Both, and the product level is the point. Every tracked product is scored on its own and broken out by topic, so you can see that one line gets recommended regularly while another is never reached for. Brand level visibility hides that. Two products in the same catalogue routinely sit a long way apart.

Branded questions name you. Unbranded questions describe a need and leave the engine to choose. Most brands score well on the first and close to nothing on the second, and that gap is the one worth money. It separates shoppers who already know you from the ones still deciding what to buy.

Being retrieved and being cited are two different things. AI engines routinely fetch a product page, read it, and then build the answer from another source, usually because the attribute a shopper asked about sits in marketing copy and never in a structured field. Monitoring shows you where that is happening. Your analytics cannot, because no click ever lands.

Engines cite the sources they trust for a category, and in commerce that is often a retailer listing or an editorial roundup ahead of the brand's own site. Autonomy classifies every citation as brand, retail, editorial or community, so you can see which domains feed the answers you want to win and how much each one influences. That classification also shows when the mix shifts.

Those tools measure what happens after a click. AI recommendations happen before one, and often in place of one. A shopper is handed a product name and a reason to buy it, and only some of them click through. Nothing in your existing stack records that an engine considered your product and passed over it.