Key findings from MHA's survey on AI and mental health

Key Findings

• — 53% of respondents had used AI for emotional support at least once (regularly/often or a few times); 33% said they usually use AI for emotional support and 29% for companionship.
• — Among respondents under 18, 65% reported using AI for emotional support and 37% for companionship, versus 29% and 18% among those over 45; 57% of under-18 emotional-support users said it helped them get help without telling their family.
• — Among all emotional-support users, 55% cited 24/7 availability, 47% feeling more comfortable talking to a bot than a human and 45% getting help without telling their family; 11% had used AI to get a mental health diagnosis.
• — Among regular emotional-support users, 29% wanted AI to ask them directly about risk of harm to themselves or others and 26% wanted AI to send resources, including how to connect to a person.
• — Regular users ranked privacy of conversations first (22%) among company protections; non-users ranked referral to a mental health professional first (25%) and a reminder that they are talking to AI second (19%). 68% named government policies and laws for technology companies as a priority.
• — Use of AI for emotional support rose by 9 percentage points among 18-24-year-olds and 10 points among 25-34-year-olds between May-September 2025 and March-July 2026.

Methodology Notes

Open, self-selected online survey hosted next to MHA's mental health screens, so respondents are a help-seeking population; MHA notes regular AI users may be over-represented. Data collected 21 May 2025 to 22 July 2026; 10,503 responses (item-level n smaller, e.g. 8,314 answered age; 39% were under 18; 56% of 7,943 lived in the US). The responsibility questions were asked from December 2025 only. Descriptive percentages, no weighting or inference reported. The page also summarises a separate qualitative study (University of Minnesota, Northwestern, Dartmouth, Microsoft; 53 adults who used AI during a mental health crisis). Date: page JSON-LD datePublished 2026-09-10 (modified 2026-09-11); MHA's 'Artificial intelligence and mental health' hub page linking to it was published 2026-09-11.

Sources