AI Product Found Rate

Found rate is the percentage of relevant shopping questions on which a retailer's product appears in the assistant's answer in any form: a brand mention, a product card, a citation or a recommendation. It is the base measure of AI visibility, roughly what impressions are in search.

It sits alongside two related measures:

• Found rate: does the product appear at all?
• Product card rate: does it appear with image, price and link, rather than as text?
• Position: is it the primary recommendation or an alternative further down?

All three are tracked per assistant, per question set, over time. The combinations are diagnostic. High found rate with low card rate usually means a schema or feed gap: the assistant knows the product but cannot render it as something shoppable. Found and carded but low in the answer usually points to thin reviews or weak category content.

A workable measurement comes down to four decisions:

1. The question set. 30 to 100 questions per category, written the way your shoppers ask them. Mix branded ("is [brand] good for X?"), comparative ("[brand] vs [competitor]") and unbranded ("best X for Y"). Keep the set fixed so the trend means something, and rotate a few each quarter as buying scenarios change.

2. Which assistants. ChatGPT, Google AI Mode, Perplexity and Google AI Overviews cover most retailers. Add Amazon Rufus where Amazon matters. Results differ a lot between assistants, so track each one separately.

3. Product level. Track found rate per product for your hero SKUs and per line for the long tail. A brand-level number hides the usual pattern: the top sellers appear, most of the catalog does not.

4. Cadence. Assistants refresh what they retrieve frequently. Weekly is a sensible floor if you want to see the effect of your own changes.

This is what the Paz platform runs for you. It asks the question set on a schedule across Google AI Mode, ChatGPT, Perplexity and AI Overviews and reports found rate, product cards, mentions, citations and position over time. A Paz expert reads the results and turns each gap into a fix. See the AI visibility service.

The levers are the fundamentals of product information:

1. Attribute completeness. Material, fit, use case, compatibility, certifications and the other details your category filters on. A missing attribute removes the product from the comparison.
2. Complete product schema. Valid Product JSON-LD with price, availability and ratings is what lets an assistant render a product card at all.
3. Feeds on each surface. Google Merchant Center for AI Mode and AI Overviews, the OpenAI product feed for ChatGPT. One feed caps your found rate at the share of questions that assistant handles.
4. Category content. For unbranded questions, which are most shopping intent, appearing depends on category authority as much as on the product record. Guides, comparisons and inclusion in third-party roundups all help.
5. Reviews. Assistants lean on review volume and sentiment when they recommend. See our branded vs unbranded visibility analysis for how the gap shows up.
6. Unblocked AI crawlers. If GPTBot, Google-Extended, ClaudeBot or PerplexityBot are blocked in robots.txt, found rate on that assistant is close to zero whatever else you do.

A Paz expert works through this list for your catalog, starting with the products and questions that matter most to your business, and publishes fixes within agreed guidelines and brings exceptions to you.