[Paper Review] Does Personalized Nudging Wear Off? A Longitudinal Study of AI Self-Modeling for Behavioral Engagement

Qing He, Zeyu Wang+4 more|arXiv (Cornell University)|Feb 27, 2026

Behavioral Health and Interventions0 citations

TL;DR

This study longitudinally evaluates video and audio AI self-modeling as personalized nudges to sustain fitness engagement, finding video self-modeling yields early gains that taper but can sustain performance, while audio self-modeling shows weaker effects.

ABSTRACT

Sustaining the effectiveness of behavior change technologies remains a key challenge. AI self-modeling, which generates personalized portrayals of one's ideal self, has shown promise for motivating behavior change, yet prior work largely examines short-term effects. We present one of the first longitudinal evaluations of AI self-modeling in fitness engagement through a two-stage empirical study. A 1-week, three-arm experiment (visual self-modeling (VSM), auditory self-modeling (ASM), Control; N=28) revealed that VSM drove initial performance gains, while ASM showed no significant effects. A subsequent 4-week study (VSM vs. Control; N=31) demonstrated that VSM sustained higher performance levels but exhibited diminishing improvement rates after two weeks. Interviews uncovered a catalyst effect that fostered early motivation through clear, attainable goals, followed by habituation and internalization which stabilized performance. These findings highlight the temporal dynamics of personalized nudging and inform the design of behavior change technologies for long-term engagement.

Motivation & Objective

• Motivate long-term fitness engagement using AI-generated self-models.
• Compare two AI self-modeling modalities (video and audio) against a control.
• Examine temporal dynamics of nudging effects and habituation.
• Identify design factors that influence the durability of AI self-modeling interventions.

Proposed method

• Two-stage empirical evaluation with 1-week exploratory (N=28) and 4-week follow-up (N=31) studies.
• Implementations of Video Self-Modeling (VSM) and Audio Self-Modeling (ASM) built on existing pipelines (FakeForward, Visomaster, ElevenLabs).
• Objective measures: wall-sit duration and crunch repetitions, normalized to Day 1 baselines.
• Subjective measures: IMI, ESES, and VAIQ questionnaires across days 1 and 7 (Day 2–6 shortened).
• Linear Mixed-Effects modeling to assess sustained performance and rate changes over time.

Figure 1 . The nudging effect of AI self-modeling across time and modalities in fitness. (a) Change of nudging effect of video self-modeling(VSM): users are motivated by the AI self-model at first, then grow numb to the repeating VSM intervention, yet the motivated self-model has internalized as the

Experimental results

Research questions

• RQ1 Does AI self-modeling sustain performance over time in everyday fitness practice?
• RQ2 Does AI self-modeling sustain the rate of improvement over time?
• RQ3 What design factors influence the long-term impact of AI self-modeling for behavior change?

Key findings

• VSM produced faster initial improvements with tapering gains over the 7-day period.
• ASM showed less consistent improvements and did not outperform controls consistently.
• Control group showed steady natural improvement over the week, indicating practice effects without nudges.
• Over 4 weeks, VSM sustained higher performance but with diminishing improvement rates after about two weeks.
• Interviews suggested habituation to continuous VSM but internalization of the goal persisted.

Figure 2 . System implementation for AI self-modeling. Interfaces are adapted from existing platforms and included for illustration only.