AI shopping assistants and the product data they read

What an AI shopping assistant is, and where its product data comes from

An AI shopping assistant is a consumer-facing AI agent that recommends specific products in its answers or completes purchases on a shopper's behalf. The term covers general assistants with shopping inside them, retailer and platform assistants, and the agentic checkout features attached to either. Operators and press also call them AI agents, personal agents, shopping agents or AI personal assistants.

A product data path is the route by which an assistant learns about a product: what it reads, in what format, how often, and what a merchant must do to be included. There are two mechanisms. Crawl-based ingestion reads what a brand publishes on the open web, chiefly product pages and structured data such as schema.org Product markup on its own domain. Brands may also publish an llms.txt file or link to a third-party machine-readable catalog such as an Agentsite, but no operator documents crawling either for shopping. Feed-based ingestion takes a catalog the merchant connects or pushes: Shopify Catalog, Google Merchant Center, Amazon's catalog, or a merchant feed exchanged over a protocol such as the Universal Commerce Protocol. Most assistants use both, and most document the feed side more precisely than the crawl side.

Documentation level records how explicitly an operator states its product data sources: explicit when it names the sources and the required fields, partial when it names some, vague when it describes ingestion only in general terms, and undocumented when it says nothing. Observed crawling is reported separately from documented sources.

Which product attributes each AI shopping assistant reads

The first table gives each assistant's US status, documented product data sources, documentation level and whether it buys for the shopper. The second shows, for fourteen common product attributes, whether each operator documents the field as required or supported. Assistant names link to their data sheets, where every statement is sourced.

The full dataset behind this page, including every data sheet, the attribute matrix, the surfaces, the rails, the changelog and the corrections register, is available as JSON.

Where AI shopping assistants read product data from

The main product data paths across the assistants tracked here. Each row lists which assistants use the path and on what basis: documented by the operator, or observed.

Data sheets: one per assistant

One sheet per assistant, each in the same order and ending with its sources and last-verified date. Sheets are alphabetical; new assistants are added as they launch for US shoppers.

Which operators are reading brand product data, by the date they were first seen

When each operator's crawler was first seen fetching the Agentsites of nine brands on agentsite.network, and how its activity has developed since, from Autopilot's October 2026 analysis. Crawling shows that an operator fetched a page, not how it uses the data. Autopilot supplies product data to these Agentsites as a partner of Everything Machines, which builds and hosts them, so the sample is not independent of this page's publisher. Figures are updated as new months are reported.

The protocols and payment rails that move product data and purchases

Shared protocols, catalog services and payment rails that carry product data and purchases between merchants and assistants, with the product fields each defines and the assistants that use it.

Dated changes to AI shopping assistants and their product data paths

Every dated event behind the sheets and the matrix, newest first: launches, new product data sources, checkout rails, merchant programs and withdrawals. Each row states its basis (official, partner, press, observed or other) and links its source. Where a widely repeated claim turned out to be outdated, the row states the current fact.

Claims about AI shopping assistants that are no longer accurate

Statements that still circulate in articles and guides but are no longer accurate, each with the current fact and its source. Claims are added as they are found and removed when they stop circulating.

About this tracker