A new randomized controlled trial out of Texas A&M University gives some of the first hard classroom evidence for a distinction we built EducAIte around: the difference between AI that answers and AI that asks.
Researchers split 90 tenth graders into three groups. One continued with regular coursework and had no argumentation activity at all, serving as a baseline. A second completed a proven inquiry-based teaching method: writing a scientific argument and revising it based on feedback from a peer. A third completed that same task, but received feedback from an AI designed around Socratic questioning, prompting students to justify their reasoning rather than handing them conclusions.
The AI-scaffolded group showed the strongest gains in critical thinking of the three groups, more than double the improvement of students who completed the same task with peer feedback alone. They also led in scientific reasoning, confidence, and engagement.
This is exactly the distinction EducAIte is built around. Most students' everyday experience with AI, whether that's Gemini, ChatGPT, or any other assistant, is asking a question and getting an answer. That's fast, but it's the opposite of what this study shows actually builds thinking skills. Instead of a student typing a question and getting a solution, with EducAIte they are guided through the reasoning with follow-up questions, until they arrive at the answer or write their paper themselves.
This study also points out that learning often requires individualized & differentiated feedback which is difficult to accomplish when you have a classroom of 25-35 students. LLMs & GenAI have the unique opportunity to provide scalable adaptive instruction if the model is built around 'socratic dialogue' in the way that EducAIte approaches student interactions. The part that excites me the most is that AI-supported students reported meaningfully higher self-efficacy (i.e., their confidence in their own ability to handle the material) than either the peer-feedback group or the no-activity group, neither of which showed a real gain in confidence at all. The AI-supported group also showed by far the largest increase in cognitive engagement (i.e., how focused and invested they felt in the work) more than the peer-feedback group and far larger than the increase seen in the group with no activity.
One caveat worth sitting with is that the AI-scaffolded group in this study didn't show growth in a student's ability to manage their own learning without outside help. This is the mental habit of stepping back and asking "wait, does this actually make sense to me?" The researchers suspect the AI may have been carrying some of that reflective load for students. That's precisely why scaffolding needs to fade over time rather than run at full support indefinitely, a design principle built into how EducAIte adjusts guidance as students demonstrate independence and why EducAIte teaches students to understand why an answer makes sense.
This study points at something we think should matter a great deal for how schools evaluate AI tools going forward, which is whether it's built to make students think like EducAIte does. Policy is shifting fast on this. Some of the largest districts in the country, including New York City, have recently moved to restrict broad AI access for younger students amid real concerns about companion chatbots and unsupervised use. That caution is understandable. But it also points to an opportunity: districts don't need to choose between banning AI and handing kids an open chatbot. A governed, Socratic-scaffolded model, like EducAIte's, is built for exactly that middle ground.
Want to give EducAIte a try? Reach out to erica@educaitelearning.com.
Source: Kao, S., Grant, P., & Woltering, S. (2025). Socratic AI in K–12 Science Classrooms: Effects on Critical Thinking, Motivation, and Self-Regulation in a Randomized Controlled Trial. Research Square preprint. https://doi.org/10.21203/rs.3.rs-8118546/v1
*Worth noting this is a preprint (not yet peer-reviewed) from a single high school, so it's early evidence and not a final verdict. However, it's a well-designed, randomized study.
Want to give EducAIte a try? Contact Erica Bishaf at erica@educaitelearning.com, or contact us here.