Evaluating ChatGPT’s Memory and Conversation Continuity: Insights from Academic Prompting

Abstract

As ChatGPT becomes a prominent tool in education, a growing challenge emerges: should educators start a new chat for each task, or continue an existing thread to retain context? This decision has become particularly pertinent with the advent of ChatGPT’s memory feature, which allows the system to remember user information and past interactions across sessions. While this capability introduces opportunities for efficiency and personalisation, it also presents risks such as context drift, irrelevant recall, or unintentional bias especially in academic settings that demand precision and adaptability. This study explores the implications of ChatGPT’s memory feature through structured and emergent prompting exercises conducted with five academics. Each academic was provided with the same sequence of prompts, addressing common academic tasks including assessment design, feedback generation, and curriculum alignment. Their interactions were observed under two conditions: (1) continuing conversations with memory enabled, and (2) starting fresh sessions without memory. The authors analysed these interactions to assess accuracy, coherence, responsiveness, and relevance. Reflections based on this analysis suggest that continuing conversations allowed for more nuanced and context-aware responses, particularly valuable for iterative educational tasks. However, these benefits were occasionally offset by outdated or misplaced contextual references that hindered objectivity and clarity. Conversely, fresh chats supported cleaner outputs in tasks requiring analytical rigor or neutrality. The study suggests that educators must adopt a context-sensitive approach when engaging with ChatGPT, selectively leveraging memory for continuity-driven tasks while favouring new chats for objectivity-critical outputs.