A recent report on how U.S. states and territories are evaluating AI found that most states are still in the early stages of developing a systematic approach to AI evaluation. Or as an AI chatbot might put it, they are still “thinking.”
The annual report released by Digital Promise looked at publicly available state-level AI evaluation guidance in 32 states and Puerto Rico, and identified three evaluation stages: Nascent & Exploratory, Emergent & Piloting, and Systematic & Evidentiary.
“Mostly we saw states in the nascent exploratory stage of the progression that we developed,” says Pati Ruiz, a co-author of the report and Senior Research Scientist for Learning Sciences and Emerging Technologies at Digital Promise.
Ruiz cautions that not every state they looked at was up-to-date with their public AI evaluation policies, and that it’s all changing fast. “We know that these documents get regularly updated by states, and so are hoping in 2026 to start seeing more states moving through the categories that we developed into the more systematic and evidentiary section,” she says.
In the meantime, Digital Promise’s report on AI evaluation in U.S. States and Territories provides a snapshot of where AI evaluation is in most states currently, and offers a closer look at states that have moved further along in their AI evaluation efforts. The report also provides guidance on what successful AI evaluation can look like.
The Importance of Tracking AI Use and States That Have Announced Systems for Doing So
Jeremy Roschelle, Digital Promise’s Director of Learning Sciences Research, says the report also highlights the states that have developed detailed AI plans and are gathering evidence about how and where it is effective. This gives other states and districts a model to look to, and points them toward those from whom they can seek guidance. Roschelle adds that this type of work is vital because “evidence is going to be an important part of this AI revolution.”
To that end, the report highlights Colorado and Louisiana, which are both states that have done more to study how AI is used in schools.
“In Colorado, they are documenting their use of qualitative and quantitative metrics on AI, access engagement, and understanding educational outcomes by student demographic to understand how their learners are leveraging AI-enabled tools,” Ruiz says. “In Louisiana, they’re also tracking student progress and measuring the effectiveness of these AI tools and resources.”
What Does A More Evolved AI Policy Look Like
Ruiz notes that states need to provide schools with guidance and suggested guardrails for classroom and administrative AI use.
“We’re finding it incredibly important for classroom teachers to also have good examples of appropriate, responsible use of AI in their classrooms, and then for educational leaders to have good examples of how they might integrate AI into their workflows,” she says. “We’re seeing an important call for more professional learning at all levels of the ecosystem, so that we are all developing AI literacy skills to be able to identify when it’s appropriate to use AI-enabled tools and when it’s not.”
Ruiz adds that it’s important for feedback loops and collaboration to be part of the development of AI policies as well as usage.
Both Ruiz and Roschelle emphasize that all this work should be done with an eye toward research and data gathering to inform future decisions.
“The reality is it’s not easy to evaluate AI, so it’s reasonable that it’s going to take some time to get it right, but if we don’t start now, we’ll never get there,” Roschelle says.
What Does Good AI Use In Schools Look Like?
The best use of AI in schools is still being studied and pioneered, but some guiding principles have already emerged. Roschelle sums these up in two words: student learning.
“We shouldn’t be using AI because it’s ‘the future’ or using AI because it’s exciting and cool,” he says. “We should be using it because students are learning something. And I just really fear that it can get too easily overlooked.”
Roschelle says AI also should not be used as a fun gimmick.
“It shouldn’t be today’s filmstrip,” he says. “They used to show us filmstrips to amuse us. And, there’s been a series of things that are just ways to give kids a break from school but don’t really contribute to learning.”
Roschelle adds that that shouldn’t happen with AI. “This is a powerful and risky technology,” he says. “It should be used where it really makes a difference.”