Skip to main content
For a question that names something to buy, an assistant can show product cards under its answer. Each card names a product, a price and the stores that sell it. A reader usually sees the first two or three cards before they scroll. The Products page lists the cards the assistants showed for this brand’s prompts.

Which platforms report them

The Platforms page marks the platforms that report their product cards. Today that is ChatGPT only, so the Products page reads ChatGPT and has no platform filter.

What the page shows first

  • One sentence: this brand’s share of the shopping answers, its rank among the brands, its change against the period before, and how often it holds the first card.
  • Your products: this brand’s products the cards showed. With a catalog, the catalog’s products.
  • Where you lose: the prompts whose shopping answers showed no card of this brand, with the product that took the first card there most often. A row opens those answers.
  • Brands in the cards: every brand’s share, change and first cards.
Details holds the rest: the daily chart, every product, the stores and the prompts.

Prompts that show products

Cards appear for questions that name something to buy and ask which one to choose:
  • “What are the best running shoes for beginners?”
  • “Best cordless stick vacuum under $300”
  • “Where can I buy a Dyson V15 Detect in Germany?”
Questions about how to do something rarely show cards. Questions about services rarely show them either. The table “Prompts that showed products” shows how often each prompt’s answers showed cards.

What the page counts

A shopping answer is an answer that showed at least one card. Every share on the page divides by the shopping answers, not by all answers. A card belongs to a brand when its product name, or the query ChatGPT ran to find it, spells the brand as the answers do. A card that names no brand belongs to the monitored brand whose own website sells it. The brand panel also names the brands the project does not monitor whose cards appear most, in grey. They are brands the answers named, placed by the query ChatGPT ran. The rest show under “Other brands”. One product can appear under slightly different names. The page merges names that differ only in size, colour, a word such as “men’s” or “running shoes”, or a closing word such as “trail”. It keeps “Bambino” and “Bambino Plus” apart, because those are different products. A row takes the product name ChatGPT looked up. It opens onto the title a store lists the product under, and every answer that showed it. The table lists the products shown most, and the sellers panel the stores with the most offers. Each says how many there are in all. The chart shows each brand’s share of each day’s shopping answers. Its note says how many days of the period have answers.

Your catalog

A catalog is the list of products this brand sells. Import it under Settings, Project, Catalog:
  • A Shopify store. Type the store’s domain. We read its public product feed.
  • A file. CSV or TSV with a header row, a Google Merchant Center feed (TSV or XML), or Shopify’s product export. The browser reads the file and sends those columns only. “The columns we read” on the card lists the column names each field accepts, in any case. Only the title is required.
An import replaces the catalog. One product in many colours or sizes is one catalog product. A catalog holds 500 products. With a catalog, the page changes in three ways:
  • Your products lists the catalog’s products the cards showed, with your price beside the price the cards showed. A price gap shows only when both prices are in one currency. Each product names the stores with the most offers of it, and how many offers came from this brand’s own website. A list then names this brand’s products on the cards that are not in the catalog, such as an older model. They count for the brand in the figures above. Last, a list names the catalog’s products that no answer showed. Export CSV saves the catalog’s products with both prices and their stores.
  • A card counts for the brand the catalog names, whatever its title says. A brand that the project does not track shows under “Other brands”.
  • A category filter narrows the product lists. It leaves the brand shares whole, because a rival’s card has no category.
A card matches a catalog product by its product name. When a match is wrong, correct it:
  • Join. Pick its catalog product beside a product in the list of products not in the catalog. For any other product, open it under Details, Every product.
  • Split. Under Your products, open the product and click Not this product beside the card.
A correction stays when you import the catalog again, or remove it.

Branded prompts

A prompt whose wording names this brand leaves every figure by default, on this page as on the others. The branded filter shows those prompts.

Where it appears