Google's Shopping Graph now holds more than 50 billion product listings, with over 2 billion refreshed every hour, per an extensive breakdown published by ecommerce SEO practitioner Demir Emektar on Search Engine Journal. That graph is built not only from merchant feeds, but from a parallel, persistent store of price history, image classifications and cross-platform signals that Google assembles on its own. Emektar labels this surface the "hidden data layer."

What exactly sits in this hidden layer?

Google maintains an internal record of every product it has seen on a merchant's site, crawled and replayed over months and years. The record crosses three domains, according to Emektar's casework and interviews:

• Price tracking. Google crawls product pages, schema and feed submissions to build a running price history independent of the merchant's current submission.
• Image indexing. Google stores and classifies product images on its own servers and uses those classifications to pick primary images for Shopping ads and free listings, regardless of what g:image_link or schema says today.
• Cross-platform data exchange. Major marketplaces may share product feed data via paid API access, so a Merchant Center feed error can surface as a rejection in Amazon as well.

These behaviors surface only after something visibly breaks: a client emails asking why a non-sale looks like a sale, or why an old lifestyle shot shows up in a Shopping ad.

Why do sale badges appear when no sale was set?

Sale price annotations look like simple feed fields, but Google validates them against its own record. In the UK, that record must show the base price for at least 30 of the past 200 days, with the discount landing between 5% and 90%.

Price drop annotations work differently. Google computes them by comparing the current submitted price against the average price it recorded for the product over the past 60 days. If the drop is large enough against a stable baseline, Google adds the badge automatically. The "Was" figure is never submitted by the merchant.

Emektar's casework shows the pattern clearly. One unnamed client was displaying "Was £474, now £355" in Shopping with a sale badge:

• The current feed carried £355 ex-VAT, with no sale_price submitted.
• The latest crawl returned £426, which is £355 inc-VAT.
• The page HTML included the word "Now."
• Schema used AggregateOffer with a priceSpecification array labeled "List price" at £395 — equivalent to £474 inc-VAT.
• An older feed export from April 15 still carried £474.

Brodie Clark, who first documented Google stepping in to populate price drop rich snippets, told Emektar: "It isn't the site owner that is specifying this in the Structured Data. It is actually Google stepping in and adding what their historical records about the page have been for the price." Three independent signals combined to produce a sale annotation nobody configured.

Why do old product images keep appearing in Shopping?

When retailers swap a product image, they typically replace the file in the CMS or feed. Google maintains its own image index populated by crawls rather than feed submissions. Without a 404 on the old URL, that index keeps associating the retired image with the product indefinitely. Shopify stores leave CDN URLs live after product deletion; Magento keeps files in a flat media directory; WooCommerce rarely clears the WordPress media library on retirement.

Structured data consultant Jarno van Driel worked a case where the wrong product image surfaced across multiple SKUs. A filename search eventually located the image on an Amazon detail page an employee had manually created two years earlier. Google kept surfacing that listing because of Amazon's authority, not because the merchant submitted it.

Are marketplaces actually sharing feed data?

Van Driel argues that they are, at least at the platform level. "Big major marketplaces pay each other for API access to their product feeds," he told Emektar. "So what ends up happening is that when you've got a mistake in your Merchant Center feed, that follows through all the way."

Van Driel says the specific mechanism is not publicly documented and comes from engineer conversations rather than published policy. The closest publicly confirmed arrangement is the Amazon Multi-Channel Fulfillment integration Google announced in 2024, which sends shipping data from Amazon into Merchant Center to power delivery estimates.

What should ecommerce teams monitor next?

The single most concrete check costs nothing. In Merchant Center, open Products > All products, click into a SKU, and scroll to "Information found on your site." That panel shows the price and availability Google last crawled, with the timestamp. A 20% gap between the crawled price and the feed price typically indicates a VAT mismatch. A gap matching a price last submitted months earlier signals a price-history exposure.

Emektar recommends three operational habits going forward:

• Date-stamp every feed export so old versions stay auditable.
• Audit orphaned images at site, CDN and third-party-marketplace levels, not just inside the CMS.
• Reconcile Merchant Center and Amazon feed exports side by side when cross-channel rejections appear without an obvious feed cause.

With Google's agentic checkout now live — where shoppers can set a target price and let Google complete the purchase via Google Pay — the cost of internal product-data fragmentation is set to rise sharply for any retailer whose private "truth" does not match what Google has quietly assembled.