Skip to main content
Visibility tells you whether your brand is in the answer. Product tracking tells you which products are, and which shop the answer sends the buyer to. Ask an assistant to recommend running shoes and the answer names specific models, from specific makers, sold by specific shops. This page is the record of those answers. Open Products from the sidebar, under Shopping.
Product tracking is in closed beta, enabled per organisation and off by default. If you do not see Products in the sidebar, it is not switched on for you yet, and the pages under it are closed to direct links too.

​
The question this page answers

When an AI answer names a shop, is it yours? Every shopping answer carries two readings: the shops it linked to, and the brands whose products it named. The Overview ranks both from the same table, with a Shops / Brands switch, and it opens on Shops.
  • If you make the product, the Brands view is your scoreboard, and you want the shops carrying you to be the ones AI names.
  • If you sell other makers’ products, brand share is not your scoreboard. The Shops view is. The question is whether the answer’s link points at you or at a rival stocking the same product.

​
How it’s determined

This page reads AI shopping answers, which today means ChatGPT only, so the model selector is locked: product carousels are captured from ChatGPT alone. Every captured answer is parsed for the product cards it contains. Each product found becomes one sighting, recording the product, the answer it appeared in, its position in the list, the price and rating shown, the market, and the shopping query behind it. Sightings are then enriched by a job that has the global view, strongest signal first:
  1. The product is resolved to one identity. The same shoe named slightly differently across two answers becomes one product, so its counts add up instead of splitting.
  2. The maker is attributed, cheap signals first. A first-party offer, a sole storefront, the global maker dictionary. Then the retailer pages ChatGPT itself linked are fetched and their schema.org markup read: a merchant’s own statement of the brand outranks every heuristic. What is still unattributed goes to an LLM, and products with no usable link go to a Google Shopping lookup.
  3. Retailers are told apart from makers. A storefront is judged a reseller by the share of its stock carrying its own name, not by how many brands it lists. Makers running their own shop measure 71-100%, resellers 0-8%, and the line sits at 50% with at least three evidenced products behind the judgement. An owner that resolves to a reseller name is re-derived, and a brand link that came from the same mistake is dropped with it. This is what stops a large retailer reading as the maker of everything it stocks.
  4. Your catalog is matched. Uploaded products are scored against everything observed (GTIN exact, then MPN, then attributes). A score above 0.95 links automatically; 0.8 to 0.95 goes to a review queue. This is also what makes a catalog gap visible: a product AI keeps recommending that you have no equivalent for.
  5. Relations are re-judged. Sightings that resolved to nothing are re-read against the catalog matches and your competitor list, so a maker resolved today upgrades yesterday’s sightings.
  6. The answer is joined to your topics through the prompt that produced it, so every number can be read per topic.
A product’s visibility is its share of the captured answers in scope that named any product at all. Answers, not sightings: a product named twice in one answer counts once. Because every product in scope divides by the same number, two products are directly comparable.It renders as a dash, not 0%, when there is nothing to divide by: no product answers were captured in the scope you selected, or the row is a catalog entry that has never been seen in an answer. A dash is unmeasured, which is not a score of zero. Widen the date range or the topic filter before reading it as an absence.

​
Is this product mine

One product sits in exactly one ownership bucket, closest to you first: you make it, it is in your catalog, an answer named your shop among the offers, a competitor makes it, or nothing matched. The precedence is the point: making it beats carrying it, and carrying it beats being named as a seller. The first three are “yours”, and that is what the page exposes:
A dash is not a verdict of “not yours”. An uploaded catalog is a sample, so absence from it proves nothing. Read a dash as a product you might want to add.
The Retailer competition column is the other half of this and is deliberately weaker: it lists the shops AI linked to in these answers, rival products included, and those are links, not confirmed sellers.

​
What the page shows

1

Shops and Brands visibility

One ranking with a switch, over the same answers. Shops is the default: every shop the answers sent readers to, by the share of answers it appeared in, your own row marked. Brands is the same read for the brands whose products were named, whether or not you track them as competitors.
2

Top products visibility

Beside it, the products named most often in the captured answers, yours and your competitors’, ranked by share.
3

Tracked topics

Per topic: Your products, Visibility share with its change against the previous period, Prompts the topic tracks, With products for how often those prompts return products at all, and Top products. Open a row to see every product that topic’s answers named.
4

Products table

Every product observed in your category, with Product, Brand, Your catalog, Visibility and the prompts behind it, Position, Rating, Retailer competition, Category and Price. Filter it with Your products, Categories and Retailers, alongside the standard date range, markets, topics and the locked model selector.
5

Product detail

One product’s full record: its metric cards, the shops that carry it with their prices, how it competes across shops, the shopping queries that surface it most, and every individual sighting.
6

Response inspector

From any sighting, open the captured answer itself, with the extracted product blocks beside the text. This is the evidence behind every number on the page.
7

Add products

Upload your catalog as CSV, from a template on the page. Re-uploading replaces the catalog and re-runs matching.
Everything except catalog gaps works without a catalog. The catalog is what turns “these products get recommended” into “these products get recommended and we have nothing like them”.

​
How to read it

Decide first whether you are reading this as a maker or as a shop. The Shops / Brands switch and the Your catalog column tell you which result you are looking at. Start with how often your prompts return products at all. The With products column on tracked topics is that number. If it is low, the rest of the page is a small sample and the work is upstream, in your prompts and topics. Read visibility per topic, not the headline. A topic where AI shops constantly and never names you is a much sharper problem than a middling average across every topic. Treat position as secondary to presence. Being named at all is the threshold. Where you sit in the list matters after that. Check the Brands view against the competitors you actually track. It counts every brand AI named, so it will surface rivals you have not added to your competitor list. Those are worth adding, and they are worth knowing about either way. Use the response inspector before you act. Every count traces back to a captured answer, and reading two or three usually explains a number faster than the number does.

​
How to improve it

  • Upload your catalog if you have not. Without it, gaps are invisible and nothing can match above the automatic threshold.
  • Fix the topics where rivals are stocked and you are not. The tracked topics table shows where your products are missing from the answers.
  • Work on the shops that carry your rivals for those topics. AI shopping answers lean heavily on merchant listings, so distribution is a direct visibility lever here.
  • If you are the shop, chase the products you already stock. Filter to Your products and look for rows where the Retailer competition column is full of somebody else. That is winnable without changing a thing about the product.
  • For high-value prompts where product visibility stays weak, paid placement is the faster route. See Ads.
Filter the products table to Your products and sort by Visibility, lowest first. Those are products AI knows about but rarely recommends, usually a cheaper fix than getting a new product recognised at all.