Skip to main content
Shopping prompts are tracked prompts written to trigger product cards. AI models only show a shopping carousel for some queries, mostly ones phrased at the moment of buying, so tracking “best running shoes for beginners” tells you about recommendations in text, while “where to buy Acme Trailrunner 2 online” is what actually surfaces a shopping block. If you want shopping data, you need prompts of the second kind.

​
How shopping results differ from normal responses

A shopping result is an enrichment on top of a response, not a different response type. The model still writes an answer; the shopping block is a set of structured product cards attached to it, each with a price, merchant, rating, review count, and a position in the carousel. Promptwatch extracts those cards separately from the answer text, which is why shopping gets its own Shopping section instead of living inside brand mentions:
  • Mentions measure the text: whether your brand is discussed in the answer.
  • Shopping measures the carousel: whether your product is one of the cards, and in which slot.
A response can mention your brand warmly and still show only competitor products in the carousel, and the other way around. Today shopping extraction covers ChatGPT (search) and Amazon Alexa responses, and even a well-phrased prompt isn’t guaranteed to trigger shopping on every run; the feature activates per response.

​
Generate suggested prompts from your products

The fastest path is to let Promptwatch write the prompts from your catalog:
  1. Go to Shopping → My products and select the products you want to target (unmatched ones are the natural pick). See Upload and manage your products if the list is empty.
  2. Click Generate suggested shopping prompts.
  3. Pick the monitor the prompts should run in. Generated prompts use that monitor’s language and country, so a “DE - German” monitor gets German shopper queries.
  4. Click Generate prompts, then review, edit, and save the suggestions in the prompt dialog before anything is created.
Generation handles up to 50 products at a time, roughly one prompt per product; select more and it takes the first 50, so run it again for the rest.

​
What makes a prompt trigger shopping

The generator is biased toward the patterns that surface product cards, and the same rules apply if you write prompts by hand:
  • Purchase-completion phrasing works: “where to buy X online”, “cheapest price for X right now”, “X on sale with free shipping”. The shopper is completing a purchase, not evaluating one.
  • Short and direct works: one product, one buying question, roughly 5 to 15 words.
  • Evaluation phrasing rarely works: “is X worth it”, “which store has X in stock”, and abstract “X vs Y” comparisons tend to return text answers without a carousel.
  • Name the product, skip the variant: color or size in the query doesn’t help trigger a carousel.

​
After your prompts are created

The prompts run on your monitor’s normal schedule alongside everything else. Extracted cards land in the Shopping overview as soon as a response includes them. A tracked product matches only when its catalog ID exactly equals the extracted product ID; that link runs at upload and again in a daily background pass. To inspect the raw responses, filter the responses view by model (ChatGPT) and the Shopping enrichment, the workflow described in Shopping overview. Expect some prompts to trigger shopping inconsistently: the same prompt can produce a carousel one day and a text answer the next, which is itself a signal worth tracking.