Prospective Teachers' Use of AI to Develop Specialized Content Knowledge: A Tool is Only as Good as Its User

While artificial intelligence (AI) systems have been in use since the 20th century, these models have recently gained the interest and use of the larger public with the release of ChatGPT by OpenAI in November 2022. Since then, several AI systems or large language models (LLMs), such as Claude, Copilot, Gemini, Grok, and DeepSeek, have been available for public use. This study examines the use of different AI systems to foster what Ball, Thames, and Phelps (2008) refer to as specialized content knowledge in six prospective primary teachers (PPTs) and six prospective secondary mathematics teachers (PSTs) in geometry. Overall, the AI systems were only able to produce a valid justification on six of 74 requests to justify geometric formulas for an 8% validity rate. This finding underscores the importance of users being cautious of the results produced by LLMs. PSTs correctly identified a valid or invalid justification 39% of the time, while PPTs correctly identified a valid or invalid justification 50% of the time. The implications of these results are discussed.