How MarketingBench Works
An independent, open benchmark for AI marketing quality. Here's the methodology behind the rankings.
Blind evaluation
Two anonymous AI models receive the same marketing task with the same system prompt. Voters see only “Response A” and “Response B” with no model names, no providers. You pick the better output. Only after voting are models revealed.
Same prompt. Same instructions. No labels. Pure output quality.
How ratings work
We use Elo ratings, the same system used in chess. When a model wins a matchup, it gains points. When it loses, it loses points. The amount gained or lost depends on the strength of the opponent: beating a top-ranked model is worth more than beating a lower-ranked one. Every model starts at 1200. Ratings include 95% confidence intervals so you can see how stable each ranking is.
Why marketing-specific
General AI benchmarks test code generation, math reasoning, and multi-turn chat. They tell you very little about which model writes the best subject line, the most clickable ad headline, or the most on-brand push notification.
MarketingBench tests models on real marketing tasks: copywriting, search ads, push notifications, SMS, and translation. Format constraints mirror actual production requirements. Every model gets the same system prompt for each task type.
Full transparency
Every model gets the exact same system prompt. You can read them all on the System Prompts page. Vote counts, win rates, and Elo ratings are public on the Rankings page. There is no weighting, no editorial selection, and no pay-to-play.