A large international education study has found that students who use AI tools to help with schoolwork generally score lower on standardized tests than those who don't. For small business owners, this isn't just a classroom story—it's a preview of what can go wrong when any organization hands people a powerful shortcut without teaching them how to use it critically.
The data comes from the OECD's Programme for International Student Assessment, known as PISA, which tests 15-year-olds across dozens of countries every few years on reading, math, and science. This year's edition included new questions about AI use in studying. The headline finding: heavier AI use tracked with weaker performance overall.
The nuance matters more than the headline. Students who used AI tools but were also taught how to evaluate the tool's output—checking its reasoning, spotting errors, cross-referencing answers—performed slightly better than non-users. Students who simply accepted AI answers without scrutiny performed worse. The gap wasn't about whether AI was used, but how.
This distinction lines up with what researchers have observed in workplace studies of AI adoption over the past two years. Productivity gains from AI tools tend to appear when users maintain a habit of verification. They tend to disappear, or reverse, when users default to trusting output without checking it. The PISA study essentially found the same pattern in teenagers doing algebra homework.
This fits a broader pattern showing up across AI research in 2024 and 2025: the tools themselves are rarely the deciding factor in outcomes. Training and habits are. Studies on AI-assisted coding, customer service, and legal research have all found similar splits—modest gains for careful users, modest-to-significant losses for people who treat the AI as an oracle rather than a draft generator.
For small business owners, the practical parallel is employee onboarding and skills training. If your team uses AI tools to draft contracts, summarize reports, or learn new software, the PISA findings suggest the real risk isn't the AI itself—it's skipping the step where someone teaches employees to question and verify what the AI produces.
This matters most for businesses using AI to train new hires quickly, a growing trend among restaurants, retailers, and service businesses trying to cut onboarding time. An AI tutor or chatbot that answers a new employee's questions can speed things up, but if nobody checks whether those answers are accurate, mistakes can get baked into how the job is done. The same logic applies to using AI for compliance training or safety certifications, where getting the wrong answer has real consequences.
The fix isn't complicated, but it does take deliberate effort. Building in a verification step—having employees explain why an AI-generated answer is correct, or cross-checking it against a manual or expert—appears to be the difference between AI helping and AI quietly eroding skill-building. That's a five-minute addition to a training checklist, not a major overhaul.
Watch for follow-up research from other education and workforce bodies over the next year, since PISA's findings are likely to prompt similar studies in corporate training and professional certification settings. Also worth tracking: whether any AI tutoring or training-tool vendors start marketing verification features—prompts that ask users to explain or challenge an answer—as a direct response to this kind of data.
The bottom line: the study doesn't say AI tools are bad for learning. It says learning still requires friction—checking, questioning, correcting—and tools that remove all the friction tend to remove some of the learning along with it. That applies as much to a new hire using a chatbot as it does to a teenager doing homework.