Reimagining Support: Exploring Autistic Individuals' Visions for AI in Coping with Negative Self-Talk

TL;DR

This paper addresses NST in autistic adults and investigates how AI, particularly LLMs, could support coping without replacing traditional therapy. It employs a mixed-methods design, combining a survey (N=200) and practitioner interviews to map NST experiences, desired AI roles, and prompt–LLM dynamics, including analysis of LLM responses to participants’ NST prompts. Key contributions include empirical insights into NST themes, diverse AI-support preferences, and actionable design implications for neuro-inclusive, multimodal AI tools that complement therapy while addressing safety and trust concerns. The findings have practical implications for developing AI-assisted mental health tools that respect neurodiversity, privacy, and the therapeutic alliance, ultimately aiming to reduce NST-related distress in autistic individuals.

Abstract

Autistic individuals often experience negative self-talk (NST), leading to increased anxiety and depression. While therapy is recommended, it presents challenges for many autistic individuals. Meanwhile, a growing number are turning to large language models (LLMs) for mental health support. To understand how autistic individuals perceive AI's role in coping with NST, we surveyed 200 autistic adults and interviewed practitioners. We also analyzed LLM responses to participants' hypothetical prompts about their NST. Our findings show that participants view LLMs as useful for managing NST by identifying and reframing negative thoughts. Both participants and practitioners recognize AI's potential to support therapy and emotional expression. Participants also expressed concerns about LLMs' understanding of neurodivergent thought patterns, particularly due to the neurotypical bias of LLMs. Practitioners critiqued LLMs' responses as overly wordy, vague, and overwhelming. This study contributes to the growing research on AI-assisted mental health support, with specific insights for supporting the autistic community.

Figures (11)

• Figure 1: Frequency of NST based on responses from the Negative Automatic Thoughts (ATQ-N 10) and Anxiety Scale for Autism-Adults (ASA-A). Distribution of participants' responses to the items in these questionnaires, with an overall mean frequency score of 3.25 (on a scale of 1–5). Participants reported a high frequency of thoughts related to the need for preparation, anxiety about social interactions, and discomfort with unfamiliar situations.
• Figure 2: Coping strategies and technologies used by participants to manage NST. Figure a (left) summarizes coping strategies derived from participants’ open-ended responses, where similar answers were grouped into broader categories. Figure b (right) shows the range of technologies participants reported using as coping mechanisms for NST, with an average of 2.3 tools selected by participants.
• Figure 3: Participants rated the helpfulness of discussing their most impactful NST with different sources: a therapist, a friend or significant other, family, and AI, using a scale from 'Not helpful' to 'Extremely helpful.' Therapists received the highest mean helpfulness score (3.29), followed by friends or significant others (2.79), family (2.29), and AI (2.13).
• Figure 4: Participants' preferences for support types, conversation structures, tone, and modalities in LLM interactions for coping with NST. Mean ratings (1 = not at all useful, 5 = extremely useful) are displayed next to each bar to provide an overview of participants' evaluations for each category.
• Figure 5: Participants' concerns about using AI for mental health support. Mean ratings (1 = not at all, 5 = a great deal) are displayed next to each bar, indicating the highest concerns for misinterpreting thoughts or feelings (Mean = 3.73) and a lack of understanding of thought processes (mean = 3.74), followed by emotional and communication style misunderstandings.