Visa tested 22 AI models as shoppers, and they failed 42.8% of the time

Visa conducted its Agentic Commerce Benchmark testing 22 AI models on 15 shopping tasks, finding they failed 42.8% of the time. Top performers from Google and Meta scored around 82%, while others like Claude and OpenAI scored lower. Success correlated with how well models uncovered hidden customer needs and their cost per run.