Abstract

Telephone fraud is pervasive and costly, but its inner workings are rarely observed at scale because complete recordings of real scam conversations are hard to obtain. We analyze a complete corpus of 10,211 real inbound scam and spam calls—913 hours of audio and 330,956 transcribed turns from 5,780 distinct originating numbers—collected over 54 days by an AI voice-agent honeypot that answered callers and kept them talking while recording and transcribing every call. We ask a set of deliberately simple questions about how phone scammers actually behave, separating outright scams (calls that solicit sensitive information) from the far larger stream of predatory but legal lead-generation (spam) that funds and feeds them. Scam operations keep office hours; a churn of thousands of distinct numbers runs only a small catalog of recycled scripts; those scripts give the scam away almost immediately; and callers solicit identity anchors (home address and date of birth) far more often than payment credentials, applying pressure through persistence and manufactured authority rather than overt threats. Our central experiment asks: does it matter who picks up? Across 1,823 randomized calls, scammers spent about 15% more conversational turns per decade of the target's apparent age, yet what they asked for did not change at all. Telephone fraud emerges as a templated industry that varies how hard it works a target, but not what it wants from them.