LearnLM tutoring boosts student problem‑solving by 5.5 percentage points
When a teacher uses LearnLM, an AI-driven tutor, their class seems to edge up on test scores. The experiment wasn’t a fancy new gadget rollout - it was a straight-forward comparison: the same instructor teaching one group the usual way, and the other group with LearnLM hovering in the background. What the researchers really wanted to see was whether students could handle a brand-new problem on their own in the next class, without any hints.
They split participants randomly, half getting the AI-assisted lesson, half not. The key number? The share of kids who solved a novel question unaided after the session.
The gain wasn’t dramatic - just a few percentage points - but it does suggest a small advantage for AI-boosted teaching. If schools start to notice even that modest bump, it might shift budgeting decisions and how teachers think about digital aides. The authors say the next move is to run more randomized trials, maybe with different subjects, to see if the pattern holds.
We also found students tutored by LearnLM were 5.5 percentage points more likely to independently solve novel problems in their next study session, indicating that a teacher using AI tools slightly outperforms a teacher who doesn't use AI. We will be building on this research with further RCTs in the U.S., U.K., India, Sierra Leone and beyond to scientifically validate AI's impact on learning outcomes globally. We're funding organizations that make learning tools more accessible Today, we are providing $30 million in new funding from Google.org over the next three years to support efforts that are focused on driving transformative learning solutions and foundational research.
To kick this off, we're announcing initial funding to organizations who are making AI and tech education universally accessible: - Raspberry Pi Foundation will lead global collaborative projects that shape how students learn to code effectively in the age of AI. - Fab AI will conduct international studies to measure AI's impact on student learning outcomes. - Playlab will build a scalable system to increase AI literacy and equitable AI access in K-12 education by partnering with nonprofits to train teachers and implement AI programs.
With Google's backing, Digital Promise, a global nonprofit working to expand opportunity for each learner, released "A Framework for Powerful Learning with Emerging Technology" to help educators use AI and new technologies in the classroom.
Did the AI-driven tutoring really move the needle? Students who used LearnLM solved new problems about 5.5 percentage points more often than classmates who didn’t, which looks like a modest boost for teachers willing to try AI. The results were shown at the Google AI for Learning Forum in London and have sparked a lot of chatter about how artificial intelligence could change the classroom.
Still, the study was small and the setting very specific, so it’s hard to say if the same lift would show up in other subjects or with younger kids. The researchers say they’ll run more randomized trials soon - that should tell us whether the effect holds up and what side-effects might surface. At the same time, the mix of scholars, teachers and students at the forum felt cautiously hopeful, but kept asking for solid proof.
As the work moves forward, we’ll be watching for data that either backs up or knocks down these early numbers, rather than assuming AI tutoring works everywhere.
Related Reading
Research & Benchmarks Consensus uses GPT-5 and Responses API to speed scientific research
Research & Benchmarks Developers say Sora, unlike Vine/TikTok, is not about people in social media
Research & Benchmarks Google AI Advisors Let Users Probe Performance with Conversational “Why” Queries
Common Questions Answered
What specific improvement did LearnLM tutoring show in the randomized trial?
Students tutored with LearnLM were 5.5 percentage points more likely to solve novel problems independently in the next study session. This modest gain indicates that teachers using AI‑driven tutoring can outperform those who do not.
How was the effectiveness of LearnLM measured in the study?
The researchers conducted a controlled experiment where the same teacher taught two groups: one with LearnLM assistance and one without. They then assessed each group's ability to tackle brand‑new problems without help in the following session.
What future research plans did the authors mention for validating AI's impact on learning?
The team plans additional randomized controlled trials in the U.S., U.K., India, Sierra Leone, and other regions to scientifically validate AI's impact on learning outcomes globally. These follow‑up studies aim to test whether the modest gains observed with LearnLM replicate across diverse curricula and contexts.
What limitations did the article note about the LearnLM study’s findings?
The article highlighted that the sample size and specific classroom context were limited, making it unclear if similar 5.5‑point gains would appear across different subjects or educational settings. Consequently, broader generalizations about AI‑driven tutoring remain tentative.
Where was the LearnLM study presented, and why is that venue significant?
The findings were presented at the Google AI for Learning Forum in London, a prominent gathering of educators and AI researchers. Presenting there underscores the growing interest in how artificial intelligence might reshape instruction and learning assessment.