Attributing digital election advertising to races and candidates: a multimodal approach to measuring outside group influence

Outside groups have become increasingly central to U.S. election campaigns since Citizens United, but their true reach remains opaque: gaps in reporting requirements and limited ad library data obscure how much they spend and which races they target. This gap limits our ability to evaluate their role in shaping electoral competition and influencing policy. We develop a scalable multimodal system that attributes digital ads to contests by identifying referenced candidates in text and images, aligning with FEC rules. Applied to 1.4 million Meta ads from 2020, our system leverages both text and images to yield more comprehensive estimates of outside-group spending than either modality alone, reallocates tens of millions in spending, and enables ad tone analysis. Our framework provides generalizable, race-level attribution and supports research on outside-group influence and campaign strategy.