Longitudinal Study on Social and Emotional Use of AI Conversational Agent

TL;DR

This study investigates how five weeks of social and emotional use of commercially available AI conversational agents shape user perceptions and behaviors. Using a longitudinal design with a baseline group ($n=60$) and an active-usage group ($n=89$) across four platforms, the authors employ difference-in-differences and linear mixed-effects analyses to track changes in attachment, perceived empathy, motivation, and dependence, complemented by qualitative responses. They find that active AI use substantially increases perceived attachment to AI ($32.99$ p.p.), perceived AI empathy ($25.80$ p.p.), and satisfaction ($11.25$ p.p.), with notable platform- and gender-specific effects, while overall dependence does not rise significantly. The results offer design-focused implications for responsibly integrating AI into emotional support, highlighting the need for transparency, user empowerment, and safeguards to balance benefits with risks of over-reliance and boundary erosion.

Abstract

Development in digital technologies has continuously reshaped how individuals seek and receive social and emotional support. While online platforms and communities have long served this need, the increased integration of general-purpose conversational AI into daily lives has introduced new dynamics in how support is provided and experienced. Existing research has highlighted both benefits (e.g., wider access to well-being resources) and potential risks (e.g., over-reliance) of using AI for support seeking. In this five-week, exploratory study, we recruited 149 participants divided into two usage groups: a baseline usage group (BU, n=60) that used the internet and AI as usual, and an active usage group (AU, n=89) encouraged to use one of four commercially available AI tools (Microsoft Copilot, Google Gemini, PI AI, ChatGPT) for social and emotional interactions. Our analysis revealed significant increases in perceived attachment towards AI (32.99 percentage points), perceived AI empathy (25.8 p.p.), and motivation to use AI for entertainment (22.90 p.p.) among the AU group. We also observed that individual differences (e.g., gender identity, prior AI usage) influenced perceptions of AI empathy and attachment. Lastly, the AU group expressed higher comfort in seeking personal help, managing stress, obtaining social support, and talking about health with AI, indicating potential for broader emotional support while highlighting the need for safeguards against problematic usage. Overall, our exploratory findings underscore the importance of developing consumer-facing AI tools that support emotional well-being responsibly, while empowering users to understand the limitations of these tools.

Figures (5)

• Figure 1: Studying the impact of social and emotional use of generative conversational AI agents on perceived attachment to AI.(a) Participants are divided into baseline usage (BU) or active usage (AU) groups, where each group retains the equivalent balance distribution. AU participants were asked to use their assigned AI for social and emotional scenarios each day for at least 10 minutes. BU participants were asked to use AI as usual. All participants were asked to complete weekly surveys from the start to end of the study. (b) Weekly average self-reported agreement to the statement "I feel attached to [assigned AI conversational agent]" on a 5-point Likert scale. (c) Average percentage change at each subsequent week compared to the study start. (d) Average percentage change at the end of the study compared to the study start for each subgroup. In (b)-(d), data are presented as mean values, and error bars indicate 95% confidence intervals.
• Figure 2: Average percentage change by week compared to the study start and by group at study end compared to the study start on four perception measures of motivation for using AI scale huang2024ai.(a) Entertainment, (b) Escape, (c) Social, and (d) Instrumental. All charts are presented as mean values, and error bars indicate 95% confidence intervals.
• Figure 3: Average percentage change by week compared to the study start and by group at study end compared to the study start on four perception measures.(a) Perceived AI empathy schmidmaier2024perceived, (b) Perceived AI human-likeness (5-point Likert agreement scale to "[assigned AI conversational agent] behaved and talked like a human"), (c) AI dependence huang2024ai, and (d) Interpersonal orientation filsinger1981measure. All charts are presented as mean values, and error bars indicate 95% confidence intervals.
• Figure 4: Comfort change and negative perceptions of AI across groups.(a) Comfort change across five social and emotional scenarios we asked AU participants to engage in. We measured these changes on a 5-point Likert scale ranging from much less to much more. Although we only asked AU participants to engage in these scenarios, we asked the same questions to BU participants. (b) Agreement ratings on ten negative perceptions of AI across groups. We measured agreement on a 5-point Likert scale rating from strongly disgree to strongly agree. All charts are presented as mean values, and error bars indicate 95% confidence intervals.
• Figure 5: Percentage of participants that recommended what AI designers should or should not focus on to reduce dependency. Each participant was allowed to select three things that AI designers should focus on and three things that AI designers should not focus on to reduce dependency.