As artificial intelligence tools like ChatGPT, Google Gemini, and Microsoft Copilot become fixtures in K-12 classrooms, school and district leaders face a new and largely invisible policy question: what is the environmental cost of student AI use, and what can districts do about it? This interactive calculator, developed by Dr. Seth B. Hunter at EdPolicyForward: The Center for Education Policy at George Mason University, is designed for education policymakers, district administrators, curriculum directors, and sustainability officers who want to move beyond awareness and into action.
With just a few clicks, users can configure the tool to reflect their own district’s enrollment, AI adoption rate, and model complexity, along with the electricity grid serving their AI vendor’s data centers. They can then explore how much additional energy and carbon emissions student AI use adds on top of existing digital infrastructure.
The tool’s central finding is that model selection matters more than usage volume. A standard chatbot query now uses roughly the energy of a web search, while a reasoning or “thinking mode” query can use more than a hundred times as much. That makes model right-sizing the highest-leverage clause a district can write into an AI vendor contract.
The tool also projects emissions trajectories through 2031 under business-as-usual versus policy-intervention scenarios, drawing on federal data, peer-reviewed research, and the environmental disclosures AI providers have published. Potential uses include briefing school boards on AI’s environmental footprint, informing AI procurement and vendor contract language, supporting district sustainability planning, and anchoring professional development conversations about responsible AI use in schools.
Estimates are order-of-magnitude rather than an emissions inventory. The tool documents its sources, assumptions, and limitations in full, including what it does not count: model training, hardware manufacturing, and water use.
Version 2.0. Data current as of September 2026. Per-query AI energy estimates are changing quickly, so check the version date before citing any figure.
Try the tool here – https://claude.ai/public/artifacts/4a76bfbf-5f29-4d90-ab06-6c8a27bcaacb.
A NOTE ON VERSION 2
The link above points to version 2 of this tool, which produces different results than version 1. The change reflects genuine advances in how AI energy use is measured and managed rather than a correction to the earlier arithmetic.
Three developments drove the revision. First, AI providers began publishing production measurements of per-query energy for the first time in 2025, and those measurements came in well below earlier published estimates. Google reported that the energy behind a median text prompt fell by a factor of 33 over a single year, the result of more efficient hardware, better models, and improved data center operations. Version 2 estimates a standard chatbot query at roughly one-sixth the energy version 1 assumed.
Second, the picture is not uniformly improving. Reasoning models, which work through problems step by step, use far more energy per query than standard models. Version 2 treats these as a separate category and estimates them substantially higher than version 1 did. The net effect on any given district depends on which models its students actually use, which is why model selection is now the tool’s central theme.
Third, the underlying reference data was refreshed: current EPA grid emission rates, current federal enrollment counts, and 2026 survey data on student AI adoption, which has roughly doubled since the figures version 1 relied on.
Users who ran version 1 should re-run their district in version 2 rather than assuming earlier results still hold. Version 2 documents every constant, its source, and its vintage, so any result can be traced or reproduced.