What’s the best way for my grade 5 students to learn how to add fractions?
This is the question I put into three of today’s most advanced AI tools. The responses were impressive: quick, clear lesson outlines with practice problems, examples, and explanations. (Here are the responses from ChatGPT, Claude, and Gemini)
But as I read through them, I noticed something troubling: Even though I asked about the best way students learn, every response centered on how the teacher could deliver the concept, not how students could discover or construct their own understanding. The AI defaulted to the traditional, teacher-directed model with differentiation as a passing afterthought. This is the very model research has shown is less effective for deep, lasting learning.
A recent study, “Pedagogical Biases in AI-Powered Educational Tools” (Chen et al., University of Pennsylvania & University of Michigan), highlights this issue (read study). Analyzing 90 AI-generated lesson plans, the researchers found most reinforced teacher-centered instruction, offering little room for student agency, problem-solving, or meaningful dialogue. Instead of supporting modern, student-centered learning, these tools mirror outdated practices.
This should surprise no one. The internet is replete with sample lessons and worksheets from thousands of different sources. It follows then that AI models trained on these sources will likely create more of the same.
And the problem is already in classrooms, because AI is.
Surveys show over 60% of teachers are experimenting with generative AI to draft lesson plans, worksheets, or assessments. While this promises efficiency, there’s a risk: if AI helps teachers do more of what they’ve always done, it locks schools into instructional models we already know fall short.
Post-pandemic, student outcomes haven’t recovered. Educators face mounting pressure to close gaps and engage diverse learners. Falling back on traditional, “sage on the stage” approaches is not the answer. We need strategies that center students as active participants in their learning: building agency, critical thinking, and collaboration.
To be clear, AI has tremendous potential. But its impact depends on intentional design. Large language models like ChatGPT are trained on vast internet content, much of it reflecting outdated pedagogical norms. Unless specifically guided otherwise, these systems replicate the past.
That’s why, at engage2learn, we took a different approach with the AI behind GroweLab, our coaching and talent development platform. Rather than relying on general models, we built our system on curated instructional practices grounded in research. Our AI supports teachers in scaffolding student inquiry, promoting goal-setting, and encouraging rich dialogue. It is built to help teachers do what works best for students, not to replace them.
Educators have seen waves of technology come and go, many overpromising and underdelivering. We cannot unleash AI into the classroom without ensuring it is improving teaching practice, meaning it’s time for district leaders to ask tough questions:
- Are the tools we adopt aligned to what research shows works?
- Do they promote student agency and critical thinking?
- How will we measure their impact on teaching and learning?
AI is not inherently flawed. Without thoughtful design, though, it will reproduce yesterday’s problems and biases. Educators and developers share the responsibility to make sure AI moves us toward more equitable, student-centered learning instead of back into ineffective patterns.
Let’s make sure the AI we bring into schools is aligned to the best research on teaching and learning, and that it helps teachers reach every student instead of automating old routines. How we choose to use these tools will decide what they do for learning.
First published June 5, 2025, by Jill Galloway.