Which software tracks which merchants AI shopping results recommend? How we do it for ecommerce clients
How we track which merchants AI shopping results recommend for ecommerce clients: what to measure, which surfaces matter, and the platform we run it on.
Ecommerce clients have started asking a new question. Not "do we rank for this keyword" but "when someone asks ChatGPT for the best product in our category, which stores does it put in the carousel, and are we one of them?" It is a fair question and a hard one to answer with older tools, because shopping results are a separate surface from the text answers most trackers measure.
Why shopping results need their own tracking
A text answer might say your brand is a good option. A shopping carousel shows a product tile, tied to a merchant, in a position. The first is a mention. The second is something closer to shelf placement. The metrics differ: you care about which merchants appear, in what order, which of your own products show up and which do not, and who is paying for placement.
A mention tracker can tell you the brand was named in a good share of runs and still say nothing about whether the product tile pointed to you or to a reseller.
What we set up for a shopping program
A shopping prompt set. Buyer-style prompts by category, price band and use case: "best standing desk under a given budget," "gift ideas for a home cook." We keep these separate from informational prompts.
Catalog coverage. Which products appear and which never do. A catalog of 400 products rarely shows up evenly.
Merchant comparison. Which retailers and competitors appear for the same prompts, and how that changes over time.
Source analysis. Which pages the engines cite when they explain a recommendation: review sites, comparison articles, retailer product pages, forum threads.
Access. Whether AI crawlers can reach the product pages.
The platform we use
Promptwatch is where we run this for clients. Its AI Shopping covers ChatGPT shopping carousels and Amazon Alexa product results, and supports uploading the client's own catalog. Shopping Insights is available from Business ($579/mo). The API and MCP catalog also includes shopping endpoints for top merchant domains and top products, which we use to pull merchant comparisons into client reports.
Ads Radar is the companion feature. It captures sponsored ads inside ChatGPT Search, Google AI Overviews and Google AI Mode answers to tracked prompts, including advertiser name and domain, title, description, image or price, target URL and the prompt that triggered it. Views include ads per day, top advertisers and a table by prompt. It does not cover Perplexity and it does not replace OpenAI's own Ads Manager for the client's campaigns. Ads Radar is on Business and on every agency plan.
For diagnosis we lean on the rest of the platform. Citation analytics shows which product pages and review sources the engines cite. Agent Analytics reads IP-verified crawler logs, so we can see whether OAI-SearchBot or ChatGPT-User reach the product pages and whether they hit errors. Visitor analytics attributes AI-referred traffic and conversions, which is how we find out whether shopping visibility is worth the effort for a given client.
Plan notes we share up front: shopping tracking is a Business feature. Agency plans include Ads Radar but not Shopping Insights, so for a client that needs both we plan a brand-level Business project.
Other options we have looked at
LLM Pulse includes ChatGPT Shopping and AI ads tracking on its Scale plan. Goodie AI focuses on commerce, with SKU-level visibility on its Pro plan and coverage of assistants such as Amazon Rufus and Walmart Sparky. Bluefish targets large consumer brands with managed work and no public pricing. Qwairy lists a shopping module, metered by credits. We treat all of these as worth a trial for a client whose main surface is one of those assistants.
Honest limits
Shopping carousels change run to run, so a single check means little. We look at trends over weeks. And tracking tells you who is recommended, not how to be. No tool can guarantee a product tile for a client. What we can do is show which merchants win, read the pages the engines trust, fix access problems and improve product pages.
We also tell clients which parts of the market a tool does not cover. For example, Ads Radar skips Perplexity, so a client whose buyers use Perplexity will not see ads there.
A first week
Write 30 shopping prompts for the client's top three categories. Run them for a week and note which merchants and products dominate. Then check whether the crawler logs show the product pages are being fetched. In our experience the first findings are often basic: product pages with thin copy, blocked bots, or a competitor retailer cited for the review content the client could have written.
If the client sells online and the surface matters, that is when we recommend evaluating Promptwatch on a Business project first.