How to Check If Product Listings Appear in AI Shopping Results
The product-visibility audit we run for ecommerce clients: buyer prompts, ChatGPT and Google AI Mode checks, feed and crawler fixes, and the tracking we set up once the baseline is done.
When an ecommerce client asks whether their products show up in AI shopping results, the honest first answer is "we'll find out, and it depends on where you look." ChatGPT and Google run different systems with different inputs. A product can appear in one and be absent from the other for reasons that have nothing to do with the product.
So we treat the question as an audit with a fixed order. This is the version we run today.
We start by agreeing on surfaces and markets
Before anyone types a prompt, we agree with the client which surfaces are in scope.
ChatGPT shopping is the one most clients ask about. OpenAI added visual browsing and side-by-side product comparison in March 2026 and extended the Agentic Commerce Protocol (ACP) so merchants can share product feeds for discovery (OpenAI). OpenAI's merchant page says shopping in ChatGPT is live in the U.S. That one line changes the scope for our European clients: if their US business is small, ChatGPT shopping may be a watch item rather than a priority.
Google AI Mode and Gemini can show a Buy button on eligible listings through Google's Universal Commerce Protocol (UCP), with checkout on Google's side and the merchant still the seller of record. It runs from Merchant Center, and the help article covers products eligible in the United States, Canada, and Australia, for select merchants in an early access program. For a client selling in Canada, that's the more relevant surface.
Then we build the prompt set with the merchandising team
We don't write buyer prompts alone. The client's merchandisers and customer service people know how buyers describe what they want, and the prompts should sound like that.
A usable prompt has constraints. "Best baby stroller" invites a generic answer. "Lightweight travel stroller that folds one-handed and fits in an overhead bin" makes the assistant commit to products. We mix category prompts like that with a handful of purchase prompts that name the client's own products ("where to buy [product] online"), because those tell us whether a known product surfaces at all.
The manual baseline
We run the set by hand once, deliberately. It's slow, but it teaches us how the client's category behaves before any tool gets involved.
Each prompt runs logged out and logged in, and we label which is which. We note the location we tested from, which matters because both surfaces are limited to specific countries. For every run we record whether product cards appeared, which products, the client's position, the merchant named, and the price displayed. Then we repeat the same set on another day.
The repeat is where clients usually see the problem. Answers vary from run to run, and they shift with persona and location. What a client saw on their own phone last Tuesday was one draw. That's why the manual baseline is a starting point and never the deliverable.
The technical checks
We check the same causes on every audit, in the same order.
Crawler access comes first. OpenAI's publisher FAQ says to allow OAI-SearchBot to appear in ChatGPT search, and its bot documentation says robots.txt changes take about 24 hours. Google's Storebot-Google crawls for Google Shopping surfaces (Google). We check robots.txt and the WAF, because an old bot rule written to stop scrapers can catch these crawlers too.
Next, the feed. OpenAI says a feed isn't strictly required if ChatGPT already crawls the site, but a feed gives more control. Shopify stores are covered through Shopify Catalog. For everyone else we check the export against the ACP product feed spec: a stable id per product, at least one variant, and a stable id and title per variant. Price, images, GTINs, and URLs are optional, and we fill them anyway because they improve display.
On the Google side we open Merchant Center with the client. UCP checkout needs approved free listings, a return policy, a support contact, and the native_commerce(checkout_eligibility) attribute set to true. Some categories are ineligible no matter what (subscriptions, pre-orders, used or refurbished goods, age-restricted products), and we say so early so nobody chases a fix that can't exist.
Last, we compare price and availability between the feed and the live product page. When they disagree, we fix the source system rather than patching the page.
Then we set up tracking
Once the baseline is recorded and the fixes are queued, the repeat checking moves into Promptwatch, the platform we run client programs on. For ChatGPT, its ChatGPT Shopping tracking takes the client's product list as tracked products and follows their positions over time, along with the top products and top merchant domains showing for the prompt set. Ads Radar sits on the same prompts and tells us when a competitor is paying for placement, which we report on a separate line from organic results. Both are on the Business plan ($579/mo), so that's where we put ecommerce clients.
We set prompts up with personas and with country tracking, and add state or city tracking, also on Business, when a client's sales are regional. Agent Analytics shows us whether AI crawlers are actually fetching product pages after a robots.txt fix. Visitor analytics, installed by script or GTM, records AI-referred visits and the conversions behind them. That's the only number we'll put next to the word "revenue" in a client report.
We're careful about scope with clients here. Promptwatch's shopping tracking is ChatGPT Shopping. It tracks the client's prompts in Google AI Overviews and AI Mode, but it doesn't track AI Mode shopping cards or Gemini product listings, so for Google we report from Merchant Center and the client's own analytics alongside it.
What the client gets
At the end of the audit a client has the prompt set, the manual baseline with conditions labelled, a fix list split by owner (dev, feed, Merchant Center, content), and live tracking that replaces the spreadsheet. If you've read our post on ChatGPT Shopping and Ads Radar, the reporting format will look familiar: organic mentions, shopping results, and sponsored units each get their own row.
We don't promise a shopping position. Nobody can. What we can promise is a clear picture of where your products stand and which fix to make first. If that's useful for your catalog, email hello@1001seomedia.com.