If you sell online, there’s a new kind of feed worth understanding before your competitors do. It’s the product feed behind agentic commerce: the structured catalog that lets an AI assistant not just mention your product, but understand it well enough to recommend it and, increasingly, help the shopper buy it.
OpenAI’s version of this is the Agentic Commerce Protocol (ACP) product feed. Here’s what it is in plain terms, how it’s different from the feeds you already run, and why it matters now rather than later.
The shift: from "show me links" to "buy this for me"
Classic e-commerce assumes a human does the clicking. They search, they browse, they add to cart, they check out. Every feed and tag you run today is built around that human in the loop.
Agentic commerce removes some of that loop. A shopper tells an assistant what they want; the assistant finds candidates, compares them, and in the most advanced flows can complete the purchase. For that to work, the assistant needs a catalog it can read and act on with confidence, not a web page it has to scrape and guess at.
That’s what an agentic feed like ACP provides: a clean, structured, machine-first description of your products, priced, available, and attributed, that an AI can trust enough to put in front of a buyer.
How ACP differs from a Google Shopping feed
People assume "I already have a Google Shopping feed, so I’m covered." Not quite. They overlap, but they’re built for different consumers:
- Audience. A Google Shopping feed feeds Google’s shopping surfaces. An ACP feed feeds AI assistants and agentic checkout flows. Different destinations, different expectations.
- Strictness. Agentic feeds lean even harder on completeness and accuracy, because an agent is making a decision (or a purchase) from the data, not just displaying it for a human to judge.
- Structure. Formats and required fields differ. A feed that passes Merchant Center isn’t automatically valid or optimal as an ACP feed, even though the underlying product data is the same.
The practical takeaway: the source of truth (your clean product data) is shared, but you need to publish it in multiple structured formats to be present everywhere shoppers and agents now look.
Why this matters now, not "someday"
Two reasons it’s worth getting ahead of:
- Recommendation is winner-take-few. When an assistant answers "what should I buy," it returns a shortlist, not a results page. Being readable is the price of entry to that shortlist. Being unreadable is invisibility, and there’s no second page.
- Almost nobody is optimized for it yet. Most stores haven’t even audited whether their catalog is agent-readable. That’s a rare, real first-mover window. The same way early structured-data adopters won rich results, early agentic-feed adopters get cited and recommended while competitors are still arguing about whether AI shopping is real.
What "getting it right" actually involves
You don’t need to become a protocol expert. The work is the same data hygiene that makes any feed strong, applied to the agentic format:
- Complete, accurate core fields: title, description, price, availability, GTIN, brand, image, category, all present and correct.
- Rich, consistent attributes: color, size, material, gender, age group, filled across the whole catalog so an agent can match a specific request.
- Machine-first titles and descriptions: specific and unambiguous, written so a model understands exactly what the product is.
- Published in the right structured formats: JSON-LD on-page, plus the agentic feed format itself, kept in sync as the catalog changes.
Get the data clean once and you can syndicate it everywhere: Google Shopping, Meta, and the AI/agentic surfaces, from one source.
What the ACP feed actually asks for
The specification is still moving, so treat any field list as a snapshot rather than a contract. But the shape of it is already clear, and it is not exotic: an agent needs enough structured detail to answer a question without opening your website.
In practice that means the fields you already owe Google Shopping, held to a higher standard. A title that identifies the product rather than describing it. A price that matches the page, in a currency, with availability that is true right now. Identifiers, because an agent comparing two listings needs to know whether they are the same product. And the attributes a shopper would filter on: size, colour, material, fit, compatibility.
The difference from a shopping feed is what happens when a field is missing. In Google Shopping a thin title costs you match quality and you still appear. In an agentic answer, a missing attribute is a question the agent cannot answer, so it moves to a listing that can. There is no partial credit and no second page to appear on.
How to tell whether you are ready today
Three checks, in order, and none of them need a tool.
Read one of your own product titles out loud. Does it identify the item to someone who cannot see the picture? “Slim fit chino, sand, 32” does. “Stylish trousers” does not, and neither does a title that is mostly your brand name.
Pick a filter your shoppers use and check whether it exists as a field. If people ask for waterproof, or wide fit, or a specific size, that has to be an attribute rather than a sentence in the description. An agent reads fields; a paragraph is not a field.
Ask what happens when something goes out of stock. If availability updates once a day, an agent will recommend something you cannot ship, which costs more trust than the sale was worth.
If those three make you uncomfortable, the same gaps are almost certainly costing you in Google Shopping already, where they are easier to see. The disapproval reasons Merchant Center gives you are the same data quality problems, reported by a system that bothers to tell you.
And if the question behind all of this is whether assistants can find you at all today, that is a separate and more immediate problem: whether your store is invisible in AI search depends on the catalog being readable long before ACP is involved.
How Verintra makes your catalog agent-ready
This is the core of what Verintra does. It grades each product on how machine-readable it is, uses AI to fix the titles, descriptions, and missing attributes that hold products back, and exports your catalog in the structured formats that matter, including JSON-LD and an OpenAI ACP feed, alongside the classic shopping channels. One clean catalog, published everywhere an agent or a shopper might look.
The scan is free and tells you, today, how agent-ready your products actually are.
Scan your feed free at verintra.com. Find out whether AI assistants can read, and recommend, your catalog. No credit card, about two minutes.
