# Visa’s AI Shopping Agents Failed in More Than Four in 10 Tests

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/49284
Published: 2026-11-05T12:57:57.000Z
Updated: 2026-11-05T12:57:57.000Z
Section: Technology

> Visa ran 15 shopping expeditions for each of 22 agents. Two tied for the top score at 82%, while an OpenAI agent scored 64%.

Visa’s test of 22 AI shopping agents found they completed purchases successfully in about 57% of trials, highlighting how often the tools struggled to turn shoppers’ incomplete preferences into suitable orders.

The test covered televisions, laptops and hair appliances. Visa ran 15 shopping expeditions for each agent. Unsuccessful runs included cases where agents selected an unsuitable product or failed to suggest a suitable match.

Two agents tied for the highest success rate at 82%. A Claude agent scored 76%, while an OpenAI agent ranked sixth at 64%.

The tasks were designed to test whether agents could learn what a shopper wanted when the person did not provide every requirement up front. In a television scenario, preferences included compatibility with a child’s gaming console, Dolby Vision and a 30-day return window. Agents had to interact with the shopper and draw out missing details within a reasonable time.

Success also varied with the cost of running a shopping inquiry. The five most expensive agents averaged a 67% success rate; the five least expensive averaged 50%.

Visa Global Head of Growth Rubail Birwadker said the technology and trust needed for consumers to delegate shopping to agents are still developing. “We’re not there yet from a trust or technology perspective. But we’ll soon get into a world where consumers can trust and delegate,” he said.

Product discovery is one part of agent-led shopping. Visa and Mastercard are working on payment tools and merchant education for agentic commerce. Responsibility for unauthorized purchases and fraud by shopping agents also remains unresolved.

Visa published the results on a website that lets consumers try the agents included in the test.