The Race To Agentic Payments In US B2C E-Commerce: Where We Are Now

In this blog, we answer five key questions about agentic payments. We also want to hear from you! What questions should we tackle next?

Question #1: WHAT
What are “agentic payments” in B2C commerce? (Spoiler alert: They don’t exist yet.)

You understand agentic AI and agentic commerce, but in order for agentic commerce to scale, our commerce infrastructure must support agentic payments. So what are agentic payments exactly, and are they actually occurring yet?

• Agentic payments do not exist yet because “agentic” requires autonomy. Agentic payments are financial transactions initiated and executed autonomously by an AI agent on behalf of a consumer (i.e., without explicit human intervention).

Status: currently hypothetical. Agentic payments are on the roadmaps of the entities (e.g., answer engines, payments processors, banks, payments networks) that are building the necessary capabilities, infrastructure, standards, and protocols for them to occur. What’s the holdup? Well, agentic payments challenge the core principles of our established online payment processing system — such as authentication and authorization — and trigger alarm bells regarding data security/privacy and liability, as well as fears of disintermediation and fairness of play by merchants and payment networks.

• AI-agent-assisted payments, or “agent-ish” payments, are starting to happen. AI-agent-assisted payments are financial transactions initiated or executed by AI agents on behalf of a customer with explicit human involvement (for example, a prompt and authorization from the customer). As such, they are not agentic; they’re agent-ish.

Status: currently limited. Today, an AI agent can gather information, add products to a cart on a merchant’s website, and populate the checkout with the consumer’s data, but it cannot safely or compliantly complete the checkout using the consumer’s payment data. It must either direct the consumer to complete the checkout themselves, or it will use a commercial card issued to the AI platform by a payments company to do so. AI-agent-assisted payments were first explicitly processed by the payment ecosystem in 2024 on merchants’ websites and on the answer engine Perplexity in the US. As of 2025, OpenAI now also supports AI-agent-assisted payments in ChatGPT in the US. But even before 2024 (and still today), brave early-adopter consumers have entrusted their unencrypted credit card details with AI agents and prompted them to make payments on their behalf.

Question #2: HOW
How are AI-agent-assisted payments happening in US B2C commerce today?

Frankly, it’s impressive that a consumer can check out on ChatGPT or Perplexity at all. That’s because agent-ish payments — and the forthcoming agentic ones — challenge the fundamental principles of our established online payment processing system (e.g., authentication, authorization). The fact that the payment isn’t the hardest part of the process is testament to how adept the payments technology and service providers have become at filling gaps in the broader payments ecosystem (e.g., as they’ve done by supporting real-time payments before banks and governments did).

AI-agent-assisted payments, like standard online payments, can be broadly categorized as happening in two ways: in closed-loop and open-loop scenarios. The difference is in whether the entity that owns the UI where the consumer’s payment intent is captured is the same or different than the entity that owns the product or service being purchased. Merchants may need to support both: that is, via AI-led guided selling on their websites and/or commerce executed through answer engines such as ChatGPT or Google Gemini.

• Closed-loop AI-agent-assisted payments enable commerce on owned site experiences. Retailers and brands are beginning to incorporate guided selling experiences that are augmented by AI agents. In 2024, Michael Kors, in partnership with Dynamic Yield (a Mastercard company), became the first notable example of an AI shopping agent capable of carrying out payments on an owned site experience (i.e., on MichaelKors.com), though the experience has since been dialed back to an AI assistant for product discovery.

How: Because these agent-ish payments are happening on the digital experiences that the merchant owns, payments typically occur through established e-commerce experiences and card-not-present processes. Merchants are beginning to embed AI-assistive shopping experiences on their sites. But as we’ve demonstrated previously, the experiences are immature and “[for merchants and] end customers, they create negative experiences at worst and vaguely helpful experiences that generate some friction at best.”

• Open-loop AI-agent-initiated payments start on a non-owned experience. Specifically, this is happening in a limited capacity in the US today on answer engines: OpenAI’s ChatGPT has launched what it calls Instant Checkout, and Perplexity has its Buy With Pro capability. Both experiences enable consumers to shop directly in the answer engine’s UI from the results to their prompts (i.e., the AI agent buys products or services sold by third parties surfaced by the answer engine).

How: It is changing very quickly. Today, the answer engines are using their payment partners to issue virtual commercial cards to make purchases from the merchants and funding those cards with consumers’ payment credentials, much like how an online travel agency such as Booking.com or Expedia would. The details of how payments are handled and which entity takes on liability vary depending on how the answer engine and its payments partners have designed the experience and the level of merchant integration. This card-on-card approach may be temporary until the card networks (or others, like Google) mature their mechanisms and standards for AI-agent-assisted and agentic payments.

Question #3: WHO
Who are the AI shopping assistants supporting open-loop agent-ish payments?

It’s a three-horse race so far.

• OpenAI’s Instant Checkout in ChatGPT is the horse to beat in this race. ChatGPT is the answer engine with the greatest market awareness and adoption. Instant Checkout is supported by the open-source Agentic Commerce Protocol (ACP) that OpenAI built with Stripe. Today, Instant Checkout is limited to the US and to single-item (i.e., not whole-cart) purchases from its launch partner, Etsy, using Stripe’s fraud screening and tokenization (processing payments through Stripe is not required). OpenAI has said that Shopify merchants will be able to onboard next. And Walmart and PayPal have made announcements of their intentions to adopt ACP. OpenAI will charge merchants fees for participating in Instant Checkout and will rank products based on merchant participation and integration.
• Perplexity’s Buy With Pro was the first to market. Perplexity was the first horse out of the gate when it launched Buy With Pro in the US in November 2024. But it’s behind both ChatGPT and Google Gemini in terms of market awareness and adoption, so it may be quickly outpaced by them. Perplexity made announcements about partnerships and forthcoming integrations with Shopify and with PayPal, but neither have materialized substantially. Perplexity is, for now, acting as the merchant of record (MOR) and therefore takes on liability for these AI-agent-assisted transactions, whereas with Instant Checkout, the merchant is the MOR and thus has more control over processing but also more liability. As it stands, Perplexity is not charging merchants fees to participate in Buy With Pro.
• Google Gemini has all the right pieces to win the race. The horse in the race with the most going for it, and the least to show for it so far, is Google’s Gemini. Its market adoption and awareness lag ChatGPT’s, but Google is still the destination for shopping-related search. And Google’s other resources put Gemini in a position to leapfrog the other answer engines as an AI shopping assistant in non-owned (or open-loop) contexts. Unlike the others, Google already receives merchant product data feeds for Google Shopping (so it doesn’t have to scrape for the data nor require new merchant integration). And its other resources such as hardware, browsers, operating systems, and consumer products can together create elegant authentication, authorization, and ongoing consumer and merchant management of settings and controls. Google also has proven experience managing open source software, something the others lack, and it has garnered goodwill from the payments ecosystem for participating in the World Wide Web Consortium. Despite announcing its Agent Payments Protocol (AP2) in September 2025, AI-assisted payments through Gemini are still more theoretical than real.
• Honorable mentions: The card networks (e.g., Mastercard, Visa) are building crucial infrastructure for agentic payments (e.g., AI agent registries, AI-agent-specific tokenization), alongside content delivery networks (e.g., Akamai, Cloudflare) to manage agentic traffic and security.

Question #4: WHERE
Where are we now? (alternatively: why agentic payments are still a future state)

Per Forrester’s data, roughly one in 10 US online adults have used OpenAI’s Instant Checkout or Perplexity’s Buy With Pro as of October 2025. Still, in the last two to three years, we went from no usage to 33% of US online adults using an AI agent to find and compare products. This trajectory could continue remarkably fast. The technology providers are proverbially “building this plane while flying it,” and the race has really just begun. It’s confusing that vendor announcements are going out much faster than usable products are being launched, but it means that merchants have time to decide their approach. See our timeline of key agentic payments announcements at the end of this blog.

Question #5: WHY
Why are the experiences mostly terrible? Because we don’t have a lot of answers yet.

So many important questions remain unanswered that will significantly impact the momentum and trajectory of agentic payments. Some are big-picture: Will merchants participate? How will merchants protect their customer relationships? When will industry standards be established and by whom? And some are tactical: How do we best validate that the agent is acting legitimately on behalf of a legitimate customer? How do we best determine that the AI agent has the authority to make the purchase and that the purchase details match the consumer’s intent? How will liability be determined and the financial repercussions of fraud or disputes handled?

And the list goes on. Which questions do you want Forrester to address next?

Stay tuned for forthcoming research on agentic payments from Forrester for B2C and B2B commerce. To discuss agentic commerce or agentic payments in more depth, please book a guidance session or inquiry with one of your Forrester analysts on the digital business strategy team!

A timeline of key open-loop (or non-owned) AI-agent-assisted payments announcements: