Across the United States, school districts are racing to adopt artificial intelligence tools in classrooms, even as rigorous evidence about their effectiveness remains remarkably thin. The result, according to observers, is a landscape that resembles a "Wild West" — a sprawling, largely unregulated frontier where experimentation is thriving but solid research is scarce.
The picture is not uniformly bleak. In some districts, teachers are using AI to draft lesson plans, shaving hours off their planning time. In others, AI chatbots are offering students feedback on their written work, supplementing the attention a single instructor can provide. These applications have generated genuine enthusiasm among educators who see them as practical solutions to real classroom pressures.
But the research community has struggled to keep pace. While anecdotes and pilot programs multiply, peer-reviewed studies that measure how AI affects student learning outcomes remain extremely limited. That gap between adoption and evidence worries some experts, who note that schools are making consequential decisions about children's education based on marketing claims and early impressions rather than controlled trials.
One of the more unusual experiments unfolded last school year in Dayton, Ohio, where a small group of students at the Alternative School at Jackson Center spent several weeks conversing with a social-emotional learning chatbot named Jordan. The school is part of Dayton Public Schools.
Jordan is not designed to be endlessly agreeable, as many commercial AI assistants are. Instead, the chatbot adopts a distinctly angsty persona. During one visit by a reporter, Jordan asked students: "Lately, I've been feeling like what's the point of any of this? I'm not failing. I just don't care. You ever feel like that?"
The approach is intentional. Students at this alternative school often arrive with behavioral and academic struggles; many would otherwise be suspended or expelled. School staff believed that a chatbot expressing its own doubts and frustrations might give students a safe, low-stakes way to practice empathy and reflection.
For at least one student, the strategy appears to have landed. A sophomore identified only as K described Jordan with affectionate exasperation. "Jordan's a little whiner, but I always help him out," K said. The student added that the conversations felt less like interacting with software and more like texting a friend — a relationship that formed over repeated exchanges.
NPR agreed to use only K's first initial because of the stigma that can accompany attending a school known for disciplinary problems. That sensitivity underscores how carefully schools must handle AI tools that touch on students' emotional lives, an area where privacy and vulnerability are heightened.
The Dayton district paid the education technology company SchoolAI $72,600 during the last academic year for a bundle of AI tools that included the Jordan chatbot. The expenditure illustrates how quickly AI vendors have moved into the education market, offering districts packaged products that promise to address everything from tutoring to mental health support.
Yet the financial commitment comes with little independent verification. Districts buying these tools typically rely on vendor demonstrations, limited pilot feedback, and the intuition of teachers and administrators rather than large-scale, longitudinal studies comparing AI-assisted instruction with traditional methods.
Teachers who use AI for lesson planning describe clear time savings, but the quality of AI-generated materials varies. Some educators report that the tools produce solid starting points requiring modest revision; others find they must heavily edit or discard outputs entirely. The net effect on instructional quality — as opposed to preparation time — remains an open question.
Student-facing applications raise a different set of concerns. When chatbots give feedback on writing, they can reinforce errors or offer generic praise that does little to improve skills. When they serve as emotional companions, as Jordan does, they raise questions about whether simulated relationships substitute for or complement human connection.
Researchers also caution that AI tools trained on broad datasets may embed biases or fail to account for the specific needs of students with disabilities, English learners, or those in under-resourced schools. Without careful evaluation, these gaps could widen existing inequities rather than close them.
The scarcity of research is partly a function of speed. AI products evolve faster than the academic publishing cycle, meaning that by the time a study is completed, the tool it examined may have been updated or discontinued. That mismatch makes it difficult for districts to demand evidence before purchasing.
Some schools are attempting to fill the void themselves, running small internal pilots and collecting their own data. These efforts can be valuable, but they are rarely designed with the controls needed to isolate AI's effects from other factors, such as a new teacher or a change in curriculum.
For now, the landscape remains one of scattered experiments rather than settled practice. Districts are learning as they go, weighing enthusiasm from students and staff against risks they cannot fully quantify. The Dayton chatbot, for instance, produced at least one promising connection, but its broader impact on attendance, behavior, or grades is unknown.
The coming years will likely bring more scrutiny. Parents, advocates, and policymakers are beginning to ask harder questions about what AI is doing in classrooms, who benefits, and at what cost. Until the research catches up, schools will continue navigating what many describe as uncharted territory — hopeful about the potential, but exposed to the possibility of unintended consequences.
(0)