Agentic Commerce: What AI Shopping Agents Mean for Your Business

Yes, but the honest answer is narrower than the hype: AI shopping agents already drive a real and fast-growing share of retail traffic and sales, and the useful response right now is to make your product data machine-readable and trustworthy, not to rebuild your checkout around a protocol that is still consolidating. Salesforce estimated AI agents influenced more than 20% of global online retail sales during the 2025 holiday season, worth roughly $263 billion, and Adobe tracked AI-referred traffic to US retail sites up 693% year over year that same season, still growing 393% year over year by the first quarter of 2026. That is not a future prediction. That already happened. This article is what agentic commerce actually is, the numbers behind it, and the specific, unglamorous first steps that matter for most businesses today.

If you want your marketing and product data built to be found and trusted by both AI agents and human buyers, see how we do AI marketing automation.

What is agentic commerce, exactly?

Agentic commerce is a purchase where an AI agent, not a person manually browsing and clicking "buy," does the work of comparing options, choosing a product, and completing the transaction, usually on explicit instructions and spending limits set by the actual buyer. A shopper might tell ChatGPT "find me a waterproof jacket under $150 with good reviews and buy it," and the agent searches, compares, and checks out, all inside the chat, without the shopper ever landing on a retailer's website in the traditional sense.

This became a working reality, not a concept, in a tight window: Stripe and OpenAI launched ChatGPT Instant Checkout in September 2025, and Google followed with its own agentic payments and commerce standards within months. The underlying idea is not new (recommendation engines have nudged purchases for years), but letting the AI agent actually complete the transaction, with real payment credentials and real fulfillment, is what changed.

How does an AI agent actually complete a purchase?

Through a specific, secured checkout flow, not a workaround or a browser automation hack. Take ChatGPT Instant Checkout as the clearest example: a user asks for product recommendations inside ChatGPT, the agent surfaces options, and once the shopper is ready to buy, a Stripe-powered checkout appears inline in the chat. The buyer picks a payment method, and Stripe issues what is called a Shared Payment Token (SPT), a credential scoped to one specific merchant and one specific cart total. That token, not the buyer's actual card number, is what reaches the merchant. The merchant approves the order, calculates tax, and fulfills it exactly as they would any other sale.

This matters for a business owner for one practical reason: the buyer's payment details never pass through the AI agent itself, and a merchant already on Stripe can turn on this flow with what Stripe describes as one line of code. The security model was designed to make this an extension of existing checkout infrastructure, not a risky new integration from scratch.

Who built the standards behind agentic commerce?

Two open, competing-but-converging efforts, both launched within months of each other in late 2025 and early 2026:

• Agentic Commerce Protocol (ACP), built by Stripe and OpenAI, launched September 29, 2025, powering ChatGPT Instant Checkout. It started with Etsy sellers and expanded to more than one million Shopify merchants, with named brands including Glossier, Vuori, Spanx, and SKIMS.
• Agent Payments Protocol (AP2), built by Google, launched September 16, 2025 with more than 60 launch partners including Mastercard, PayPal, Coinbase, American Express, and Salesforce. AP2 uses three signed "Mandates" (Intent, Cart, Payment) that act as guardrails: the shopper sets spending limits and named brands or products up front, and the agent can only complete purchases that match those signed criteria.
• Universal Commerce Protocol (UCP), also built by Google and launched January 11, 2026, is a broader integration standard endorsed by more than 20 major names including Shopify, Etsy, Wayfair, Target, Walmart, Best Buy, Macy's, Mastercard, Stripe, Visa, and Home Depot. UCP is explicitly designed to work alongside AP2 for payments and alongside the agent-interoperability protocols A2A and MCP for agent communication, rather than forcing merchants to choose a side.

The practical read for a business owner: these are not two incompatible walled gardens fighting for exclusivity. Major retailers, payment networks, and even Stripe itself show up as partners across more than one of these standards, which is a strong signal the industry is heading toward interoperability rather than a format war.

How big is agentic commerce, actually, right now?

Big enough to already show up in the numbers, and growing fast enough that "wait and see" has a real cost. During the 2025 holiday season, Salesforce estimated AI agents influenced more than 20% of global online retail sales, roughly $263 billion in orders. Adobe's independent traffic data backs that up from a different angle: AI-referred traffic to US retail sites was up 693% year over year over that same holiday period, and the momentum did not fade after the holidays, Q1 2026 traffic from AI sources was still up 393% year over year, with March 2026 alone up 269%.

The behavior of AI-referred shoppers is also notably different from typical traffic, which matters more than the raw volume:

Retailers that deployed dedicated "shopper agents" of their own saw 59% higher sales growth (6.2% versus 3.9% for non-adopters), according to the same reporting. That gap is the kind of number that tends to close a debate inside a leadership team fast.

Is this just retail, or does it reach beyond consumer shopping?

It reaches well beyond consumer retail. Gartner's headline strategic prediction for 2026 is that 90% of B2B buying will be intermediated by AI agents by 2028, moving more than $15 trillion through agent-brokered purchases. Gartner separately forecasts that AI agents will outnumber human sellers by 10 to 1 by 2028, and that enterprise adoption of agentic AI will jump from under 1% today to 33% by 2028. If your business sells to other businesses, procurement and vendor comparison work is a plausible early target for this shift, not just consumer checkout.

How do you know if AI agents are already visiting your site?

Check before you guess, because for most businesses the traffic is arriving quietly, inside normal analytics, mislabeled as something else. AI-referred visits typically show up as direct traffic or as unfamiliar referrer domains (chat.openai.com, chatgpt.com, gemini.google.com, perplexity.ai) rather than a clean "AI agents" bucket, so if you have not built a segment for them, you are likely undercounting. Three practical checks:

1. Segment your referrers. In whatever analytics platform you use, build a segment for traffic from ChatGPT, Gemini, Perplexity, and Copilot domains, and watch it over the next full quarter rather than a single week, since the growth curve is what matters, not one snapshot.
2. Watch conversion and time-on-page for that segment specifically. Given Adobe's numbers (31 to 42% better conversion, longer time on page), a segment behaving unusually well is a strong sign it is agent-referred, even if the referrer data is incomplete.
3. Check whether your site is reachable at all. Fetch your own top product pages the way an agent would: no JavaScript rendering assumed, structured data extractable in the raw HTML. If your product details only render after client-side JavaScript runs, many agent crawlers will simply miss them, the same failure mode that hurts traditional SEO.

What mistakes are businesses making with agentic commerce right now?

The most common one is treating this as a checkout-integration problem before it is a data-quality problem. Teams jump straight to "should we integrate ACP or AP2" while their product pages still have missing prices, generic stock descriptions, and no structured markup at all, which means no agent can recommend them regardless of which checkout protocol they eventually support. A second common mistake is the opposite extreme: dismissing agentic commerce entirely as hype because the checkout volume still looks small relative to total sales. That misses the point of the traffic and conversion numbers: the shoppers arriving through AI agents already convert better and spend more time engaged than average traffic, so even a modest current share is a high-value channel worth capturing cheaply now.

A third, quieter mistake is inconsistency between systems. A product page that says "in stock" while the separate merchant feed says otherwise does not just confuse a human buyer occasionally, it actively teaches an AI agent to distrust your entire catalog, since agents cross-reference structured data against feeds and live pages. Fixing that consistency is unglamorous work, but it is exactly the kind of foundational fix that determines whether an agent recommends you at all.

Do you actually need to optimize your store for AI shopping agents now?

Start with the cheap, high-leverage work, and hold off on the expensive full-checkout integration unless you are already deep in ecommerce infrastructure. That is the honest, non-hyped answer, and it comes straight from where Adobe's own data points: most retail sites are not yet "machine-readable" enough for AI agents to parse reliably, which means the traffic and buying intent are already showing up while most competitors are not ready for it. Being ready earlier than your competitors, on the cheap part, is a real and currently available advantage.

Here is a realistic, staged first-steps checklist:

1. Fix your structured data first. Google's own developer documentation specifies the baseline: name, image, offers, price, priceCurrency, availability, and condition, implemented as JSON-LD schema.org markup on every product page. This is inexpensive, well-documented, and is the single biggest lever for whether an AI agent can even understand what you sell.
2. Make your data consistent, not just present. Google's crawlers now cross-check your on-page structured data against your separate product feed (like a Merchant Center feed). If the two disagree, an agent has reason to distrust both. Sync them.
3. Be citable, not just findable. Write product descriptions that answer the specific questions a buyer, or an agent acting for a buyer, would actually ask: "what is the battery life," "is this true to size," "what is the return window." Vague marketing copy is invisible to an agent looking for a factual answer to cite.
4. Keep pricing and availability accurate in real time. An agent that recommends a product and then hits stale inventory or a mismatched price does not just lose that one sale, it teaches the agent (and its provider's ranking signals) to trust your catalog less next time.
5. Only then consider checkout protocol support. If you are already on Stripe or Shopify, ACP and UCP compatibility may already be close to available to you with limited engineering work. If you run custom commerce infrastructure, treat full agentic checkout as a deliberate roadmap item, not an emergency, while the standards keep consolidating.

What agentic commerce does not change

Your website is still the source of truth. AI agents read from it, cite it, and route purchases through it, they do not replace the need for it. The businesses that lose out are not the ones without a checkout-ready AI integration, they are the ones whose product data is too thin, inconsistent, or marketing-fluffy for an agent to confidently recommend at all. If your site would fail a plain factual question like "is this in stock and what does it cost right now," fix that before worrying about which payment protocol to support.

It is also worth being clear-eyed about what agentic checkout does not solve on its own. A shopper who lets an agent buy on their behalf under a signed Intent Mandate still expects the product to match what was described, still expects normal returns and support, and still forms an opinion of your brand from that experience. Agentic commerce changes how the sale starts, not the standards a business is held to once it closes.

How to get started this quarter

Audit ten of your best-selling product pages against the structured data checklist above, and fix the gaps first, since this is the cheapest, fastest-payoff work and it is also what most competitors have not done yet, per Adobe's own findings. Then check whether your existing payment processor already offers ACP or AP2 support (Stripe and Shopify merchants often do, or will soon, with minimal extra work). Everything past that, custom checkout engineering, agent-specific merchandising, is worth revisiting once you have real traffic data showing AI-referred shoppers actually reaching your site, which, per the numbers above, is likely sooner than most teams expect.

If you would rather have this handled for you, that is our work. We audit your product and content data for AI readability, fix the structured data and citability gaps that keep agents from recommending you, and help you decide when full agentic checkout is actually worth building. Book a free consultation and we will show you exactly where your store stands today.