eCommerce

Agentic commerce: getting your store ready for AI shopping agents

10 min readBy Jaskirat Singh Talwar

As of October 2026, AI shopping agents mostly do discovery, not checkout. ChatGPT launched in-chat "Instant Checkout" with Stripe in September 2025 and scaled it back in March 2026 to focus on product discovery, sending shoppers to the merchant's own checkout. Google's Universal Commerce Protocol (UCP), announced in January 2026 with Shopify, Walmart, Target and others, is the second open standard aiming at the same flow. What a store needs today is the same in both worlds: a clean, complete product feed, product structured data that matches the page, crawlable product pages, accurate policies, and a checkout that works for a referred visitor. The agent-specific protocols sit on top of that and are mostly handled by your platform.

Key takeaways

  • Two open, Apache-2.0-licensed protocols exist: the Agentic Commerce Protocol (ACP, OpenAI and Stripe, September 2025) and the Universal Commerce Protocol (UCP, Google with Shopify, Etsy, Wayfair, Target, Walmart and others, January 2026). In both, the merchant stays the merchant of record and keeps the customer relationship.
  • OpenAI refocused ChatGPT shopping on discovery in March 2026 after limited merchant uptake of in-chat checkout; merchants can now use their own checkout. Treat agent checkout as optional and discovery as the main event.
  • Semrush's six-month study found AI Overviews grew 71% on commercial-intent searches but fell 5% on transactional ones: research is moving into AI answers, the purchase is still mostly on your site.
  • Google's advice for AI features is to keep Merchant Center data current, keep structured data consistent with the page, and allow crawling. Product structured data plus a Merchant Center feed "maximizes your eligibility".
  • Your platform does most of the protocol work (Shopify's agentic storefronts, Stripe's seller integrations for ACP and UCP). Your job is data quality, policy accuracy and measurement.

What "agentic commerce" means in practice

An AI agent is software that takes actions on a person's behalf. In shopping, the actions are: find products that match a need, compare them, build a cart, and, in some implementations, pay. Three layers are involved:

  1. Discovery: the assistant (ChatGPT, Google AI Mode or Gemini, Perplexity, Copilot) needs to know your products exist, what they cost, whether they are in stock, and what your shipping and return terms are. This comes from crawling your pages and from structured feeds.
  2. Transaction: a protocol that lets the assistant start a checkout with you, pass payment credentials securely, and receive order confirmation. This is what ACP and UCP standardise.
  3. Post-purchase: order status, tracking, returns, queried by the assistant on the customer's behalf. UCP's published scope includes orders and identity linking; most live implementations stop at discovery.

Stripe's documentation for sellers now frames the choice as "Sell through agents" using "UCP or ACP", with the agent side of its product still in private preview. That is a fair summary of the market: the plumbing exists and keeps changing; the demand signal is discovery traffic.

The two protocols, without the marketing

ACP (Agentic Commerce Protocol)

Developed by Stripe and OpenAI, published as "an open-source standard" under the Apache 2.0 licence. The specification covers a checkout interface that businesses implement so agents can start a checkout, and secure handling of payment credentials. The merchant "maintain[s] their customer relationships as the merchant of record", controls what is sold and how products are presented, and fulfils the order. Integration is through a REST API or MCP; any compatible payment processor can be used, with Stripe first via its Shared Payment Token.

What happened with it: OpenAI launched "Instant Checkout" in ChatGPT in September 2025 for US users, starting with Etsy sellers and promising Shopify merchants. Retail press reported in March 2026 that OpenAI was scaling back in-chat checkout after weak uptake, letting merchants use their own checkout flows while ChatGPT focused on discovery, comparisons and conversational filtering. The protocol still exists; the in-chat purchase is no longer the default path.

UCP (Universal Commerce Protocol)

Announced by Google in January 2026 and published at ucp.dev as "an open standard" under Apache 2.0, "built by the industry". Its documented capabilities: discovery (catalogue search and lookup), cart, checkout (unified checkout sessions), identity linking via OAuth 2.0, orders (status, tracking, returns) and payments via the Agent Payments Protocol (AP2), with each authorisation backed by cryptographic proof of user consent. Named co-developers for shopping include Google, Shopify, Etsy, Wayfair, Target, Walmart, Amazon, Microsoft, Meta, Salesforce and Stripe. Businesses "remain the Merchant of Record" with "full ownership of customer relationships". Lodging has a draft specification; food is "coming soon".

For a mid-size store, the practical difference is small: Shopify and Stripe support both, and Google's own AI Mode checkout is built on UCP.

What the platforms do for you

  • Shopify. Trade coverage in March 2026 reported Shopify switching on agentic storefront features by default for eligible US merchants, with catalogue data exposed to assistants including ChatGPT; later coverage lists Google AI Mode, Microsoft Copilot and Perplexity, each with its own eligibility rules (US customer base, guest checkout, completed store policies). Check the Shopify admin for the agentic/AI channel settings and read Shopify's own documentation for the current list; eligibility and defaults have changed several times this year.
  • Adobe Commerce (Magento). Live Search, the SaaS search service included in the licence, provides "AI-powered dynamic faceting and re-ranking of search results in response to in-session shopper behaviors" and a merchant-maintained synonym dictionary. That improves on-site discovery; for external agents, the feed and structured data work below is what matters.
  • WooCommerce. No native agent channel at the time of writing; the route is the Merchant Center feed, product structured data, and (if you want checkout) Stripe's seller integration.

Whatever the platform, the assistant can only present what your data says. The readiness work is data work.

The 10-point readiness checklist

  1. Product feed complete and current. Every sellable product in Google Merchant Center with price, availability, GTIN/brand/MPN where they exist, shipping and return policies set at account level. Google's guidance for AI features explicitly says to keep Merchant Center information current. Stale availability is the fastest way to be dropped from an answer.
  2. Product structured data on every product page, matching the page. Google's Product documentation distinguishes product snippets (pages where people cannot buy) from merchant listings (pages where they can) and recommends adding structured data and a feed, since doing both "maximizes your eligibility to experiences". Google also warns to make sure "structured data matches the visible text on the page".
  3. Crawlable product and policy pages. Allow Googlebot, OAI-SearchBot, PerplexityBot and Claude-SearchBot in robots.txt and at the CDN. Blocking "AI bots" wholesale in 2024 removed many stores from 2026 answers.
  4. One canonical URL per product. Variant and filter URLs canonicalised; no duplicate product pages across collections.
  5. Policies written as answerable text. Shipping times by region, return window, restocking fees, warranty, in plain sentences on a crawlable page. Assistants quote these when a shopper asks "can I return it".
  6. Attributes a shopper would filter by. Material, size, compatibility, dietary flags, certifications, in structured fields, not only in a description paragraph. Agents filter on data.
  7. Guest checkout and a referred-visitor path that works. With discovery as the main channel, the shopper arrives from an assistant with a product in mind. Deep links must land on the product, not the home page; checkout must not demand an account.
  8. Reviews that are real and marked up correctly. Aggregate ratings in structured data only where the page shows the reviews; fabricated or imported-without-consent reviews violate platform policies and will eventually be detected.
  9. Measurement. A GA4 channel group for AI referrers (chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com), UTM discipline on any feed-driven links your platform lets you tag, and a monthly check of orders attributed to AI sources in your platform's reports.
  10. A decision on agent checkout. Enable platform-provided agent checkout only once the first nine points are done and you can see referred orders converting; otherwise you add a surface to support with little volume.

What to expect from the traffic

Be realistic about scale. Semrush measured AI Overviews on more than 600,000 US keywords between November 2025 and April 2026: they "grew an average of 71% across search engine results pages (SERPs) with commercial intent" and "decreased by 5%" on transactional queries. Shoppers researching ("best waterproof hiking boots for wide feet") increasingly get an AI answer that cites products and merchants; shoppers typing a product name to buy it still get a classic results page. The win is being the merchant cited at the research stage, with a product page that converts the referred click.

Retail press in 2026 also reported that in-chat checkout converted worse than merchants' own sites and that only a small number of Shopify merchants went live with it before OpenAI's pivot. Those are secondary reports, not published data, but they match the direction the platforms took.

A 30-day plan for a store

  • Week 1: audit the feed (disapprovals, missing identifiers, stale stock), run the product-page structured data through Google's Rich Results Test, check robots.txt and the CDN for crawler blocks, read the shipping and return pages as a stranger would.
  • Week 2: fix feed and markup; rewrite policies as plain answers; add missing attributes for the 50 best-selling products.
  • Week 3: deep-link QA (product URLs from the feed land correctly on mobile), guest checkout QA, set up the GA4 AI-referrer channel group.
  • Week 4: enable the platform's agentic channel settings where eligible; run 20 shopping prompts across ChatGPT, Google AI Mode, Perplexity and Copilot and record whether your products appear and which page is cited; repeat monthly.

FAQ

Can customers buy my products inside ChatGPT right now?

As of October 2026, mostly not by default. OpenAI launched in-chat Instant Checkout in September 2025 and in March 2026 refocused on discovery, letting merchants use their own checkout. ChatGPT shows products and comparisons and sends shoppers to your site; the Agentic Commerce Protocol still exists for merchants who integrate it.

What is the difference between ACP and UCP?

Both are open, Apache 2.0 protocols that let an AI agent start a checkout with a merchant while the merchant stays the merchant of record. ACP was developed by Stripe and OpenAI (2025) and is used by ChatGPT; UCP was announced by Google (January 2026) with Shopify, Walmart, Target, Etsy and others and powers checkout in Google AI Mode and Gemini. Shopify and Stripe support both.

Do I need to do anything special for my products to show up in AI answers?

No new file or markup is required. Google's guidance is to keep Merchant Center current, keep product structured data consistent with the page, and allow crawling. For other assistants, allow their search crawlers in robots.txt and keep product and policy pages in plain, crawlable text.

Will AI agents replace my website?

Not on current evidence. Research-stage queries are increasingly answered by assistants, while purchase-stage queries still go to classic results and merchants' own checkouts, and in-chat checkout saw limited uptake in its first six months. The website remains the place the order happens; the agent is a new referral source.

Which platform is best placed for agentic commerce?

Shopify has moved fastest on turning catalogue exposure on by default for eligible US merchants, and both Shopify and Stripe support ACP and UCP. Adobe Commerce and WooCommerce stores can achieve the same discovery outcome through Merchant Center feeds and product structured data; agent checkout there depends on a payment-provider integration.

Impinge Solutions builds and maintains Shopify, Adobe Commerce and WooCommerce stores and offers an AI for eCommerce readiness check. Ask for the 10-point audit on your store.

Sources

Are you ready to partner with us?

Our good-natured team is waiting to hear your business requirements.

Book a free project call

Stay informed with our newsletter