Political campaigns increasingly run experiments to learn what persuades voters, but the consequences for elections and democratic influence remain unclear.

🧾 What the Archive Contains

• 146 advertising experiments run by U.S. campaigns in 2018 and 2020 using the Swayable platform
• 617 unique advertisements produced by 51 campaigns
• Over 500,000 respondents who evaluated the ads
• Analysis of the complete archive, avoiding publication bias

🔎 What Was Measured

• Variation in the persuasive effects of advertisements across the full archive
• The predictive power of common theories about what makes advertising persuasive

📌 Key Findings

• Persuasive effects vary: there is small but meaningful variation in how much different ads persuade voters
• Predictive limits: common theories about advertising have limited and context-dependent power to predict which ads will be persuasive

🛠️ Why It Matters

• Experiments reduce uncertainty about which messages work, but it is difficult to predict ex ante which ads will persuade
• Because experiments identify effective ads, campaigns that can afford to test and then deploy those ads at scale gain an advantage
• These patterns suggest that experimental learning can compound money’s influence in elections by turning uncertain creative differences into actionable, scalable advantages