JMIR Publications

ABSTRACT

Background:

The growing demand for mental health support has led to exploration of digital interventions to bridge the gap between available resources and those seeking help. Large language models (LLMs) have emerged as a promising technology for creating more personalized and adaptive mental health chatbots. Prospection—the ability to envision and plan for future outcomes—represents a transdiagnostic process altered across various mental health conditions that could be effectively targeted through such interventions.

Objective:

To investigate how people use "Future Me," an LLM-powered chatbot designed to facilitate future-oriented thinking, and what types of support they seek from it; to examine how Future Me enables future-oriented thinking; to assess its acceptability and usability across different user groups; to understand postgraduate students' specific perceptions and needs regarding chatbot-based stress management; and to identify how chatbot design can be improved to better address users' mental wellbeing needs.

Methods:

Two complementary studies were conducted. Study 1 (n = 20) examined how postgraduate students used Future Me during a single guided session, followed by semi-structured interviews. Study 2 (n=14) investigated how postgraduate students interacted with Future Me over a one-week period, with interviews before and after usage. Both studies analyzed conversation transcripts and interview data using thematic analysis to understand usage patterns, perceived benefits, and limitations.

Results:

Across both studies, participants primarily engaged with Future Me to discuss career/education goals, personal obstacles, and relationship concerns. Users valued the chatbot's ability to provide clarity around goal-setting, its non-judgmental nature, and its 24/7 accessibility. Future Me effectively facilitated self-reflection and offered new perspectives, particularly for broader future-oriented concerns. However, both studies revealed limitations in the chatbot's ability to provide personalized emotional support during high-stress situations, with participants noting that responses sometimes felt formulaic or lacked emotional depth. Postgraduate students specifically emphasized the need for greater context awareness during periods of academic stress.

Conclusions:

Future Me demonstrates promise as an accessible tool for promoting prospection skills and supporting mental wellbeing through future-oriented thinking. However, effectiveness appears context-dependent, with prospection techniques more suitable for broader life decisions than acute stress situations. Future development should focus on creating more adaptive systems that can adjust their approach based on the user's emotional state and immediate needs. Rather than attempting to replicate human therapy entirely, chatbots like Future Me may be most effective when designed as complementary tools within broader support ecosystems, offering immediate guidance while facilitating connections to human support when needed.