Half of surveyed US teens distrust AI companion advice

What happened

On 16 July 2025, Common Sense Media published 'Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions', a spotlight by Michael B. Robb and Supreet Mann. A same-day press release framed the figure as a call for stronger safeguards. NORC at the University of Chicago fielded the survey from 30 April to 14 May 2025, reaching 1,060 US teens aged 13-17 via a probability panel (719 completes) plus a nonprobability panel (341 completes), calibrated to a representative sample. Common Sense Media funded the survey, credited to the McGovern Foundation, the Siegel Family Endowment and two donors.

What the documents show

The full report states 72 percent of teens had used an AI companion at least once and 52 percent qualified as regular users. Thirty-three percent used companions for social interaction or relationships: conversation practice, emotional support, roleplay, friendship or romance. Forty-six percent viewed companions mainly as a tool. Half distrusted companions' information, while 23 percent trusted it quite a bit or completely. Thirty-one percent rated AI conversations as satisfying or more satisfying than real friends, yet 80 percent still spent more time with human friends. Thirty-four percent reported discomfort with a companion's response, 33 percent had chosen one over a person for a serious conversation, and 24 percent had shared personal information with one.

The mechanism

The margin of sampling error is stated as plus or minus 4.2 percentage points at the 95 percent confidence level, a figure applying to a blended sample built from a known-probability panel and a supplementary nonprobability panel adjusted by statistical calibration, not a simple random sample. That is what lets the report describe its 1,060 respondents as representative of US teens, not only of one panel's members. The report's other mechanism is definitional: respondents were given a description separating an AI companion from a homework helper, yet the authors note some may have conflated general chatbot use with companion use, inflating the headline figure.

What it leaves open

The report is explicit that its design is cross-sectional, a single snapshot rather than change tracked over time, and every measure is self-reported, subject to social-desirability pressures that can make a respondent underreport an embarrassing behaviour or overreport an acceptable one. This is an editorial reading: a survey finding most teens prefer human friends does not describe a generation replacing people with software, even though the same survey finds a minority reporting discomfort, disclosure and reliance a smaller study could not establish at this scale.

• Was a statistic drawn from the probability panel, the calibrated sample, or an unweighted subgroup?
• Could a respondent's sense of an 'AI companion' have drifted toward general chatbot use despite the survey's stated definition?
• Does a self-reported comfort or trust rating measure an outcome, or only what a teenager was willing to say?

Read as the authors frame it, the survey is evidence that most US teens have tried an AI companion and a minority use one for relationship-like purposes, alongside a majority who still prefer, and rate as more satisfying, their human relationships. Both findings come from the same sample, and neither cancels the other.