Turkish-language Large Language Model-supported Voice Companion for Cognitive Monitoring via Social Artificial Intelligence in Older People Living Alone: A Single-Arm Pilot Feasibility Study (Preprint)

BACKGROUND Older adults living alone are particularly vulnerable to reduced social interaction and loneliness, which are associated with adverse mental and physical health outcomes. At the same time, age-related cognitive decline becomes more prevalent, highlighting the growing need for scalable, technology-enabled interventions that can both mitigate social isolation and support routine cognitive monitoring in home settings. OBJECTIVE To develop iShe (Intelligent Social Health and Interaction), a Large Language Model (LLM)-supported mobile voice assistant designed to increase social interaction, reduce loneliness and depressive symptoms, and enable routine monitoring of the cognitive status of elderly adults living alone at home, and to evaluate its feasibility and acceptability. METHODS In this single-arm pilot feasibility study, ten community-dwelling adults aged 65 years and older living alone used the application for 30 days and were encouraged to complete one daily voice session. Loneliness and depressive symptoms were assessed at baseline and endline using the De Jong-Gierveld Loneliness Scale (DJGLS) and Geriatric Depression Scale-15 (GDS-15). Cognitive status was monitored in-app daily using the Short Portable Mental Status Questionnaire (SPMSQ), and the Standardized Mini-Mental State Examination (SMMSE) was administered face-to-face at baseline and endline. Paired changes were analyzed using Wilcoxon signed-rank tests and descriptive feasibility metrics (retention, adherence, session duration, support needs). RESULTS Retention was 100%. Participants completed all 30 sessions. Mean session duration was 18.6 ± 3.2 minutes. DJGLS decreased from 8.50 ± 5.28 to 5.70 ± 4.92 (p = .007; Cohen’s dz = 0.98; median change 2 [IQR 1–3.75]). GDS-15 decreased from 4.50 ± 4.22 to 2.30 ± 2.26 (p = .039; Cohen’s dz = 0.66; median change 1 [IQR 0–2]); the proportion scoring 6 or higher declined from 30% to 10%. SMMSE remained stable (29.3 ± 1.1 to 29.6 ± 0.7), and all SPMSQ results remained in the normal range (0–2 errors), consistent with SMMSE categories. CONCLUSIONS This pilot study provides strong feasibility and acceptability signals for daily use of an LLM- supported voice companion by older adults living alone, with exploratory within-group improvements in loneliness and depressive symptoms and stable, concordant cognitive screening outputs. Larger controlled studies are needed to establish efficacy and longer-term cognitive monitoring value. CLINICALTRIAL