Leveraging artificial intelligence to identify the psychological factors associated with conspiracy theory beliefs online

The paper, in plain language

Measuring behaviour, not just answers

Nearly everything known about conspiracy believers comes from what they tell researchers. Here we measured what they do: 2,506 US-based Twitter users completed psychological surveys, and an AI pipeline classified their 7.7 million posts, likes, replies and reposts across the pandemic for support of six common COVID-19 conspiracy theories. The classifier was validated against human raters before we trusted it. Interviews cannot reach this scale; this is also what the machine-learning turn in our lab is for.

What actually predicted support

Three factors were most consistent: being older, placing oneself at the far left or far right of the political spectrum — extremity on both ends, not one side — and general susceptibility to believing false information. Equally telling is what did not predict behaviour: the widely used conspiracy-mentality questionnaire showed no significant association with real online conspiracy support, and neither did narcissism nor need for chaos. Education was not the shield it is often assumed to be.

Why it matters

If a standard questionnaire fails to predict who actually spreads conspiracy content, then part of the field has been measuring an attitude that does not travel into behaviour — a methodological warning well beyond this topic. The honest caveats: the design is correlational, one platform, and explained variance was modest. But as a proof of method, pairing self-report with AI-classified behaviour at scale opens a door the field needed opened.