Observation
Participants and Methodology
Five participants with varying experience levels were observed managing and deleting chats to clear ChatGPT’s memory. The task required them to locate old conversations, delete unnecessary chats, and verify available memory. Performance metrics included time taken, errors made, and frustration levels. Data was collected using time logs, error tracking, and post-task surveys.
Description of Findings
The observations revealed notable usability issues in memory management. Frequent users navigated the system more efficiently but expressed frustration over the absence of bulk deletion features. Intermediate users made errors when distinguishing between chats and reported moderate frustration. The beginner user faced the most challenges, frequently selecting incorrect options and struggling with unclear labels. Across all participants, the lack of visual cues for memory-heavy chats was a common barrier.
Visual Representation of Findings
The data is summarized in three bar graphs, depicting completion time, error rate, and frustration levels for each participant.
A.Completion Time vs. Errors
A scatter plot visualizes the relationship between completion time and errors. It reveals a positive correlation: as completion time increases, so do the number of errors. This suggests that participants who took longer to complete the task were more prone to making mistakes, possibly due to frustration or lack of understanding.
B.Completion Time by User Experience Level
A bar chart illustrates the completion time for each user experience level. As expected, frequent users completed the task significantly faster than intermediate and beginner users. This highlights the impact of user experience on task efficiency.
C. Frustration Level by User Experience Level
A bar chart illustrates the completion time for each user experience level. As expected, frequent users completed the task significantly faster than intermediate and beginner users. This highlights the impact of user experience on task efficiency.
Analysis and General Observations
Frequent users exhibited faster completion times due to familiarity but voiced concerns about repetitive manual deletions. Intermediate users experienced moderate frustration and errors caused by difficulty distinguishing chats. The beginner user struggled significantly with task comprehension and interface navigation. Notably, the absence of actionable feedback from the “Memory Full” label and the uniform chat title styling were major usability bottlenecks. Visual cues and clearer action prompts emerged as key areas for improvement.
Human systems engineering (HSE) Principles
I.Cognitive Load
ChatGPT’s design partially manages cognitive load. Temporal categorization simplifies chat retrieval, but the uniform styling of chat titles increases mental effort when distinguishing between them. This issue becomes even more apparent when users must navigate through lengthy lists of chats. The “Memory Full” label offers limited actionable feedback, leaving users uncertain about how to address memory issues. The ambiguity surrounding memory management exacerbates cognitive load, particularly for novice users who may not know where to start. An interface that offers clear, immediate guidance could help alleviate this burden by streamlining decision-making and reducing mental effort (Baddeley, 1992; Sweller, 1988; Wickens et al., 2017).
Recommendations:
● Transform the “Memory Full” label into a clickable element offering options such as clearing chats or upgrading memory.
● Display memory usage in the sidebar for greater transparency.
● Provide contextual tooltips or on-hover descriptions to guide users through memory management actions.
II. Decision-Making
The interface prioritizes the chat input field, aiding decision-making for core tasks. However, the sidebar’s uniform text styling complicates prioritization, making it difficult for users to quickly locate frequently accessed or pinned conversations. Additionally, the understated “Upgrade plan” button reduces its visibility, potentially delaying critical decisions related to memory management. These design elements can lead to decision fatigue, particularly for users managing multiple chats or tasks simultaneously. By enhancing visual hierarchies and introducing intuitive cues, the interface can better support user decision-making processes, encouraging timely and efficient actions (Gigerenzer & Goldstein, 1996; Payne, Bettman, & Johnson, 1993; Norman, 2013).
Recommendations:
● Enable users to pin or prioritize chats for easier access.
● Redesign the “Upgrade plan” button with stronger visual emphasis.
● Use color-coded signals or icons to differentiate between priority and secondary actions.
III. Attention
The chat input field effectively captures user attention with its central placement. However, the sidebar lacks a clear visual hierarchy, and white-colored feature buttons below the chat field do not stand out. This lack of contrast makes it harder for users to identify key functionalities quickly, especially under time constraints. Additionally, the repetitive styling of chat titles does little to guide user focus, often leading to unnecessary scanning and delays. Leveraging visual design principles such as contrast, grouping, and alignment can significantly enhance attention management, ensuring that users can locate and interact with essential features more efficiently (Eriksen & Hoffman, 1973; Itti & Koch, 2001; Wickens et al., 2017).
Recommendations:
● Use distinct typography or color highlights to differentiate sidebar categories.
● Enhance the contrast of frequently used feature buttons.
● Introduce visual grouping techniques, such as dividing the sidebar into collapsible sections, to streamline navigation.
IV. Working Memory
Chronological organization of chats partially supports working memory by enabling easier recall. However, uniform presentation of chats increases reliance on memory for context, forcing users to remember specific details about each conversation. This issue is further compounded when managing memory-heavy chats, as the lack of contextual cues makes it harder to identify and prioritize items for deletion. Additionally, feature buttons below the chat field lack immediate explanations, requiring users to recall their functions. Implementing strategies that offload cognitive demands can help users focus on critical tasks without overburdening their working memory (Cowan, 2001; Baddeley, 1992; Miller, 1956).
Recommendations:
● Add icons, tags, or labels to the sidebar for better chat recall.
● Use progressive disclosure to reduce memory strain for feature options.
● Implement a search or filter function to help users locate specific chats quickly.
V. Visual Search
Visual search efficiency is hindered by the uniform styling of chat titles. Users must scan each item individually, slowing task completion and increasing frustration. The absence of visual cues for memory-heavy chats further exacerbates this issue, making it difficult to identify items that require immediate attention. A dedicated memory management section with clear actions could enhance search efficiency, allowing users to quickly locate and address memory-related concerns. By integrating visual aids such as icons, progress bars, and sortable categories, the interface can significantly improve the speed and accuracy of visual search tasks (Wolfe, 1994; Treisman & Gelade, 1980; Wickens et al., 2017).
Recommendations:
● Incorporate visual cues for memory-heavy chats, such as icons or progress bars.
● Introduce a memory overview section with aggregated memory usage and actionable insights.
● Allow users to sort or filter chats based on criteria like memory usage, recency, or frequency of access.
Design Innovations
The proposed enhancements to the application's memory management enhancements leverage Cognitive Load and Visual Search HSE principles to address complex challenges through two innovative features: a categorized memory interface with filtering capabilities, and a streamlined image management system.
By minimizing cognitive burden and enhancing visual search efficiency, these features simplify memory organization and provide intuitive content navigation. The design enables users to quickly locate and manage their content through two access points: the "(i)" icon and the "Manage memory" button in the memory-status widget.Our solution creates an efficient, user-friendly memory management approach that aligns with core human-system interaction principles, empowering users with greater control and ease of content management.
New Redesigned Interface to Manage Memory
I.Categorized Memory Interface
Challenge: The current linear design increases cognitive load and slows navigation.
Innovation: Introduce a filter to sort the chats in the memory management interface being added. The sorting dropdown offers six strategic filtering options to meet diverse user needs:
●'High to Low' and 'Low to High' memory usage sorting helps users quickly identify memory-intensive chats for storage management.
● 'Old to New' and 'New to Old' temporal sorting assists in finding chats based on chronological relevance.
● 'Irrelevant' filter surfaces potentially disposable chats that haven't been accessed recently.
● 'With Images' filter specifically identifies image-heavy chats that typically consume more storage
Categorized Memory Interface to manage memory/Images and Sort
II. Image deletion
Challenge: Giving the user access to all the images uploaded to the application for direct deletion.
Innovation: Add a button in the memory management interface to access all uploaded images for easy deletion. The image gallery displays images in a grid layout with selection options for individual deletion or a "Clear all images" button for bulk deletion. A filter drop-down, similar to the chat view, allows users to sort images by memory usage, date, or relevance. This dual deletion approach—individual or bulk—and intelligent sorting enhances efficiency and gives users precise control over storage management.
Memory Management Interface with Image Gallery View and Sorting Controls
Conclusion
This study highlights critical usability challenges in ChatGPT’s memory management, including high cognitive load, suboptimal decision-making support, and inefficient visual search. Observations and HSE principles guided actionable recommendations such as transforming the “Memory Full” label and enhancing the sidebar design. These innovations aim to reduce cognitive effort, improve task efficiency, and create a user-friendly experience. Applying HSE principles ensures that user-centric designs optimize both functionality and satisfaction, demonstrating the value of research-driven design improvements.
