Articles — Shopping inside chat
Shopping inside chat
ChatGPT can look up your stock only after you connect it
A product page is for people. An assistant wired into your shop needs labeled stock and price, and even that will not get you named on a public shortlist.
Someone asks whether your watch is in stock at today’s price. Your marketing page may say yes, but the assistant in the chat still cannot answer from live catalog facts unless it is wired into your shop.
The page was built for people and crawlers. An assistant that is wired into your systems may need something else: labeled product facts it can request during the chat, rather than a URL it has to read like a brochure.
Getting recommended in a public answer is still a different job. See How AI chatbots build a shortlist.
A live catalog the assistant can query
Illustrative example, not a logged test. The Northloop Trail Watch is a fictional product at €300.
Public webpage
Northloop Trail Watch
Story, photos, and layout, for humans and crawlers. This has to be opened and read.
Live catalog hook-up
Labeled product, price, and stock the assistant can request during this chat.
That hook-up is MCP (Model Context Protocol), a tool call during this chat. How to publish the tool is in The assistant checks your stock by calling a tool you publish.
While the chat is running, a connected assistant can ask for a product code, a price, or availability without reading your marketing page like a brochure.
That sits next to the webpage, the structured block, and the catalog file in Your product lives in three places, and it does not replace them:
- Your webpage — story, photos, and layout
- JSON-LD on the page — labeled facts for machines that open that URL
- A product feed — rows of price and stock for shopping systems
The connection answers a narrower question: can this connected assistant read your product facts on demand?
What it will not do
It will not make ChatGPT recommend your watch to everyone. Most people asking what they should buy never use this connection, and a public answer still leans on reviews, shopping indexes, feeds, and memory. See How AI chatbots build a shortlist.
When it is worth doing
MCP-style access is useful when an assistant is actually connected to your shop, such as a support bot, an internal catalog, or an experiment with buying in chat. You want the reply to use your price and stock, not a guessed snippet from HTML.
It is less useful if your only goal is being named when a stranger asks a generic buying question in ChatGPT.
An assistant that is connected to your catalog and can also complete a purchase skips the moment when a person would normally stop and check. ChatGPT recommended a deodorant that didn’t exist is that escalation, with a fake brand that sounded vetted in chat.
When someone sells you “AI commerce,” ask which of these they mean: a public page, a feed, a live catalog hook-up, or checkout. Then ask whether they mean being named, being described correctly, or finishing a purchase. If they mix those, ask them to name the connection and the job before you pay.
How this sits next to feeds and checkout: Checkout in chat comes after the shortlist.
Read next if: The assistant checks your stock by calling a tool you publish · Checkout in chat comes after the shortlist