Track products & collections
Product tracking measures how often AI engines recommend a specific product or surface your brand for a focused collection, not just your brand overall. Paste a product detail or collection URL, review the extracted page, and track the buyer questions attached to it.
Before you start
Product tracking sits inside a project, so create the brand first. You need a live product or category URL that an AI engine could read, for example a product detail page or a collection page.Brand-level vs product-level tracking
Track at the level your customer actually chooses. If buyers pick the brand and the specific product barely matters, brand-level is enough. If different products have different buyers and different questions, track each product. The table below is the quick decision.
| Use brand-level when | Use product-level when |
|---|---|
| The brand is the unit of choice (people pick "Acme," not a SKU). | Individual products have distinct buyers and search queries. |
| You sell one product or a tight, similar range. | You sell a catalog where products compete on different attributes. |
| Customer prompts name the category, not a model. | Customer prompts name a model, size, or use case. |
| You want one headline visibility number. | You need to compare products against each other over time. |
The two are not exclusive. Many brands keep a brand-level visibility view and add product tracking for the few SKUs that earn their own buyer journeys.
A worked example
Acme sells running shoes. The flagship is the Acme Trailblazer. Buyers do not ask "is Acme good?" They ask "best trail running shoes for wide feet" and "Acme Trailblazer vs Beta Summit". Those are product questions, so Acme tracks the Trailblazer as a product. The walkthrough below sets that up.
Set up product tracking
Paste a product or collection URL
Inside the project, choose Track a page and paste a PDP such asacme.com/shoes/trailblazer, or a focused collection such asacme.com/collections/trail-shoes. AppearIn AI detects the page type before extracting its details.
Review what AppearIn AI extracted
Check the extracted product name, key attributes, and any variants AppearIn AI found. Correct anything wrong here, because the prompts and mention detection are built from this. For the Trailblazer, confirm the name and that it is a trail shoe, not a road shoe, so prompts target the right queries.
Accept the auto-generated prompts
AppearIn AI generates prompts a buyer would ask about this product, such as "best trail running shoes for wide feet" and "is the Acme Trailblazer good for long distances?" Accept the ones that match how your buyers actually ask, edit the rest, and drop any that do not fit. Accepted prompts join the project library without changing the current locked set. See build a prompt set for choosing well.
Choose the product prompts for the next set
Select the product prompts you want to optimize next month. A project's library may hold prompts for many products, but each monthly set can include at most 200 prompts total. When that version locks, the selected prompts join the weekly run.
Compare products and track the trend
Read each included product's mention rate: the share of its prompts where an AI engine names the product, broken down by AI engine. If the Trailblazer is mentioned in 60% of its prompts on Perplexity but the Acme Glide only 20%, you know where to put your content work. Open any answer to inspect the citations behind it.
What good looks like
Well-set-up product tracking has:
- Products that genuinely have their own buyers, not every SKU in the catalog.
- Extracted product facts you have checked and corrected.
- Prompts that name the model or use case the way buyers do.
- Only the products you are actively optimizing represented in the monthly set.
- Weekly runs on, so per-product mention rate is a trend you can compare over time.
Common mistakes
Tracking every SKU. If a product has no distinct buyer or query, it adds noise, not signal. Track at the brand level instead.
Skipping the extraction review. A wrong product name or category quietly poisons every prompt and mention count that follows.
Reading one run as truth. Data is probabilistic. Compare products across several runs, not a single snapshot, before drawing a conclusion.