AI personality pairing improves ad quality; text quality lifts clicks on X

For 40 minutes, more than 1,200 participants from the U.S. worked with AI agents to create display ads for a year-end report from a real think tank. The match between person and machine mattered: Sinan Aral of MIT Sloan School of Management and Harang Ju of Johns Hopkins University found that some human-AI personality pairings produced markedly better ads than others.

The experiment used Pairit, a platform developed by Aral and Ju, to randomize the teams. Each AI agent was prompted to independently exhibit high or low levels of the Big Five personality traits — openness, conscientiousness, extraversion, agreeableness and neuroticism. Participants and agents collaborated through real-time chat and synchronized text- and image-editing tools, while a separate group of approximately 1,100 people rated the resulting ads.

The highest-rated text and/or images came from extraverted humans with extraverted AI, conscientious humans with conscientious AI, and open humans with conscientious AI. Some combinations moved in the opposite direction: extraverted humans with conscientious AI, conscientious humans with agreeable AI, and neurotic humans with conscientious AI appeared to significantly reduce ad quality.

The researchers then took the ads beyond the session. They ran them on X for a two-week period, during which they received approximately 5 million impressions. The ratings tracked market performance: text quality increased click-through rate by about 7% and decreased cost per click by about 30 cents.

Concretely, a business could use personality personalization to match an AI agent to the person using it for this kind of marketing task. That could make the same kind of marketing workflow more effective without changing the task itself. The boundary is clear, however: these findings come from a randomized experiment focused on ad creation, and Aral and Ju plan to study whether the pairings hold in sales, engineering and health care.