For years, AI visibility meant one thing: getting mentioned when an assistant answered a recommendation question. That is changing fast. AI assistants are no longer just naming products. They are selecting, comparing, and buying them on a shopper’s behalf.
OpenAI now lets people complete a purchase inside ChatGPT, Google and Gemini are wiring checkout into AI shopping, and Amazon’s Rufus answers product questions before a shopper ever reaches a listing. What wins a sale shifts with that: the product an agent can read, trust, and transact with gets picked, whatever its landing page looks like.
This guide covers what agentic commerce is, the two protocols quietly deciding how it works, and a practical way to get your products recommended, and bought, inside AI assistants.
What is agentic commerce?
Agentic commerce is when an AI agent completes part or all of a purchase for a person: discovering options, comparing them, and in some cases checking out, instead of only returning links for the person to click. The shopper states an intent (“find me waterproof trail runners under $150 that ship by Friday”), and the agent assembles candidates, weighs them, and either recommends or buys.
It is the commercial layer of AI search. Where generative engine optimization (GEO) asks whether an assistant mentions your brand, agentic commerce asks something sharper: when the assistant shops, does it pick your product, and can it complete the sale?
The protocol layer: ACP vs UCP
An agent cannot buy from your store unless there is a shared language for carts, checkout, and payment. In 2026 two competing standards are forming, and your store will likely need to work with both.
| Aspect | Agentic Commerce Protocol (ACP) | Universal Commerce Protocol (UCP) |
|---|---|---|
| Backed by | OpenAI and Stripe; open-sourced | Google and Shopify, with a reported coalition of retailers and payment networks (Etsy, Wayfair, Target, Walmart, Visa, Mastercard, Stripe) |
| Powers | Buying inside ChatGPT (Instant Checkout) | Agent-led checkout across Gemini and Google shopping surfaces and participating retailers |
| Shape | A direct protocol between a merchant and the AI app | A cross-retailer standard for agent-driven purchases |
The details will keep shifting, and the two camps already overlap (Stripe appears on both sides). Rather than bet on a winner, make your catalog readable and buyable through standard, well-structured product data, so whichever protocol an agent speaks, your products are eligible.
What this changes for your store
The shift is early but moving quickly. ChatGPT alone is reported to serve hundreds of millions of weekly users, and widely cited industry estimates put agentic commerce in the hundreds of billions of dollars of retail within a few years. Treat the specific figures as estimates. The direction is not in doubt.
What matters for you is narrower. A meaningful share of buyers will start shopping inside an assistant, and the assistant, not the shopper, does the first round of filtering. If your product is not legible to that agent, you are cut before a human ever sees you. This is part of why measuring AI search like old organic traffic falls short: the funnel now starts one step earlier, inside the model.
How an AI shopping agent actually picks a product
An agent shopping for a product leans on structured, machine-readable signals more than prose. A long article about your category helps less than a clean, accurate product feed. In practice, agents assemble candidates from three inputs:
- Your product feed. The structured catalog (title, price, availability, attributes, shipping) an agent reads to know what you sell and whether it fits the request. Stale or incomplete feeds get filtered out first.
- Structured data on your pages. Product, Offer, and AggregateRating schema in JSON-LD let an agent confirm the details on the page itself. JSON-LD is the format these systems parse most reliably.
- Earned trust off your site. Reviews, ratings, and third-party mentions tell the agent your product is real and well regarded. As with the rest of AI search, earned evidence often decides who gets recommended.
Miss the feed or the schema and you are filtered out before selection. Get those right but skip the trust signals, and you stay in the running yet rarely become the pick.
The AI shopping visibility playbook
None of this requires a trick. It comes down to making your catalog easy for an agent to read, trust, and buy from. Start here:
- Ship a clean, real-time product feed. Accurate title, price, availability, and key attributes, refreshed often. For agent shopping this beats another blog post.
- Add Product, Offer, and AggregateRating JSON-LD to every product page, so an agent can confirm the details on-page.
- Keep price and stock identical across feed, page, and marketplace. Agents drop products on mismatch.
- Check that AI shopping agents can actually reach your pages in robots.txt, your WAF, and any CDN bot rules. If crawlers cannot reach you, none of the rest matters.
- Make pages usable by an agent, not just readable: clear pricing, variants, and availability that do not hide behind heavy JavaScript or multi-step flows. Browser agents need websites they can actually use.
- Earn reviews and third-party coverage. That is the trust layer agents lean on.
- Re-check after each change, since AI uptake usually takes a few weeks to show.
What this does not mean
Agentic commerce is not an invitation to game the agent. The same caution that applies to AI search applies here.
- Stuffing fake attributes or inflated claims into your feed backfires; agents and marketplaces reconcile against reality.
- Do not fabricate reviews. Rating systems penalize it, and agents weigh review quality.
- Do not block legitimate shopping agents by accident while trying to stop scrapers.
- One protocol or one schema tag is not a shortcut. Clean data, accurate pricing, and real trust are the inputs that last.
How to measure agentic commerce
You cannot improve what you cannot see. The question worth tracking moves from whether you got mentioned to whether an AI-driven visit turned into a sale. That means joining three things: AI-referred sessions, the conversions they produce, and the revenue behind them. Tying AI traffic to conversions turns agentic commerce from a trend you read about into a number you can act on.
One honest caveat: some assistants strip referrer data, so a share of agent-driven visits goes uncounted. Pair referral detection with analytics attribution, and read the trend rather than any single day.
Where SurfacedBy fits
SurfacedBy tracks how AI assistants mention, cite, compare, and recommend your brand, and, through GA4 and first-party attribution, how much traffic and revenue those AI surfaces actually drive. As shopping moves inside the assistant, the same loop applies to products: see where an agent picks a competitor over you, find the feed, page, or trust gap behind it, and watch whether fixing it moves the number.
No tool can guarantee an agent buys your product. The value is measurement, diagnosis, and proof that a change moved the result.
Where to start
Agentic commerce moves the first round of filtering from the shopper to an agent, but it does not need a brand-new playbook. The same work that earns AI citations, a clean feed, accurate structured data, reachable pages, and real reviews, is what lets an agent find, trust, and buy your product. Our take: the stores that fix the feed and the schema first will see movement well before the ones still writing another category explainer.
