Personalized Intelligent Chatbot Based on AI-Generated Content Assists Memoir Writing for Older Adults With Cognitive...

Older adults with cognitive impairment often face significant challenges in memoir writing, including memory fragmentation, emotional loneliness, and speech and language disorders. Although AI-generated content (AIGC) technologies such as GPT-3.5 show potential in creative tasks, they often lack the personalization and adaptability required for users with dementia. Generic AIGC tools often fail to address the heterogeneous cognitive and emotional needs of this population.

This study aimed to design and evaluate, as a proof-of-concept, a personalized AIGC-powered chatbot to assist older adults with cognitive impairment in memoir writing and emotional support.

We developed a multimethod collaborative design framework integrating Kansei Engineering, Quality Function Deployment, Axiomatic Design, and the Technique for Order of Preference by Similarity to Ideal Solution decision model. The system dynamically adapts interaction strategies based on users' Mini-Mental State Examination (MMSE) scores, using a clinical threshold of 20 to distinguish mild (20-26) from moderate-to-severe (<20) impairment. In a single-session, nonrandomized, matched-pair evaluation using minimization-based allocation with an active control condition, performance was assessed via usability testing (System Usability Scale), affect assessment (Positive and Negative Affect Schedule), and blinded psychiatrist-rated memoir quality among 20 participants (10 per arm).

The experimental group showed significantly greater improvement than the active control group in positive affect, negative affect, and psychiatrist-rated memoir quality (all Ps≤.03 by matched-pair analysis; all comparisons remained significant after Benjamini-Hochberg correction), with moderate-to-large effect sizes (Cohen d=0.85-1.03 for affect outcomes; rank-biserial r=1.00 for memoir quality). Within the sample (MMSE range 11-23), participants in the lower MMSE tier (<20; n=12) appeared to benefit more from AI-driven narrative generation, while those in the higher tier (20-23; n=8) responded better to keyword-based prompting; these subgroup observations are descriptive, given the small cell sizes. The mean System Usability Scale total score of 92.2 (SD 5.1) exceeded the acceptability threshold and fell in the excellent range, though single-session exposure and potential acquiescence bias warrant cautious interpretation.

These findings provide preliminary, hypothesis-generating evidence that a personalized AIGC-based chatbot may support short-term affective and narrative outcomes among older adults with cognitive impairment. The adaptive, multimodal design shows promise for human-AI collaboration in memoir writing and older adult care contexts, but larger, adequately powered randomized controlled trials with verified active control fidelity, multisession follow-up, and content source-differentiated outcome scoring are needed before clinical conclusions can be drawn.