AI Shopping Agents Complete Purchases 57% of Time
Visa's test of 22 agents shows wide variation in reliability, with liability for unauthorized buys still unresolved.
Updated on Nov. 5, 2026 in Artificial Intelligence
Live Poll
Would you trust an AI agent to make purchases for you without checking?
Visa tested 22 AI shopping agents and found they completed purchases successfully in about 57% of trials. Visa published the results on a website that lets consumers try the agents included in the test. The work is a company-run evaluation of existing agents, not a shipped consumer standard.
Why it matters
The test measures whether agents can learn what a shopper wants when the person does not state every requirement up front, a core capability for agent-driven commerce. Results show wide variation across agents, suggesting reliability is still a differentiator.
The evaluation was a company-run test of 22 agents, each given 15 shopping runs across televisions, laptops and hair appliances. Successful purchases averaged 57% overall, with the best two agents at 82%, the Claude agent at 76%, and the OpenAI agent at 64%. Who bears liability for unauthorized purchases or fraud is not yet settled.
The players
Visa
Global payments network whose test evaluated third-party AI shopping agents for purchase accuracy.
Claude
Anthropic's AI model, which powered one of the tested shopping agents scoring 76%.
OpenAI
AI developer whose agent ranked sixth in the Visa test with a 64% success rate.
The details
Each run asked an agent to complete a purchase, and tasks were designed to see whether it could draw out missing requirements through conversation within a reasonable time. Failures included choosing an unsuitable product or not surfacing a suitable match. The results were published on a site where consumers can try the tested agents.
Timeline
Visa published the AI shopping agent test results on November 5, 2026.
The Tech Race
Visa's test puts a measured baseline on competing shopping agents, with OpenAI and Anthropic-powered agents among the tested field. The gap between the top agents at 82% and the middle of the pack shows that agent reliability is becoming a basis for competition.
Shoppers who hand purchases to AI agents face roughly even odds of a clean completion based on this test, so checking the final order matters. Availability and pricing of the tested agents for everyday consumers were not stated in the source.
The takeaway
Watch whether agent success rates rise across future tests, since the 57% baseline is the number to beat. Consumers should review agent-placed orders carefully until liability for unauthorized purchases is clarified.
Further reading
For broader context on how these tools are reshaping online buying, explore our Artificial Intelligence coverage.
Live Poll
Would you trust an AI agent to make purchases for you without checking?
