AI ads beat human designs in live Instagram campaign

On Instagram, background images promoting nature exploration left the design room and entered a live campaign for an outdoor activities company. The AI-generated portfolio recorded a 0.98% mean click-through rate, beating both a professional human designer and an aesthetics-focused AI benchmark.

The engine was not simply asked to make attractive pictures. Daviet of INSEAD and Nishimura of the University of Wisconsin–Madison used active learning to explore possible visuals while jointly optimizing predicted performance and alignment with the brand. Marketers vetted the images against brand standards before testing them in the campaign.

The comparison was clear: the human designer's portfolio averaged 0.65%, with a standard deviation of 0.31%; the aesthetics-optimized AI benchmark averaged 0.78%, with a standard deviation of 0.41%. The researchers' analysis found that the system's creatives beat the human designer's batch in 99.59% of samples.

The result also held up when the campaign was revisited 18 months later, during a high-stakes booking season and without retraining the model. The AI portfolio reached 3.38%, against 3.24% for the company's contemporary human-designed campaign, again with lower variance.

So what changes, concretely? For advertisers and agencies, the potential gain is not just faster image production: it is a way to search more deliberately among AI-generated options while keeping brand constraints in the loop. The evidence remains bounded by one company's live Instagram campaign and by the specific system tested, so the figures describe a demonstrated use case, not a universal replacement for human designers.