Consumer Reports (CR) is proudly presenting two prototypes as part of an Agentic Commerce event hosted by Skyfire at Visa’s San Francisco HQ, and sharing our vision for how shopping agents can provide trusted information to inspire the right purchase.
We’re seeing increasing interest in “agentic commerce”: shopping experiences through which consumers rely on AI-powered tools to not only help them research products, but also make purchase decisions and carry out transactions. The data shows that consumers are already using ChatGPT and other generative AI experiences to solicit product recommendations and make major purchases.
At CR, we’re excited to put new technologies to work to make shopping even more empowering and information-rich for consumers. AI experiences in the future could allow us to discover, select, and purchase products within the same interface, and with minimal friction.
When discovery, selection, and purchase all happen in one place, maybe even through a single prompt, the need for consumers to trust the systems acting on their behalf becomes all the more critical. The prototypes we’re sharing offer two concepts for embedding consumer trust in AI-driven purchase journeys.
Prototype 1: Buy It For Me — Agentic Purchase via AskCR
In the first demo, a consumer uses CR’s expert-powered advisor, AskCR, to find and purchase a stainless steel pan. AskCR identifies a top-rated product based on CR’s independent testing and mission-driven evaluation criteria. When the consumer decides to move forward with the purchase, they tap “Buy It For Me” and proceed to enroll their credit card. From then on transactions can be executed seamlessly by the agent, without the consumer needing to visit the retailer’s site or enter payment information manually.
This prototype is for demonstration purposes only, and not publicly available.
Technically, this is enabled by a combination of Visa’s Intelligent Commerce infrastructure and Skyfire’s secure payment credentialing. The consumer’s Visa card has been tokenized, bound to their device, and authorized to complete purchases in a privacy-preserving way. CR’s agent makes the purchase using a headless browser, acting on the consumer’s behalf to complete the purchase flow. The agent confirms the transaction and provides delivery details.
While simple for the user, this interaction requires a complex orchestration of trust under-the-hood. CR serves as the product advisor, Visa enables payment settlement, and Skyfire allows AskCR to access merchant sites using verified identity credentials and make purchases using the Visa payment information. It’s a first step toward demonstrating what an end-to-end trusted purchase flow might look like in agentic commerce.
We’re already thinking about how to get this functionality to market, as it will require new guardrails, consent frameworks, patterns for agent authorization and permissioning, and novel experience design to work in a production setting. We’re also tracking ongoing developments in the space, like Google’s new open protocol for purchases initiated by AI agents.
Prototype 2: Pay for Trust — Embedded Trust Signals via CR’s API
In the second demo, a consumer is using a home renovation agent called “KitchenShop” to research blenders. They’ve narrowed down their options but remain uncertain about which blender to purchase. They ask the KitchenShop agent to pull in blender ratings to help them make their decision.
At this moment, KitchenShop connects to CR’s MCP server and discovers that CR has ratings available for all of the blender options under consideration. KitchenShop then executes a micro-license to use the CR rating in its answer, paying CR and retrieving the rating before presenting it in its response to the consumer. Armed with information about the highest performing blender, the consumer can then make their decision and complete the transaction with confidence.
This prototype is for demonstration purposes only, and not publicly available.
This demo highlights a new role for CR: not just as a trusted agent, but as a provider of trusted inputs into others’ agentic experiences. It reflects the idea that trusted information can be discovered and licensed in real-time, right when it’s needed. Skyfire once again powers the identity and payment layer, while CR’s role is to stream our verified data to the agent in order to more effectively guide and advise the consumer at the point of decision.
Some questions remain about how CR’s MCP server should be discovered and queried in practice, and we look forward to continuing to develop this concept with AI-forward companies who recognize the value of CR’s brand and data.
Building Toward an Agentic Economy That Serves People
While these concepts represent early R&D that we’re cooking up in our labs, we believe they give a glimpse of where agentic commerce is heading.
The demos also raise important questions: how do we ensure complex agentic systems are aligned with consumers’ preferences and interests? How do we ensure agents are loyal to users, rather than advertisers, affiliates, or optimization targets? What kind of infrastructure is needed to preserve consumer intent as transactions are processed through multiple agents? And how might we embed verified, high integrity information to power shopping experiences across the agentic web?
Through CR’s work on trust signals, payment protocols, and ongoing research collaborations on loyal agents with Stanford’s Digital Economy Lab and others, we hope to greatly enhance consumer agency so that AI helps consumers make the best choices.
We’re thrilled to be working with Visa and Skyfire to envision these first bridges. If you’re thinking about trust in agentic commerce and want to build alongside us, we’d love to talk. Drop us a line at innovationlab@cr.consumer.org.
And many thanks to the teams at Carnegie Mellon’s Human Computer Interaction capstone program, Ocupop, Visa, and Skyfire who helped contribute to these prototypes.
