Title: Large Language Models in K-12 Education: Alignment with State Curriculum Standards and Student Personas

Abstract:

As Large Language Models (LLMs) gain traction in educational environments, their ethical deployment has sparked significant concern. The rapid enhancement of publicly accessible online chatbots in terms of accuracy and capability has driven broader adoption, particularly among students seeking assistance with academic assignments. Consequently, it is imperative to assess whether these models adhere to established educational standards. Since curriculum frameworks in the United States are determined at the state level, there are substantial disparities across states regarding content requirements, emphasis, and narrative perspectives.

This study introduces an LLM-driven pipeline designed to detect variations in U.S. History curricula among different states and to measure how well various LLMs mirror these state-specific educational nuances. Furthermore, we performed controlled experiments that modified user personas by specifying attributes such as geographic location, grade level, gender, and race. This approach allowed us to evaluate how sensitive LLM responses are to such user characteristics.

Our analysis reveals that while models can modify their presentation of historical subjects, these adjustments appear to stem from the perceived political leanings of states rather than actual curriculum content. Conversely, the models demonstrated a successful ability to adapt to a student’s grade level, yet exhibited minimal bias related to race or gender. This suggests that LLMs can effectively tailor their output to student personas with limited demographic prejudice. Collectively, these results underscore potential risks to student learning outcomes associated with unrestricted access to LLM chatbots, particularly due to misalignment with state curriculum standards, and emphasize the urgent need for more robust alignment strategies.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC