i ran every online ad in the alaska senate race through a model that flags manipulation techniques
i've been collecting every facebook/instagram and youtube ad in the alaska senate race and running the text through a model that flags manipulation techniques (loaded language, name calling, appeal to fear, that kind of thing). it doesn't say whether an ad is true, just how it's worded.
142 ads so far, 93 of them (65%) use at least one. that's the highest of the 13 senate races i'm tracking.
share of the 142 ads with each technique:
- loaded language: 46% (66 ads)
- appeal to prejudice: 42% (59)
- name calling: 28% (40)
- false urgency: 27% (38)
- glittering generalities: 24% (34)
- black-and-white fallacy: 22% (31)
- card stacking: 20% (28)
- conspiracy appeal: 13% (19)
the words that keep showing up in the flagged phrases: "liberal" (20 ads, e.g. "lower 48 liberals") and "sold" (10 ads, e.g. "sold Alaskans out").
biggest outside spender per fec filings: SLF PAC, $8.2M opposing Peltola.
every ad is on one page with what got flagged and who paid for it: https://semblen.com/races/ak-senate?ref=reddit
online ads only, no tv. the model gets things wrong sometimes, so each ad shows the exact words it flagged.