The Rise of AI Companions: A national public opinion survey on the relationships people form with artificial intelligence systems

Key Findings

• — 27% of US adult internet users qualify as AI companion users, defined as having used AI in the past six months for at least one of six purposes: entertainment, sharing life experiences, advice on personal matters, companionship when lonely, discussing feelings, roleplay, or romantic chat
• — Among companion users, 35% said their bot at times agrees with them too much and 27% said it at times wants to keep them talking with it
• — 15% said their bot at times makes them feel less in touch with reality; 7% said it at times makes them feel lonely; 6% said it at times discourages them from talking with other people; 4% said it at times hurts their feelings
• — Asked how they would feel if their AI companion disappeared, 36% would not miss it at all, 40% would miss it a bit, and 24% would miss it a lot
• — Asked whether they would rather converse with AI or with friends and family, 59% chose friends and family, 27% valued both similarly, 3% chose AI, and 11% did not know
• — 20% said their AI companion is at times more reliable than friends or family, and 20% said they have used AI to assess whether their own AI use was healthy or dependent
• — The report states that a majority of respondents predict AI will deepen social isolation and loneliness in the years ahead

Methodology Notes

Online YouGov panel survey fielded 18-22 May 2026; 4,268 US internet-using adults aged 18+ interviewed and screened, matched down to 4,031, yielding the analysed subsample of n=1,000 AI users with emotional or social use of AI. Matched on gender, age, race and education with four-way stratification; weights trimmed and recentred; stated margin of error +/- 3.68 points. The headline percentages therefore describe companion users, not the general population, apart from the 27% incidence figure itself. Publication date 2026-06-15 from the landing page JSON-LD datePublished, corroborated by the PDF Subject line ('June 2026') and the /wp-content/uploads/2026/06/ path; the report PDF was regenerated on 2026-07-28 (PDF CreationDate) with a cover relabelled August 2026, so the cover date is not the publication date. Several figures in the report appear in multi-column infographics that do not survive text extraction with reliable label-to-value pairing; only figures readable unambiguously are recorded here. FURTHER REVISION (recorded 2026-09-03): the report PDF at the same /uploads/2026/06/ URL was regenerated again on 2026-08-29 (PDF CreationDate and HTTP Last-Modified 2026-08-29; 15,433,787 bytes, 26 pages) with the masthead now reading 'September 2026' while the PDF Subject metadata still says 'June 2026' and the fieldwork line still reads May 18-22, 2026; Elon University publicised the report on 2026-09-02. The published_date here remains the original 2026-06-15 release; the document has now carried three cover dates (June, August, September 2026) under one URL.

Sources