Merchant Center Help says about six feed fields that conversational attributes "help AI systems and conversational agents better understand specific details" of a product. For product data, that sentence is the only place where Google names shopping agents.
Yet agentic shopping is often written about in forecasts. What you want to know is what of it stands in Google's and OpenAI's documentation and what is a guess.
I'll go through both sets of documentation and then through what the documented sentences mean for an e-shop's product data. You won't find a date for shopping agents arriving in Czechia here: the sources don't give one.
Agentic commerce and how it differs from advertising in ChatGPT
A working definition to start: agentic commerce is shopping in which an AI agent picks products on the user's behalf, or buys them outright. This is not a quote from Help.
Three neighboring topics swirl around agents, and each is about something else.
| What it is | Who is behind it | What the source says about it | Where it is on the blog |
|---|---|---|---|
| ChatGPT Ads | OpenAI | OpenAI's ad platform. Ads are separate from product results. | ChatGPT Ads as OpenAI's ad platform |
| AI Overviews in Search | The sources on agentic shopping don't name them. | AI Overviews and conversational attributes | |
| AI Max for Shopping campaigns | A beta Google Ads feature, not an AI agent. | AI Max for Shopping campaigns |
Advertising often gets mixed up with product selection. OpenAI Help explicitly separates ads from product results. You set up AI Max in Google Ads, while you send conversational attributes in the feed.
Google: conversational agents and Merchant Center data
That sentence about conversational agents sits in Help next to six fields. I cover what each of them carries in the article on conversational attributes in Merchant Center. The exact wording is in Merchant Center Help.
The feed doesn't end with six fields. The product data specification lists dozens of required, recommended and conditional attributes: ID, title, description, price, availability, brand, GTIN, material, gender, age group, size, custom label or Google product category.
An example of how a query meets the feed fields. Someone is looking for a two-person tent under CZK 4,000. The price ceiling points to the price attribute, being in stock to availability. The type of product corresponds to the Google product category. How an agent chooses from such fields, the sources don't say.
Example
The query “two-person tent under CZK 4,000” and the feed fields
- Price ceiling under CZK 4,000price
- In stockavailability
- Product type: tentGoogle product category
The beta ad feature AI Max for Shopping campaigns stands on the same data. It is not an agent. It takes its data from the same feed. So you don't fill in feed fields just for Shopping.
OpenAI: Instant Checkout and the Agentic Commerce Protocol
On September 29, 2025, OpenAI announced purchases directly in ChatGPT. It is called Instant Checkout and it runs on the Agentic Commerce Protocol, which OpenAI developed with Stripe and released as an open standard.
At the start, ChatGPT Plus, Pro and Free users in the US got it, along with US sellers on Etsy. The announcement adds that over a million merchants on Shopify will follow. At the time, it covered single-item purchases.
OpenAI's Help on shopping in ChatGPT today describes what the user sees: for a shopping query, ChatGPT shows products with an image, details and links, and for eligible products and merchants it also offers Instant Checkout. The purchase then happens in ChatGPT.
ChatGPT picks the product results independently. Help says they are not ads.
What it means for an e-shop
Both sets of documentation meet at structured product data. Google describes it in the Merchant Center specification, OpenAI in its feed specification: one row per sellable item or variant and nine required fields, among them title, description, brand, image, availability and price.
Help describes two paths into ChatGPT. Merchants on Shopify have their data integrated through Shopify Catalog and don't do anything else. Anyone who wants to send a direct product feed to OpenAI applies for access on the developer site.
In principle, both specifications ask for the same work: complete product data in the feed. For clothing, do you send only the title and price, with material and size missing from the feed? You can add them with a supplemental feed, no developer needed.
I won't claim what an agent reads on a website. I don't have an official source for that. What Google already assesses on a website today is in the Misrepresentation policy.
Summary
- An agent picks or buys on a person's behalf. In ChatGPT, ads and product results are two separate things.
- Google's only mention of shopping agents leads to Merchant Center. It concerns the six conversational attributes in the feed.
- OpenAI announced Instant Checkout on September 29, 2025, and with it the Agentic Commerce Protocol; US sellers on Etsy were there at launch.
- The feed for OpenAI has nine required fields, while the Merchant Center specification lists dozens. Filling them in is on you.
- Merchants on Shopify don't do anything extra. Their data gets into ChatGPT through Shopify Catalog, everyone else applies for access to the direct feed.
- Google assesses websites even without agents. The Misrepresentation policy applies today.
There are few documented sentences so far. But they describe the same work the Merchant Center product data specification asks of you.
Take your ten best-selling products and compare their rows with the required and recommended attributes from the specification. Fill in the missing fields with a supplemental feed, and in two weeks check on the same ten products what made it into the feed.
