Seek and De-Stress: Matching Supporter Identity to Seeker Need in AI Emotional Support
The study asks whether AI emotional support improves when help-seekers can choose among differentiated supporter identities rather than one generic assistant. The authors built 42 help-seeking personas grounded in COVIDiSTRESS pandemic-era mental-health survey data and four synthetic supporter identities with distinct support philosophies; an intake battery separated seekers into two need patterns (emotional steadiness versus practical direction), which preferred different supporters. Simulated seekers ranked the supporters, then held conversations with their top choice, bottom choice and a generic baseline across four commercial model providers, giving 1,008 conversations. Matched supporters produced longer conversations, more reported relief and more feeling understood; process audits show they scored higher on problem-solving and personalisation while generic or unmatched support sounded more validating.
Publisher
ACM (Proceedings of the 26th ACM International Conference on Intelligent Virtual Agents, IVA 2026); Cold Spring Harbor Laboratory
Published
6 Sept 2026
Added
today
Key Findings
- 42 simulated help-seeker personas (COVIDiSTRESS-grounded) x 4 supporter identities x 4 commercial providers = 1,008 conversations
- Two need patterns (emotional steadiness versus practical direction) prefer different supporters
- With their top-ranked supporter, simulated seekers more often prolonged the conversation, reported greater relief and felt more understood than with the generic assistant
- Matched supporters scored higher on problem-solving and personalisation in process audits; generic or unmatched support more often sounded validating without addressing the need
Methodology Notes
Fully simulated: LLM help-seekers rate LLM supporters; no human participants. Authors at Cold Spring Harbor Laboratory. Published 2026-09-06 (Crossref published date; print 2026-09-07; DOI created 2026-09-03); CC BY 4.0; conference IVA 2026, Puebla, Mexico. dl.acm.org is walled from this network; verified through the Crossref record (authors, dates, licence, event) and the OpenAlex record (full abstract). No arXiv preprint found. The IVA 2026 proceedings volume (48 DOIs) had not been swept before this sweep.
Authors
Ben Wigler, Maria Tsfasman
Tags
Cite This
APA
Ben Wigler, Maria Tsfasman. (2026). Seek and De-Stress: Matching Supporter Identity to Seeker Need in AI Emotional Support. ACM (Proceedings of the 26th ACM International Conference on Intelligent Virtual Agents, IVA 2026); Cold Spring Harbor Laboratory. https://doi.org/10.1145/3806774.3827995