Google's education and AI research divisions released new data evaluating how the Gemini AI platform functions in European classrooms. The trials, conducted over the past school year in the United Kingdom and Italy, examined how AI-driven math tutoring impacts students and how AI-generated lesson planning affects teachers. The results show that while AI can offer instructional support, human supervision is necessary to prevent errors and ensure student progress.
What Happened
According to the Google Education announcement, the trials tested different applications of Gemini in two countries. In the United Kingdom, researchers analyzed one-on-one AI math tutoring. They found that Gemini-driven tutoring offered step-by-step math support that performed similarly to traditional, human-authored materials when monitored by teachers.
In Italy, teachers used Gemini to design interactive lesson plans. The Google Education blog reports that teachers using the AI tool noticed higher student participation and more peer collaboration. These teachers also reported lower daily stress levels.
The Bigger Picture
While Google's findings show potential benefits, independent academic research reveals a more complex reality. For instance, while AI can generate immediate math hints, reliability remains a persistent challenge. A study published in PLOS One confirmed that AI-generated help produced math learning gains equivalent to human-authored materials. However, the study also revealed that the AI initially failed basic math quality checks on 32% of algebra and statistics problems, showing the danger of using unmonitored software.
Purely automated systems also struggle to keep students motivated over time. A longitudinal study published in Artificial Intelligence in Education compared seventh-grade students using AI-only tutoring against those receiving a hybrid human-AI tutoring model over an entire school year. By the end of the year, the hybrid group was 0.36 grade levels ahead of the AI-only group. The researchers found that human interaction increased the software's effectiveness and cautioned against using AI as a standalone tutor. This need for human oversight aligns with new teacher training standards. As we previously reported, future educators are now learning to evaluate and audit real-time classroom AI software.
The claim that AI reduces teacher stress also faces academic skepticism. A systematic review published in Acta Psychologica warns that while generative tools promise efficiency, they often lead to heavier academic workloads. This occurs because teachers must spend substantial unpaid hours fixing mistakes and adjusting AI outputs. According to a study in Computers and Education Open, the effectiveness of AI in lesson preparation depends heavily on structured training. Without proper training, teachers struggle to deal with what researchers in Learning, Media and Technology call the "educational frailties" of AI tools, which often creates more administrative work.
Safety remains a primary concern for school administrators. AI models generate text by predicting the most likely next word, rather than verifying factual truth, as explained in a guide by the University of Michigan. This calculation process leads to errors, such as fabricated historical facts or incorrect math guidelines. To prevent these errors from reaching children, school leadership must establish strict K-12 guardrails and student data protections. According to SchoolAmplified, schools should treat all AI outputs as unverified drafts and implement formal verification workflows. This ensures no automated claims are shared without explicit human validation.