Vermont School AI Guidance Limits K-8 Chatbots but Skips Tracking

A recommendation without a record

Vermont has said clearly that it doesn't want general-purpose chatbots in younger students' hands. The state has not tried to find out whether schools follow that advice. Guidance from the Agency of Education says children through eighth grade should not use ChatGPT-style tools. Yet the agency cannot tell parents which schools have turned on Google's Gemini chatbot for students, because it does not collect that data.21 When asked which school systems allow Gemini, and for which grades, the agency answered that it does not "collect this level of specificity."21

That answer sums up how Vermont is handling AI in schools. The state publishes detailed advice and then leaves the decisions to 54 local school systems. Most of those systems had no written AI policy as recently as the 2023–24 school year.21 Meanwhile, a growing body of research suggests the difference between an open chatbot and a structured tutor matters a great deal. Treating that difference as optional local preference is a bigger risk than Montpelier admits.

What the guidance actually says

The agency issued the 50-page document in late January and called it a practical framework to support local decision-making.1 The grade bands are specific. Students in pre-K through second grade should not use AI chatbots at all. Third through fifth graders should avoid open-ended chatbots and use AI only when it is built into curriculum-specific applications.2 In grades six through eight, students may use education-specific chatbots under a teacher's direction. The focus should be on tools that don't give them access to generic chatbots outside class.2 Students start using AI as a "cognitive assistant" in ninth grade.2

The reasoning is about development. The document warns that younger children may treat AI tools as if they were human. It also says tools that write far beyond a child's ability can let them skip "the essential struggle" that builds core literacy.2 For middle schoolers, it warns that unstructured use can become a crutch.2 The guidance names ChatGPT, Claude and Gemini as general-purpose tools and says the agency recommends that students in K-8 not have access to them.22

The guidance is also explicitly not binding. Districts are told they "should feel empowered" to change the timeline, and the grade table is described as "one possible framework for discussion."2 Every outlet covering the rollout described it the same way. Kids VT reported that Vermont has no official statewide restrictions on AI in education and that the document "stops short of imposing restrictions."10 Early TV coverage presented the release as boundary-setting, with Secretary Zoie Saunders saying it was the state's responsibility to set limits.8 There's some tension between those two descriptions. The limits exist on paper, but nothing makes districts follow them.

Why the Gemini gap matters

The untracked Gemini settings are not a small detail, because of how Google's tools reach classrooms. The state's own guidance says every Vermont school system uses Google for Education. Google gives Gemini and NotebookLM to schools for free and builds them into Classroom and Docs.22 In August, Google opened Gemini inside Google Classroom to students of all ages, including elementary students. Google says the Classroom setting is on by default, but students get it only if their district has also enabled the Gemini app or NotebookLM for them.21

In practice, this means the line between "general-purpose chatbot" and "education tool" is now set by an administrator's settings, not by which product a school buys. A district that turned on Gemini for older students and left the defaults alone elsewhere could be outside the state's recommendations without realizing it. Nobody at the state level would see it either.

The guidance itself pointed out a second blind spot. When it was released, Google's auditing features were limited and required a Google administrator. Gemini used inside Docs and other apps was not logged at all.22 So teachers could not review all of their students' chatbot conversations.21 For a policy built on teacher-directed use in middle school, that weakens the main safeguard.

Data privacy coverage is also uneven. In the most recent state survey, 16 of the 54 school systems did not require their software vendors to sign student data privacy agreements.21 The guidance says districts should get vendor promises not to train AI models on student chats.22

The state's case for staying flexible

The agency's argument isn't unreasonable. A spokesperson said AI changes quickly enough that formal policy would make it impossible to update the rules fast enough.21 The guidance tells schools to pilot tools with a few classes before larger rollouts.2 Education tech program manager Josh Blumberg described it as "a starting point for collaboration across districts."1

The results so far are mixed. In the 2024 technology survey, about five of 54 systems had issued any AI policy or notice. Forty-seven called their approach cautious, and 34 said teachers were already using AI in instruction.21 Some districts have done serious work. The guidance cites Champlain Valley, Essex Westford and Missisquoi Valley as examples.21 Champlain Valley Union High School's English department, for instance, uses a red-yellow-green system for student AI use.10 Earlier reporting found that districts with dedicated tech specialists were better prepared than smaller rural ones. That gap is exactly what a hands-off approach tends to make worse.

Vermont has turned guidance into law before. Fewer than half of schools had a cellphone policy in 2023–24, and lawmakers then passed Act 72, which requires every board to adopt a policy at least as strict as the state model.21 Senate Education Chair Seth Bongartz has said his committee will likely take up AI next session. He wants a model policy local schools can adapt, though he has said he is not prepared to support a blanket ban.21 The agency has not committed to writing one.21

What the research says about tutors and chatbots

The strongest argument for taking the state's grade lines seriously comes from the research. In a randomized trial with nearly 1,000 Turkish high school math students, those using a standard ChatGPT-style interface did 48% better on practice problems. But they scored 17% worse on a later exam without AI than students who never had access.12 A version built to give hints instead of answers mostly removed that harm, but it did not produce gains.12 The students also didn't notice the damage: those using the open tool did not think they had learned less.16 A recent scoping review of K-12 risks cites that "learning penalty" and connects it to cognitive offloading and an inflated sense of mastery.17

The evidence does not all point one way. A Harvard physics trial found that a carefully designed AI tutor produced more than double the median learning gains of an in-class active-learning lesson, in less time.13 A meta-analysis cited in the same scoping review found positive effects overall, strongest when AI was built into teacher-supported activities.17 Against that, a new randomized trial of 2,379 undergraduates found that access to a course-integrated AI tutor lowered final grades by up to 0.37 standard deviations. Losses for first-generation students were more than twice as large.14

Taken together, these studies don't show that AI in schools is good or bad. They show that results depend on design and supervision. Open chatbots that hand out answers tend to hurt learning. Tightly structured tutors used with teacher support sometimes help. Even well-meant tutors can push out other learning activities. Vermont's grade bands follow that distinction closely, keeping open tools away from younger students and favoring constrained, teacher-led use. That is the strongest reason the state should know whether districts follow them.

Academic integrity without detectors

Cheating is what worries local leaders most. In the state survey, 40 of 54 systems said they were concerned about students using AI to cheat.21 The agency's own workshop materials call academic integrity the top concern school leaders raise. They admit there is "no magic bullet" and that most students can easily reach AI outside school systems.9

The guidance's answer is mainly about teaching, not technology. It tells schools not to rely on AI detection tools, calling them unreliable, prone to false positives and easy to get around. It urges teachers to grade "process" over "product" and to respond to suspected misuse with curiosity, not accusation.222 That is sound advice. But it depends on knowing which tools students can reach on school accounts, which is the information the state doesn't collect.

Vermont compared with other districts

Vermont sits between the more aggressive approaches elsewhere. Boston is teaching AI in every high school, while New York City and Los Angeles restricted student use this fall.21 New York's version bans companion chatbots in all grades and removes other student AI tools through eighth grade for 2026–27 while the city studies the effects. On paper, Vermont's K-8 line is close to New York's. The difference is that New York enforces its line, and Vermont leaves its line to each district.

The state is right that AI changes faster than statutes. But updating rules quickly and knowing what's happening in schools aren't in conflict. A one-line survey question about which grades have Gemini turned on would not lock the agency into anything. Until the state asks, its eighth-grade line is only a recommendation, and parents have to find out from their own schools whether it's being followed.