Perfectly to a Tee: Understanding User Perceptions of Personalized LLM-Enhanced Narrative Interventions

TL;DR

This study investigates dynamic personalization of narrative interventions for mental health by comparing personalized LLM-enhanced stories to human-written ones in two young-adult populations. Using a randomized single-session design, the authors assess narrative qualities (authenticity, clarity of takeaways, reflection, and balance) and intervention outcomes (reduction in belief in negative thoughts, emotional intensity, relatability, and action-oriented thinking). Findings show LLM-enhanced stories generally improve key takeaways, reflection, perceived solution capacity, and future engagement, with authenticity comparable to human-written content though some concerns about AI tone and over-personalization. The results yield design recommendations for future DMH tools, emphasizing a balance between relatable content and plausibility, refined AI tone, structured reflection, peer-support integration, and longitudinal deployment considerations. Overall, the work demonstrates that AI-assisted narrative interventions can meaningfully support young adults in managing negative thoughts while outlining practical guidelines for safe and effective deployment in digital mental health ecosystems.

Abstract

Stories about overcoming personal struggles can effectively illustrate the application of psychological theories in real life, yet they may fail to resonate with individuals' experiences. In this work, we employ large language models (LLMs) to create tailored narratives that acknowledge and address unique challenging thoughts and situations faced by individuals. Our study, involving 346 young adults across two settings, demonstrates that personalized LLM-enhanced stories were perceived to be better than human-written ones in conveying key takeaways, promoting reflection, and reducing belief in negative thoughts. These stories were not only seen as more relatable but also similarly authentic to human-written ones, highlighting the potential of LLMs in helping young adults manage their struggles. The findings of this work provide crucial design considerations for future narrative-based digital mental health interventions, such as the need to maintain relatability without veering into implausibility and refining the wording and tone of AI-enhanced content.

Figures (10)

• Figure 1: Mean perceived authenticity across Non-LLM and LLM stories addressing six challenging thoughts, shown separately for crowdworkers (left) and students (right). Error bars represent the standard error of the mean, with sample sizes indicated above each bar.
• Figure 2: Mean perceived ability to communicate key takeaways across Non-LLM and LLM stories addressing six challenging thoughts, shown separately for crowdworkers (left) and students (right). Error bars represent the standard error of the mean, with sample sizes indicated above each bar.
• Figure 3: Mean perceived ability to promote reflection across Non-LLM and LLM stories addressing six challenging thoughts, shown separately for crowdworkers (left) and students (right). Error bars represent the standard error of the mean, with sample sizes indicated above each bar.
• Figure 4: Mean perceived balance of positivity with realistic struggles across Non-LLM and LLM stories addressing six challenging thoughts, shown separately for crowdworkers (left) and students (right). Error bars represent the standard error of the mean, with sample sizes indicated above each bar.
• Figure 5: Mean perceived reduction in belief in negative thoughts across Non-LLM and LLM stories addressing six challenging thoughts, shown separately for crowdworkers (left) and students (right). Error bars represent the standard error of the mean, with sample sizes indicated above each bar.