What question did this study set out to answer?

This research aims to explore the role of personalization in AI use and its impact on user engagement.

April 8, 2026Open Access

Toward an AI Personalization Index: A 157-Day Single-User Case Study

Key Points

This research aims to explore the role of personalization in AI use and its impact on user engagement.
Conducted a single-user exploratory study over 157 days on the Claude platform.
Compared performance metrics of an established session with memory slots versus a new session without memory slots.
Analyzed context re-explanation frequency, correction requests, and time to first meaningful output.
Identified significant differences in performance metrics between the established and new sessions.
Established a conceptual framework defining the relationship between model performance and personalization index.
Proposed four candidate variables related to personalization that may influence AI engagement.

Abstract

AI model benchmark performance has reached a saturation point. According to the Stanford 2025 AI Index Report, the performance gap among the top 10 models has narrowed to 5.4%, and the U.S.–China gap stands at only 1.70%. Yet the EY 2025 Work Reimagined Survey (29 countries, 15,000+ respondents) reveals that while 88% of employees use AI, only 5% engage in advanced utilization. This study proposes that the absence of personalization may be an important explanatory variable for this discrepancy. The author conducted an exploratory single-user manual personalization study over 157 days (~1,000 hours) on the Claude platform (Max plan). As the primary evidence, a same-time comparison on the identical model (Opus 4.6) between an established session (30 memory slots) and a new session (0 memory slots) revealed marked differences in context re-explanation frequency, correction requests, and time to first meaningful output. This study proposes a conceptual organizing framework—Experienced Performance ∝ Model Performance × Personalization Index—and preliminarily defines four candidate variables.

Ask AI

Helpful

Bookmark

View Full Paper