Stress in America™ 2026: The stress of uncertainty

Key Findings

• — 70% of US adults reported intentionally using AI in 2026, up from 62% in 2025
• — Among AI users, 65% had ever used AI to help make day-to-day decisions, 62% to get input on treatments, 60% to self-diagnose and 59% to discuss their physical health
• — Among AI users, 49% had engaged in conversations with AI as a friend, 49% had discussed their relationships, 50% had discussed their mental health and 47% had used AI in a role comparable to a mental health professional (e.g. practising therapy exercises or getting a second opinion on a therapist's suggestions)
• — Regular or frequent use (topline 'regularly/often'): 22% of AI users converse with AI as a friend, 18% discuss their mental health and 16% use it in a mental-health-professional role
• — Asked how using AI for personal support, advice or conversation had affected them, 51% of AI users selected at least one positive impact and 29% at least one negative impact: 12% said it makes them feel more emotionally dependent on technology, 10% that they rely on it more than on friends and family for advice, 9% that it makes them feel more disconnected from other people, 8% that it has steered them in the wrong direction and 7% that it has made them feel worse about themselves; 18% said it has affirmed their thoughts, feelings or decisions
• — 47% of all adults said they had pulled back from at least one form of social participation in the past year because of stress, uncertainty or feeling overwhelmed; about half reported at least sometimes feeling isolated (51%), left out (49%) or lacking companionship (48%)

Methodology Notes

Online survey by The Harris Poll for APA, 29 July to 24 August 2026; 3,132 US adults 18+ (national sample) with oversamples of 801 Black, 803 Latino/a/e and 827 Asian adults; Latino/a/e interviews in English and Spanish. Weighted to the March 2025 Current Population Survey and Pew NPORS 2025/Pew 2023 benchmarks; respondents drawn from people who agreed to take part in Harris surveys (opt-in panel); Bayesian credible interval ±2.5 points for the general population. AI items (Q4024, Q4033, Q4034) asked about generative AI tools 'e.g., ChatGPT, Microsoft Copilot, image generators'; AI-use and impact questions were asked of AI users (unweighted base 2,353, weighted 2,191). Self-reported, cross-sectional; impact items are multi-select perceptions, not measured outcomes. Report page read via r.jina.ai (apa.org HTML returns a ~1 KB bot shell to curl); topline and questionnaire PDFs fetched directly.

Sources