Visa tested 22 AI agents on shopping tasks for a TV, laptop and hair appliance. Across the experiment, agents completed purchases successfully about 57% of the time.
Key points
Google’s Gemini Aragon 4 and Meta’s Wonder agents each had an 82% success rate
A Claude agent scored 76%; an OpenAI agent scored 64%
Visa ran 15 shopping expeditions for each of the 22 agents
Fortune reports that Visa evaluated the agents on whether they could identify and buy products that met shoppers’ needs. The tasks included missing or incomplete information, which agents had to elicit through interaction; some bought unsuitable products or failed to suggest an appropriate option.
Visa’s results site lists the outcomes for each agent. The five most expensive agents succeeded an average of 67% of the time, compared with 50% for the five cheapest. Visa also launched a site where consumers can interact with the agents used in the experiment.