Appeals to authority in 73.8% of robocall scams aimed at older adults, against 7.1% of the rest

GASA’s Research Working Group met 17 September, and the writeup carries two studies rather than the membership news the source usually produces. The robocall one is Ryan Moore (UT Austin) and Doug Shadel (Fraud Prevention Strategies), working from 4.49 million transcripts collected on 295,749 decoy lines between January 2018 and December 2022. The preprint names Nomorobo as the honeypot operator and Shadel as a past Nomorobo consultant.

Scam calls score above spam calls on threat of loss (d = 0.45), time pressure (d = 0.83) and emotional arousal (d = 0.63). The sharpest figure is the age split: appeals to authority — IRS, FBI, Social Security, Medicare — appear in 73.8% of scams aimed at older adults against 7.1% of those aimed at the general public. The arousal gap between the two groups is 0.03 on a nine-point scale.

Two things constrain what the corpus can say. The decoy lines answer with “hello” and record, so every transcript is an outbound pitch and none of it describes what happens once a person engages. And the labels are model-assigned: scam-versus-spam comes from a GPT-4o-mini prompt hand-trained on 1,499 transcripts, scoring F1 = .70 on a 650-transcript held-out set. Older-adult targeting is inferred from content cues — Medicare consultations, final expense insurance, sweepstakes — which the authors acknowledge overlap the LIWC health, risk and illness categories that then come out as the top discriminators.

The AI extension Moore presented at the meeting is not in the preprint: five generative models prompted for 600 scam pitches, showing higher arousal and heavier use of all four tactics. Separate, unpublished work.

Yaniv Hanoch’s ticket study, 604 participants across three studies, found people correctly identified 42% of fraudulent event tickets and wrongly flagged 36% of genuine ones. Teaching the warning signs improved detection 25 to 30 percent, and almost 40 percent for tickets carrying misspelt text. It did nothing for tickets listing seats that do not exist, because the consumer has no way to check seating. The intervention moved the attribute the consumer can verify and not the one they cannot.