How to Track Product Visibility by SKU in LLM Answers

Why track by SKU rather than by brand?

A brand can appear in most answers while the wrong products are recommended. The flagship may be missing, or a discontinued model quoted, or a cheaper variant shown at the wrong price. Brand-level tracking cannot show:

• which products assistants recommend for which needs;
• whether the product you want to sell for a use case is the one named;
• whether prices, variants and availability are right;
• whether links go to your store, a marketplace or a retailer.

How do you build the matching dictionary?

Answers usually use short or informal names, so the short-name column matters most. Add names from reviews and marketplace listings, where customers' own wording appears. Mark discontinued products and their successors, so mentions of old models are flagged rather than counted as wins.

Which questions should you run?

• Use-case questions: "Best waterproof trail running shoes for wide feet"
• Category questions with constraints: "Trail shoes under $150 for beginners"
• Comparison questions: "[Your product] vs [competitor product]"
• Price and availability questions: "Where to buy [product] and how much does it cost?"
• Replacement questions: "What replaced [discontinued product]?"

Keep brand-name questions separate. They show accuracy, not whether you are chosen.

What should you record for each answer?

• products named, matched to SKU or family, with a match confidence;
• position of each product in the answer or carousel;
• the variant, if one is stated;
• the price shown, and whether it matches your current price;
• the merchant and link destination: your store, a marketplace or a retailer;
• availability stated;
• the image shown in product cards, if any;
• the cited sources.

Review low-confidence matches by hand. They often reveal new nicknames to add to the dictionary.

How do shopping carousels change tracking?

ChatGPT, Google's AI Mode and other assistants increasingly show product cards or carousels with prices and merchants, separate from the answer text. These need their own capture:

• Record carousel products separately from products named in the text. They are often different.
• Track Google Shopping as a leading indicator. Search Engine Land found 83% of products in ChatGPT's shopping carousels came from Google Shopping data. A Semrush test found ChatGPT's top product was in Google Shopping's top three 75% of the time.
• Check prices and merchants in cards, which come from feeds and may differ from your site.

How do feeds and identifiers help?

Consistent identifiers make both assistants and your tracking more accurate:

• use the same product name, model number and GTIN on your site, in Google Merchant Center and in any feed you share with ChatGPT's merchant programme;
• keep availability and price current in feeds, since carousels use them;
• mark discontinued products and point to successors on your site.

OpenAI's product feed specification requires a stable item ID and availability for each item. GTINs are optional there, but they help match products across sources. Structuring product data is covered in how to use Product JSON-LD so AI can understand products, prices and offers.

Which metrics should you report?

• Family mention rate: share of runs naming each product family, by question group.
• Top-product share: share of runs where your product is the first or recommended product.
• Price accuracy: share of your product mentions with a correct current price.
• Wrong-variant rate: mentions of variants that do not fit the question, such as the wrong size range.
• Discontinued mentions: mentions of retired products.
• Own-store link share: share of your product links that go to your store rather than elsewhere.

Tracking how assistants describe products is covered in how ecommerce brands can track how LLMs describe their products.

What might an illustrative result show?

Consider an illustrative outdoor footwear brand with 120 SKUs in 14 families. It runs 60 use-case and comparison questions monthly in four assistants. Its brand appears in 45% of answers. SKU tracking shows that 70% of those mentions are one older family, and that a discontinued model accounts for 12% of mentions. Prices are correct in only 58% of carousel cards, and most links go to a marketplace. The brand adds "replaced by" pages for the discontinued model and fixes price syncing in its Merchant Center feed. It also improves product titles to use the short names customers use. Three months later, its newest family appears in answers about waterproof shoes. Carousel price accuracy has risen above 90%.

What do people ask about SKU-level tracking?

Do answers ever use SKU codes?

Rarely. Match on names, nicknames and model numbers instead, and use GTINs where carousels expose them.

How many products should we track?

Start with product families that drive most revenue, then add others. Tracking every size and colour separately adds little.

Can AI visibility tools track SKUs?

Some offer product-level tracking. Ask how they match variants, whether they capture carousels, and whether they record prices.

How often should SKU tracking run?

Monthly for most catalogues, and weekly around launches, promotions and price changes.