Portugal’s PISA gains have largely faded
Scores peaked around 2015. The 2025 results show where the decline is largest and what the proficiency data add.
Read the Portugal story →OECD PISA 2025 · released 8 September 2026
Portugal’s 2025 scores are close to its 2006 results. Reading and maths have fallen most since 2015. Science has also slipped, while staying closer to its earlier level.
Scores peaked around 2015. The 2025 results show where the decline is largest and what the proficiency data add.
Read the Portugal story →PISA records a sharp rise in fast, incorrect reading responses. AI use is already common in Portugal, but the survey cannot connect it to the score decline.
Read the reading and AI story →01 · The long view
The line shows Portugal against the OECD average in every available cycle. The score is a scale score for 15-year-olds, not a percentage correct. Reading starts in 2000, mathematics in 2003 and science in 2006.
Portugal and OECD average · PISA scale points
02 · The comparison
Portugal is close to the OECD average in all three subjects. The chart places 95% score-estimate intervals beside each score, so a one-place rank difference does not look more exact than the data allow. These intervals describe score estimates, not rank intervals.
Top 10 participating systems plus Portugal · mean score and 95% score-estimate interval
| Country/economy | Score | 95% score interval | Rank |
|---|
Rankings are indicative. Score intervals do not translate directly into rank intervals.
03 · The recent turn
This view compares 2022 with 2025. The headline counts systems with lower, similar or higher average scores; the chart highlights the largest movements, Portugal and the OECD’s 35-system comparable benchmark. Direction counts are descriptive and do not test statistical significance.
Change in PISA score points · 2022–2025
| Country/economy | 2022 | 2025 | Change |
|---|
Includes systems with comparable 2022 and 2025 estimates in the OECD table. Changes are raw score-point differences; some may not be statistically significant.
04 · The distribution
Mean scores compress a distribution into one number. These bars show how the share below Level 2 and the share at Level 5 or above changed between 2015 and 2025. The comparison is in percentage points.
Portugal · percentage of students · 2015 and 2025
Below Level 2 is the OECD’s basic proficiency threshold. Level 5 or above denotes high-performing students.
| Subject | Below L2, 2015 | Below L2, 2025 | Change | Level 5+, 2025 |
|---|
05 · Inside Portugal
That gap is larger than the difference between many countries near Portugal in the international table.
The OECD estimate describes an association with socio-economic status. It does not identify a single cause.
National rankings describe system-level averages. They do not show how evenly results are distributed inside a country.
06 · Classroom context
The digital measures are self-reported. The adjusted curves account for socio-economic status, yet they remain associations. They cannot tell us that a device, an app or AI caused a score to rise or fall.
Adjusted science score by reported learning-device use at school
The curve is highest in the middle categories. It is a descriptive adjusted association, not a recommended dose.
Portugal reports fewer learning-device hours at school, slightly more leisure hours, and more classroom distraction than the OECD average.
Share reporting selected weekly uses · Portugal and OECD average
Portugal is slightly above the OECD average on these reported uses. Frequency is not a measure of learning gain, quality or independent work. Source: OECD, PISA 2025 Results, Volume I and Portugal country note.
Adjusted odds ratio for distraction in most or every science class · reference line = 1
Both estimates are below 1 after accounting for students’ and schools’ socio-economic profile. This is a cross-sectional association. OECD found that changes in system-level phone bans were not associated with reading or science score changes between 2022 and 2025. A policy association with distraction does not establish an effect on achievement. Source: OECD PISA 2025, Table I.B1.4.51 and chapter on school life.
07 · Reading in a digital world
The OECD reading analysis finds both falling scores and more answers delivered too quickly and incorrectly. PISA also asks how students evaluate scientific information. These results raise a sharper question for an AI-rich classroom: can students check what they read? They do not measure AI’s effect on learning.
Share of all student responses · OECD PISA reading assessment · 2018 and 2025
This metric counts responses, not students. It is distinct from the OECD’s “hasty readers” student classification. In the same period, the OECD average reading score fell 25 points. Source: OECD, reading trends 2018–2025, Table I.A1.12.
OECD students’ reported approach when they encounter scientific information · 2025
The remaining profiles reveal a gap between checking and evaluating: 37% check sources but still rely on common sense; 11% do not check sources but trust scientific studies; 5% do neither. Percentages may not total 100 because of rounding. These self-reported profiles are not direct tests of misinformation detection. Source: OECD, PISA 2025 Results, Volume I.
What we can conclude: information literacy is a real educational challenge, while the available PISA data do not show that AI caused the reading decline. The strongest evidence here is the simultaneous rise in rushed incorrect responses and reported AI use, not a causal link between them.
08 · School experience
PISA also asks students about home and school life. These figures add context to test scores, though they do not explain score changes by themselves.
Share who say a parent or guardian does this at least weekly · Portugal · 2022 and 2025
Asking about the school day fell from 84% to 73%; discussing school problems fell from 69% to 53%. The surveys represent different cohorts of 15-year-olds. Source: OECD, Portugal PISA 2025 country note.
Any reported bullying · Portugal and OECD average · 2025
Portugal’s reported rate is about 5 percentage points below the OECD average. This is self-reported exposure and does not say how severe or persistent each student’s experience was. Source: OECD, PISA 2025, Figure I.4.29.
Difference in science score associated with each student-reported family-support item · OECD average · adjusted for socioeconomic profile
For example, students who said family members asked what they did at school scored 26 points higher on average, after adjustment. These cross-sectional score differences are associations, not gains caused by a conversation. Source: OECD, PISA 2025, Figure I.4.25 and Table I.B1.4.124.
Science score-point difference associated with reported bullying · OECD average · adjusted for socio-economic profile
Among the examples shown, students who said upsetting information about them had been published online without consent scored 48 points lower on average. This association does not establish that bullying caused the score difference; other factors may remain. Source: OECD, PISA 2025, Figure I.4.31 and Table I.B1.4.144.
Portugal’s context is mixed: lower reported bullying than the OECD average sits alongside a marked fall in weekly family-school contact. The dashboard can show these changes and associations; it cannot establish that either explains Portugal’s score trajectory.
09 · Beyond test scores
That compares with 62% across the OECD. Students were asked whether they would like to contribute to protecting the environment in their future job. This is an aspiration, not a forecast of career choices. Source: OECD, Portugal PISA 2025 country note.
Share interested in environmental protection in a future job
A separate OECD survey of 15-year-olds connects extracurricular participation with reported social and emotional skills. It also shows that opportunities differ across students and places. These are contextual findings, not PISA scores or evidence that activities cause stronger skills.
Standardised regression coefficients · average and range across participating sites
Each dot is the cross-site average coefficient; each whisker spans the lowest to highest site estimate. These are coefficients, not PISA points, and the whiskers are not confidence intervals. The OECD models control for gender, socio-economic and migrant status. Emotional control has the smallest average coefficient and was associated with participation in 8 of 15 sites.
Source: OECD, SSES 2023, Figure 2.12 and Table A2.11 · download the underlying table.
Average participation rate and range across sites · percent
The dot is the average participation rate; the line spans the lowest and highest site rate. Participation and activity offers vary between sites.
Source: OECD, SSES 2023, Figure 2.13 and Table A2.9 · download the underlying table.