The agent reads price, stock and delivery. What is not in the data does not exist.
In the United States, assistants already complete purchases inside the conversation. Everywhere else the storefront arrived first: product panels in Google AI Mode and carousels in ChatGPT. We prepare the data layer that decides who shows up, and measure that presence week by week.
The storefront came first, the counter follows
The state in August 2026, with dates. Google agentic checkout shipped in November 2025, Copilot Checkout in January 2026, and buying inside Perplexity in November 2025, all on United States inventory. OpenAI pulled its checkout back in March 2026 and kept the discovery surface. Two protocols are competing underneath: ACP, from OpenAI and Stripe, and UCP, from Google and Shopify. Neither has won, and building only for one is a bet, not a plan.
The layer we prepare
The feed, complete and fresh
Google Merchant Center is the spine. Price, availability, shipping and returns, GTIN or MPN on every item, and a refresh cadence that matches how fast your stock moves. A feed that is 94% complete is a feed with a 6% hole where the agent stops reading.
Product pages a machine can read
Product and Offer schema with price, currency, availability, shipping details and return policy, matching what the feed says. Disagreement between the two is worse than silence: the engines resolve it by trusting neither.
Assistant feeds as they open
The OpenAI product feed specification, and the equivalents that follow. Registration, format and cadence handled as each channel opens to your market, rather than discovered six months after competitors got in.
Measured across the catalogue
The Visibility Tracker checks the fill rate across a sample spread over the whole catalogue, then product by product. In a catalogue of 10,000 items what counts is the percentage, not the one page that happens to be complete.
Price Radar
Our observatory of the ChatGPT shopping carousel logs, item by item, what a buyer actually sees: title, brand, merchant, price, rating, position, and the sponsored label when there is one. If your product shows the wrong price, an unexpected merchant, or sits behind a sponsored item, you hear it from us, not from a customer.
Behind it sits our own instrumentation: we run an automated observatory of the ChatGPT shopping carousel, logging which products appear and which sources feed each panel, alongside the weekly per-product measurement in the Tracker.
Common questions
+Does AI already complete purchases on its own?
In the United States, yes, on several surfaces: Google agentic checkout since November 2025, Copilot Checkout since January 2026, and buying inside Perplexity since November 2025. OpenAI pulled its own checkout back in March 2026 in favour of discovery. Outside the United States what exists is the storefront rather than the counter: product panels with prices in Google AI Mode and shopping carousels in ChatGPT. The race for the storefront is running now, and the counter arrives later for whoever is already standing there.
+What does a shopping agent actually read to pick a product?
Structured data, not the copy on your site: the Google Merchant Center feed (price, availability, shipping, GTIN), Product and Offer schema on the page, and the assistants' own feeds, where the OpenAI specification accepts a refresh every fifteen minutes with inventory quantity and delivery estimate. Around 83% of the products in ChatGPT carousels come from the organic top 40 of Google Shopping (Search Engine Land, March 2026), so the Google feed moves both surfaces at once.
+Does this apply to services, or only to shops?
Agentic checkout is mostly a product story. The underlying mechanic is universal: the agent answers with whoever is in the data. For services the equivalent layer is the entity, the service schema and presence in the sources that get cited, which is the ordinary method. For shops, the feed and a machine-readable product page are added on top.
See where your products stand, before touching the feed.
The free measurement puts real category questions to the engines and shows what comes back. If there are products, we read the commerce layer with it.
AI ads: advertising inside AI answers → · How the weekly measurement works →