One in Five AI Loneliness Scenes Include a Phone or Social Media

The phone has become a recurring visual language for loneliness. In a privacy-safe analysis of completed Magic Hour projects, 1,034 observed creators explicitly depicted loneliness, feeling lonely, dying alone or eating alone. Of those creators, 226 also referenced a phone, screen, social media, scrolling, online activity or text messaging.

That means technology appeared in 21.9% of explicit AI loneliness scenes. The finding describes how Magic Hour creators represent loneliness. It does not show that phones cause loneliness.

What creators put in loneliness scenes

The broader phone, screen or social-media category appeared for 226 creators. Among individual physical objects reviewed, windows appeared for 168 creators, literal phones for 129, mirrors for 105, tables or desks for 82, beds for 78, photographs for 78, and chairs or benches for 76. Categories can overlap.

What the finding means

Creators repeatedly place digital connection inside scenes about emotional disconnection. That contrast makes technology a prominent part of the visual vocabulary of loneliness in this dataset.

This is an observed association, not a causal result. Magic Hour did not survey creators about their emotional state or infer why they selected particular objects.

Methodology

The analysis covers completed, enabled, nondeleted Magic Hour image, video and audio projects from June 16, 2023 through September 19, 2026. It counts distinct observed creators, not prompt volume alone.

The loneliness cohort required explicit language such as loneliness, feels lonely, feeling lonely, dies alone or eating alone. The technology category required references to phones, smartphones, screens, social media, scrolling, online activity, text messages or related terms.

Templates, logged automated free-tool tasks, blank prompts, exact prompts reused by five or more creators in the same month, adult content, named political and celebrity subjects, and prominent copyrighted-character contexts were excluded. Creator counts were checked against heavy-user concentration.

The 2x estimate for unlogged free-tool usage is not included in observed counts or percentages. Magic Hour users are not representative of everyone, and the findings should not be generalized to the public.

No raw prompts, personal information, project identifiers or user media are published.