Uruguay: 86.7% of teenagers use AI as reading and math decline; correlation is not causation

By Cadena del Mar · October 21, 2026.

Uruguayan ninth-grade students performed below the targets set by the National Public Education Administration (ANEP). That is the finding of Aristas Media 2025, the national assessment by the National Institute for Educational Evaluation (INEEd), presented on Wednesday. According to the report, 86.7% of teenagers say they use artificial intelligence tools.

The sample includes 9,508 students in 564 groups across 306 schools. The tests were conducted between October 6 and 24, 2025.

In reading, ANEP aimed for at least 77% to exceed the second performance level; 72% did. In math the target was 36%, and 30.6% was reached. According to the report, the trend compared with 2018 and 2022 is downward in both areas, with more students at the lower levels.

On technology, 66% of students admit to using AI to complete activities during class. Some 83.3% say several classmates are on their cell phones in class, even though 72.9% note that phone use is prohibited. Among teachers, 68.4% of language teachers and 61.7% of math teachers say they have incorporated AI into their practice. Six in ten teachers of both subjects cite the cell phone as the main distraction. Among adscriptos (staff who support students at school), the share rises to 79.6%.

The social gap still weighs heavily. In math, 60.6% of students from very advantaged backgrounds exceeded level two, compared with 12.5% of those from very disadvantaged backgrounds. The report notes that, once sociocultural context is accounted for, there are no differences in reading between public and private secondary schools.

Reading for pleasure is also falling. In 2022, 71.7% liked to read. In 2025, it is 47.2%. Some 66.6% did not read on the day before the survey (54.8% in the previous measurement), and only 7.4% spent more than an hour reading.

**Our take.** The temptation is to connect the dots: more AI and more phones, worse grades. But the available material does not demonstrate that causality. The report shows a downward trend compared with 2018 and 2022, but it does not isolate the effect of generative AI, which became popular from late 2022 onward. In general, education systems in many countries have been dealing with a decline in learning that predates chatbots, with the pandemic and inequality as factors. The AI figure is a legitimate alarm, not a verdict.

What the report does say, and what deserves attention, is that two-thirds of students use AI to complete tasks in class. If the tool does the cognitive work, the student does not train it. It is a reasonable hypothesis, but it needs evidence that measures it. It is also striking that teachers and students are both using AI without the data reflecting any common guideline on how to do so. The phone ban exists and, according to the students themselves, is not enforced.

Inequality is the starkest figure. A gap of 60.6% versus 12.5% in math matters more than any gadget. If AI reaches everyone but only some know how to use it as a tutor rather than a shortcut, it could widen that gap. That is the short-term risk, and it should be said plainly: a generation could end up less well educated in a poorly managed transition.

In the long run, the same technology can be a personalized, patient tutor, something that today only well-off families can afford. That points to a society with more opportunities. But that potential will not fulfill itself. It requires teaching students to read, reason and ask AI for help without outsourcing their thinking. The collapse in the enjoyment of reading, from 71.7% to 47.2%, is exactly the kind of habit no tool can replace.

Our thesis: the answer is neither to ban nor to embrace uncritically, but to measure the effect of each use and redesign tasks so that the effort cannot be outsourced. Regulating panic is easy; governing with evidence is what is missing.