27% of US adult internet users have social or emotional interactions with AI, new survey finds
A YouGov poll for Elon University's Imagining the Digital Future Center, covered the same day by The Washington Post, found that among the AI-companion users it isolated, 31% consider their bot a friend and 39% tell it things they wouldn't tell another person.
Elon University's Imagining the Digital Future Center published survey results on September 2, 2026, covered the same day by The Washington Post, putting a number on something that's mostly been anecdotal: how many people treat an AI chatbot as something closer to a companion than a tool. Across a screened sample of US adult internet users, 27% reported having meaningful social or emotional interactions with an AI large language model.
How the survey was built
YouGov fielded the poll for Elon between May 18 and 22, 2026, screening 4,268 US adult internet users and matching them down to a working sample of 4,031, from which it drew a target subsample of 1,000 people with social or emotional AI use, matched to the broader population on gender, age, race, and education. That subsample is where the more specific figures below come from.
Inside the 1,000-person companion sample
31% consider their AI a friend, 59% agree it gives them support they need, 39% say they've told the AI things they wouldn't tell another person, and 37% say they'd feel a personal loss if they lost access to it. Center director Lee Rainie called the results "the first wave of insights about these emerging relationships" as large language models become part of daily life.
This site's own tracker is built around a narrower, harder-edged version of the same question -- not whether an employee likes talking to a model, but whether the work that comes out the other side is worth what it cost. The two questions turn out to be entangled: if more than a third of people already say losing AI access would feel like a personal loss, that's not just a wellbeing statistic, it's a signal about how embedded these tools already are in daily judgment and decision-making, inside and outside of work, well ahead of most organizations having any systematic way to tell whether that embedding is producing good outcomes or just familiar ones.