Summary
Computational ad analysis is the academic-research discipline of applying text-classification, image-classification, and content-analysis methods to large political-ad corpora — broadcast, cable, social-platform, and search-engine ad transparency streams. It produces the empirical evidence on what political ads look like in volume: framing, emotional appeals, issue frames, sponsor concentration.
Body
The discipline sits at the intersection of political communication, computer science, and electoral research. The Wesleyan Media Project (WMP) anchors the field with its open dataset on broadcast-cable and social-platform political-ad spending, content, and target audience, and the CREATIVE project that adds machine-readable ad-creative metadata (the actual ad scripts, visual frames, and content). The Wesleyan-Media-Project GitHub publishes CSV exports for the academic community. [source: wesleyan-media-project]
The AdImpact-Wesleyan Media Project partnership layers broadcast/cable signal accuracy onto social-platform transparency data — the combined stream produces a cross-platform dataset usable for content analysis of US political advertising across cycles. The data are the empirical basis for cross-cycle comparison studies: are campaigns spending differently on broadcast vs. social platforms; which issue frames dominate; how do emotional vs. policy appeals distribute; what is the concentration of spend by sector. [source: wesleyan-media-project]
The methodological challenge for computational ad analysis is that transparency data from platforms is partial (Facebook and Google historically release some transparency streams; TikTok and others vary), that ad-creative-text scraping is brittle (creatives change, become inactive, get pulled for violation), and that scaling content-analysis labor across thousands of ads requires either supervised ML labeling or crowdsourced labeling pipelines (Galway, Young, Pompilio).
The recent emergence of generative AI in political advertising (AI-generated audio, video, image, and text in ads) adds a new methodological frontier: detecting synthetic-media political advertising requires its own detection pipeline and disclosure regime. IPIE (the Institute for Public Integrity and Election) research on generative-AI in US campaigns is the leading frame for this frontier in 2025–2026. [source: wesleyan-media-project]
Computational ad analysis treats the political-ad ecosystem as an observable, measurable system. The output is publishable, citable empirical evidence on what campaigns actually do, not what campaigners say they do — the distinct research contribution of the field.
Use it for
Citing the empirical literature on US political-ad spend, concentration, and creative content; designing a research project using the WMP or CREATIVE datasets; understanding the transparency-disclosure status of social platforms; writing data-journalism or academic papers on the contemporary ad ecosystem.
Related
Open Questions
- The generative-AI frontier: how will academic detection pipelines keep up with the pace of generative-AI ad innovation? What disclosure regime would make the pipeline unnecessary?
- The cross-platform-generalisation question: are methods developed on US broadcast+cable+Facebook/Google data portable to other countries and other platforms, or do they need to be re-derived per jurisdiction?