Beyond Speed: Peru’s AI Experiment Shows Why Teacher Skills Matter More Than Technology Access

Peru’s AI experiment shows that teacher training improves prompt quality and judgment but does not significantly reduce time spent on core teaching tasks. The findings suggest governments should pair AI access with skills training, professional oversight and evidence of student-learning gains before scaling investment.

Peru's experiment with generative artificial intelligence in public schools is challenging a popular assumption about AI: that its biggest benefit is making workers faster. A randomized study by researchers associated with the World Bank's Development Research Group and Education and Skills Global Department, the University of Wisconsin–Madison and the University of Toronto found something different. AI training did not reduce the time teachers spent on their main duties. Still, it significantly improved their ability to communicate with AI, detect weak responses and use the technology for professional purposes.

The experiment involved 1,269 sixth-grade teachers across 390 public schools in Metropolitan Lima. Of these, 196 schools with 660 teachers were assigned to the treatment group and 194 schools with 609 teachers formed the control group. The intervention combined one day of in-person training, access to Microsoft Copilot and eight months of follow-up through WhatsApp.

AI Improves Teacher Skills, but the Clock Barely Moves

The programme produced a clear increase in AI adoption. Copilot activation increased by 20.4 percentage points, while teachers accumulated an additional 3.15 active weeks on the platform during the 28-week programme. The share of teachers identifying Copilot as their most frequently used AI tool increased by 29.3 percentage points from 9.6% in the control group.

Yet greater AI use produced no significant time savings across seven major professional activities, including lesson planning, differentiated instruction, assessment design, grading, collaboration, administration and communication with parents.

The stronger gains appeared in quality. Teachers receiving the intervention improved their prompting performance by between 4.6% and 16.9% across different measures. Overall prompt quality increased by 9.2% using an assessment based on the training programme and by 11.1% under an independent assessment measuring qualities such as clarity, specificity and context.

Trained teachers were also 9.9 percentage points less likely to incorrectly judge the quality of a weak prompt and 6.7 percentage points less likely to accept an inadequate AI response as good.

Peru's Digital Challenge Is Shifting From Access to Skills

For policymakers, the findings suggest that simply providing AI platforms will not guarantee better education. Peru already has relatively high levels of basic digital access among teachers. Data cited by the researchers show that 99% of public-school teachers had cellphones, 81% had laptops or tablets and 88% had mobile data plans in 2021.

More recent evidence cited in the study indicates that roughly 70% of teachers in grades 4, 8 and 11 were already using AI to create mathematics teaching materials, with about 40% doing so for at least half their classes.

The policy challenge is therefore moving from access toward capability. Governments considering AI investments may need to devote greater attention to teacher training, curriculum-aligned prompting, verification of AI-generated information and continuous professional support.

This becomes especially important when AI is used to produce materials for children. A convincing AI response can still contain inaccurate information or inappropriate teaching material, making teachers' professional judgment an essential safeguard.

Older Teachers Show AI Training Can Narrow Skill Gaps

One of the study's most important findings concerns older workers. Teachers above the median age of 54 initially performed worse than younger teachers on measures of AI prompting, but experienced considerably larger gains after training.

Instead of widening the digital divide, structured AI training narrowed age-related differences in prompting skills. Gains were otherwise broadly found across male and female teachers and schools with different performance levels.

For governments and international development partners, this suggests that AI programmes can be designed to support workers who begin with weaker digital capabilities. Training could therefore become an important component of inclusive digital transformation rather than an optional addition to technology procurement.

Development institutions supporting education systems may also need to reconsider how AI projects are evaluated. The number of licences purchased, teachers registered or platforms installed says little about whether technology is improving professional capability. Measures such as prompt quality, error detection, curriculum alignment and eventually student learning could provide stronger indicators of impact.

AI Investment Opens Opportunities but Demands Better Evidence

The shift from AI access toward AI capability could create opportunities for technology companies, EdTech providers and professional-training firms. Governments may increasingly require curriculum-integrated AI tools, teacher-training systems, verification mechanisms, localized educational applications and platforms that help educators develop effective prompts.

But private-sector providers also face a higher evidence bar. Increased platform usage should not automatically be presented as higher productivity or better education. The Peru experiment shows that even when AI adoption rises significantly, measurable time savings may not follow.

The biggest unanswered question is whether stronger AI skills among teachers ultimately improve student learning. The study measured teacher AI capabilities rather than changes in student achievement. Researchers therefore caution that better prompts cannot yet be assumed to produce better classroom outcomes.

For Peru, the next logical step is to examine whether stronger teacher-AI interaction improves lesson quality, differentiated instruction and student achievement. Governments and development partners considering wider deployment could link future investments to evidence from these outcomes before committing substantial resources to nationwide expansion.

The broader lesson extends beyond education. Generative AI may create value in complex public-sector jobs not simply by helping employees work faster, but by helping them perform difficult tasks better. For policymakers, development institutions and private companies, that means successful AI investment will require more than buying technology: it will require building human skills, maintaining professional oversight and measuring whether improved AI capability ultimately produces better public services.