part of everyday work?
What the survey actually measures
Arianna Valentini Céspedes’s methodological paper specifies that the questionnaire was circulated between August and October 2025. It contained 33 closed-ended questions and was aimed at institutional leaders: rectors, senior management, and academic or strategy leads. It did not directly ask students about their learning.
After 375 responses were cleaned, 200 were retained. The survey was distributed through institutional networks, emails, LinkedIn and a newsletter, without a predefined sampling frame. The author therefore describes the work as exploratory and descriptive, without claiming statistical representativeness.
This qualification changes how the findings should be read. Saying that 87 % of the responding institutions use AI does not justify claiming that 87 % of all Latin American universities do so. Nor does it mean that all their courses or all their students use it.
Teaching, research and common rules are advancing at different rates
In the UNESCO IESALC summary, teaching and learning top the list of reported uses, at around 74 %, ahead of research, at 57 %. The general-purpose tools cited include ChatGPT, Copilot and Gemini. Some of the adoption is driven by teachers, researchers and students, before an institution-wide strategy is in place.
The presence of a tool can meet very different needs: preparing a resource, exploring a question, analysing data or helping with an administrative task. Grouping these uses under a single label obscures what is at stake. Assistance with preparing a lesson raises different questions from use in an assessment.
Governance means deciding who makes decisions, under what rules and with what oversight. A strategy sets a direction; it then needs to be translated into clear guidance. For a student, the useful information is practical: is assistance allowed for this assignment, does it have to be declared, and what must remain their own work?
A better assignment does not prove better learning
The OECD report “Digital Education Outlook 2026” offers another perspective. Its summary distinguishes between completing a task with AI and actually acquiring knowledge. The research it reviews indicates that work can improve with the help of a general-purpose tool, without that advantage persisting when the tool is removed.
The OECD also notes that uses designed around a learning objective can support learning. The difference therefore does not depend solely on the model selected. It depends on the activity set, the teacher’s role and how what the pupil or student can do afterwards is assessed.
This report is not a second assessment of the 200 institutions in the Latin American survey. It helps to interpret a wider issue: counting uses is not enough to establish their educational effectiveness. Adoption, the quality of work produced and the acquisition of skills are three distinct questions.
What questions should be asked before rolling out a tool more widely?
A university can start by defining the need that AI is meant to address. Is the aim to provide more frequent feedback on a line of reasoning, make a resource easier to access or reduce a repetitive task? Without a specific objective, it is difficult to choose a measure of success.
Next comes the rule that applies to the activity. An exercise intended to teach a method should make it possible to see the steps in the reasoning, not just the final result. Any permitted assistance should be made explicit before the work begins, so that students do not encounter conflicting expectations at the point of assessment.
Finally, the institution must be able to observe the effects and revise its decision. UNESCO IESALC’s recommendations concern, in particular, training, student participation in governance and monitoring of use. These are organisational conditions, not an automatic guarantee of progress.
The issue therefore goes beyond the opposition between a blanket ban and enthusiastic adoption. The challenge is to make expectations clear and retain a central question: after using the tool, what has actually been learned?
Illustrative photograph: the central library of UNAM in Mexico City. It does not identify an institution that responded to the survey.
