← PISA 2025 dashboard

OECD PISA 2025 · data story

Fast, incorrect reading responses nearly doubled. AI’s role is unknown.

PISA finds more rushed wrong answers in reading. Portuguese students already report using AI to learn. The data raise questions about literacy, but cannot link AI to the score decline.

Results released 8 September 2026 · estimates describe successive samples of 15-year-olds

Between 2018 and 2025, fast and incorrect or partial responses rose from 4.6% to 8.9% of all reading responses. Portugal’s weekly AI use for learning is 48%, close to the OECD’s 46%. These are separate measures, not evidence of cause and effect.

Generative AI is now part of schoolwork for many students. PISA 2025 offers two useful pieces of context: students are answering some reading questions too quickly and incorrectly, and many report using AI weekly. The results do not connect those two patterns.

−25OECD average reading points, 2018–2025
4.6 → 8.9%fast and incorrect/partial responses
48%Portuguese students use AI weekly to learn

More answers are fast, and wrong

Across OECD countries, the share of all student responses classified as fast and incorrect or partial nearly doubled between 2018 and 2025. This counts item responses, not students. It is one signal of hasty responding, alongside a 25-point decline in average OECD reading performance.

The share of fast, incorrect or partial responses rose to 8.9%

OECD PISA reading assessment · percent of all responses

Do not confuse this response-level measure with the separate OECD category for students classed as “hasty readers”.

AI use is common in Portugal, close to the OECD average

In Portugal, 48% of 15-year-olds say they use AI weekly for learning. The OECD average is 46%. Portuguese students also report weekly AI use for research, summaries and drafting. PISA measures frequency here, not whether AI helped them learn or weakened independent reading.

Portugal reports slightly more weekly AI use across these tasks

Share of students · Portugal and OECD average · 2025

Copper marks Portugal; blue marks the OECD average. The category “learning” is the same measure highlighted in the country note.

Checking sources remains the harder question

In the OECD average, 46% of students say they check sources and trust scientific evidence over common sense when they encounter scientific information. Another 37% check sources but still rely on common sense. That distinction matters: finding a source is different from judging whether it is reliable.

Most students report checking sources, but not all then trust the evidence

Self-reported approaches to scientific information · OECD average · 2025

Rounded percentages sum to 99. These profiles are not a direct misinformation test.

What the data can and cannot say: PISA links a reading decline with more hurried, incorrect responses and documents widespread AI use. It does not estimate AI’s effect on reading scores. The questions worth asking next are about reading comprehension, source evaluation, and how students use AI while doing the work of learning.

See the full reading and information section →