Shopping Graph

Google's AI-powered knowledge graph of products — an entity database (not a page index) that underpins Google Shopping, AI Mode, AI Overviews, and the Universal Cart, holding over 60 billion product listings refreshed at more than 2 billion updates per hour as of May 2026.

Architecture

The Shopping Graph models products as nodes rather than pages. A single product node aggregates seller offers, prices, images, reviews, variants, and attributes from multiple input streams, reconciling them into a unified entity.

Input streams confirmed by Google:

Google Merchant Center feed submissions (structured product data from retailers)
Web crawl of retailer and manufacturer product detail pages
Structured Data markup (Schema.org Product, Offer, Review) parsed from those pages
Google's proprietary signals (user behaviour, reviews, image analysis)

Additional streams identified by practitioners (BrightonSEO October 2024, Imogen Groome): Google Manufacturer Center submissions, YouTube product videos, and user-generated product reviews also feed the graph.

Entity reconciliation runs identifier-first: GTIN (Global Trade Item Number) where available, then brand + MPN (manufacturer part number) to attach seller offers to the correct product node. When identifiers are absent, Google uses textual and image signals. (Sources: Web — Shopping Graph 2026-09-11, YouTube — Shopping Graph 2026-09-11)

Scale

Milestone	Listing count	Refresh rate	Source
September 2022	35 billion	1.8 billion/hour	blog.google Search On 2022
May 2025 (I/O 2025)	50+ billion	2+ billion/hour	Google I/O 2025 keynote (Lilian Rincon, VP Consumer Shopping)
January 2026 (NRF)	50+ billion	—	Sundar Pichai NRF 2026 remarks, blog.google
May 2026 (GML 2026)	60+ billion	2+ billion/hour	blog.google Universal Cart launch, GML 2026

Scale benchmark: People shop across Google more than 1 billion times a day, powered by the Shopping Graph (as-of 2026-05-20). (Source: blog.google Universal Cart post)

How it powers Google surfaces

The Shopping Graph is the single data source underlying all Google shopping features — organic free listings, Shopping ads, AI Overviews, AI Mode panels, and the Universal Cart.

Google Shopping — organic free listings and Shopping ads draw product data, pricing, availability, and imagery from the graph.

AI Mode (Google) — queries routed to AI Mode trigger Gemini to retrieve product nodes from the Shopping Graph via RAG (Retrieval Augmented Generation) and render product panels with synthesised summaries. AI Mode had surpassed 1 billion monthly users by GML 2026. (Source: searchengineland.com GML 2026 recap)

AI Overviews — when a shopping query triggers an AI Overview, product nodes from the graph are used to populate the overview, alongside web content. AI Overviews appeared on approximately 14% of shopping-related queries as of March 2026 (as-of 2026-03, secondary analysis via almcorp.com; no primary Google publication confirmed this exact figure).

Zero-Click Search — the Shopping Graph enables richer on-SERP answers that reduce the need to click through. Overall zero-click rate reached 68.01% of searches (Jan–Apr 2026; Search Engine Land / Datos study). Within AI Mode specifically, zero-click reached 93% (as-of 2026, Nobori AI; vendor source).

Virtual Try-On (Google) — launched September 2025, powered by the Shopping Graph and a custom AI image generation model for fashion, allowing virtual try-on of items in the graph. (Source: blog.google, 2025-09)

Shopping ads in AI Mode — from GML 2026, shopping ads inside AI Mode include Gemini-generated copy explaining why a product matches the query, using data from the Merchant Center feed. (Source: ppc.land GML 2026 recap)

Shopping Graph and Universal Commerce Protocol

Universal Commerce Protocol (UCP) (launched NRF January 2026) establishes a standard API surface for AI agents to discover products, manage carts, and complete checkouts across merchant systems. The Shopping Graph supplies the product-discovery layer UCP agents query.

Practitioners describe a data feedback loop: UCP feeds transaction data back into the Shopping Graph; the graph trains Gemini models; the model executes better agentic transactions. Google has not publicly confirmed this loop in these exact terms, but the architecture is consistent with official UCP documentation. (Source: hypotenuse.ai, 2026 — vendor analysis)

UCP timeline:

January 2026 (NRF): UCP launched with Shopify, Etsy, Wayfair, Target, Walmart; 20+ partners including Visa, Mastercard, Stripe, American Express, Best Buy, The Home Depot
March 2026: simplified Merchant Center onboarding for UCP
May 2026 (GML 2026): Universal Cart (Google) launched cross-retailer (Search, Gemini, YouTube, Gmail); BNPL (Affirm, Klarna) inside Google Pay; geographic expansion to Canada, Australia, UK; Amazon, Meta, Microsoft, Salesforce, Stripe join UCP steering group (per ppc.land — should be cross-checked against official Google blog post)
November 2025: agentic checkout launched with Wayfair, Chewy, Quince, select Shopify merchants; Business Agent (Google) went live January 2026 with Lowe's, Michaels, Poshmark, Reebok

Structured data and the Shopping Graph

The Shopping Graph ingests Schema.org Product markup from pages alongside Merchant Center feed data. The two sources serve overlapping but distinct functions:

Merchant Center feed: primary vehicle for offer data (price, availability, variants, images) and eligibility for Shopping ads
Structured data on page: enables rich results in organic search; supplies machine-readable product signals to AI Mode and AI Overviews; cross-validates feed data inside the graph

Key rule: when structured data on the page, the Merchant Center feed, and the visible page content disagree on price or availability, rich results are suppressed. Data alignment across all three layers is a core merchant optimisation challenge. (Source: panem.agency, 2026)

2025–2026 structured data updates:

February 2024: Google added Product Variants structured data support, allowing declaration of variations in markup (size, colour, material) rather than only through feeds. (as-of 2024-02; stale-risk: may have been extended since)
November 2025: Google added two new pathways for shipping and return policies — via Search Console settings or organisation-level structured data — for merchants without Merchant Center. (as-of 2025-11)
2026: Conversational Attributes launched as a new optional Merchant Center attribute type — machine-readable answers to common product questions, compatible accessories, and substitutes, designed for AI Mode and Gemini discovery. (as-of 2026; volatile — feature in rollout)

Merchant optimisation

Core requirements for Shopping Graph inclusion:

Submit a complete Merchant Center feed with accurate GTINs, high-quality images (from June 2026: minimum resolution requirements; enforcement January 2027), pricing, availability, and description
Add Schema.org Product markup on PDPs, aligned with the feed
Keep pricing and availability current — outdated data can prevent surfacing in AI Mode altogether
For AI-layer visibility: submit Conversational Attributes answering natural-language product questions

Feed completeness matters more than domain authority for AI Mode visibility: machine-readable product data, feed completeness, and structured markup matter more than homepage authority (practitioner consensus, consistent with Google guidance; source: panem.agency, 2026).

New 2026 Merchant Center developments:

Conversational Attributes attribute schema launched (help Google understand products in natural language)
AI performance insights in Merchant Center to show how products surface across AI Mode, AI Overviews, and Gemini
Content API for Shopping sunsetting progressively from September 2026; replaced by Merchant API (Google)
"Merchant Center Next" branding retired July 2026, reverting to "Google Merchant Center"
Image minimum resolution requirements: warnings from April 14, 2026; enforcement from January 31, 2027 (as-of 2026-04-14)
New optional attributes: video_link (eligible June 30, 2026), handling_cutoff_time, minimum_order_value

Controversies

Surveillance pricing concerns

Senator Elizabeth Warren published a letter on February 4, 2026 raising concerns that Google's retail AI partnerships could enable personalized pricing — adjusting prices to the highest amount an individual user is likely to pay — using personal data collected across Google surfaces. (Source: warren.senate.gov, 2026-02-04)

Tech Policy Press (2026) reported that Big Tech's AI shopping tools raise legislative stakes for "surveillance pricing" laws, as AI systems can in theory set personalized prices based on inferred willingness to pay. A Morning Consult survey found 59% of US adults consider AI-powered price gouging a "major concern" (cited in secondary analysis; original survey URL not verified — treat as low confidence).

AI Mode price discrepancies

A study by Productrise (2026) identified price discrepancies between what Google's AI Mode shows and what regular search results show for the same products, questioning how AI Mode arrives at its recommendations.

Key terms

Term	Meaning
GTIN	Global Trade Item Number — primary identifier for entity matching in the Shopping Graph
MPN	Manufacturer Part Number — secondary identifier when GTIN is absent
Merchant Center	Google's primary feed submission portal for Shopping Graph ingestion
Conversational Attributes	New 2026 Merchant Center attribute type for AI-layer product discovery
RAG	Retrieval Augmented Generation — technique by which Gemini retrieves product nodes from the Shopping Graph to answer shopping queries
UCP	Universal Commerce Protocol — open standard for agentic commerce built on top of the Shopping Graph
Content API for Shopping	Legacy API (sunsetting 2026) for programmatic Merchant Center feed management
Merchant API	Replacement for Content API, available from 2026

Benchmarks (as-of 2026-05-20)

60+ billion product listings
2+ billion listing refreshes per hour
1+ billion shopping interactions per day on Google

What practitioners report

BrightonSEO October 2024 (Imogen Groome): optimize for Shopping Graph by conducting sentiment analysis on product reviews; list specific attributes per product and develop targeted review questions; prioritize products using sales data. RAG acts "like a librarian" pulling relevant product information from the graph to power AI Overviews.