At several point, most of us operating in ecommerce have had that email. A client, a colleague, person senior, asking why a merchandise looks akin it’s on transaction whenever it isn’t. Or why an old merchandise depiction is showing up in Shopping. Or why a run that looks fine on Google is getting rejected location alternatively entirely.
You inspect what you continually check. The feed. The schema. The website. But you can’t fig out what’s causing the issue.
That’s normally the archetypal sign you’re dealing alongside Google’s hidden data layer.
Google isn’t lone operating alongside what you’re currently telling it. It builds a persistent recollection of your products. Prices, images, and another merchandise data are gathered complete period and sometimes stitched together throughout sources.
Like an elephant, it never forgets what it’s earlier seen on your own site, lengthy following you’ve moved on. And akin a truffle pig, it sometimes goes foraging fine beyond your domain entirely, into market listings and third-party pages you may not have idea concerning in years.
Meanwhile, nobody on the SEO or PPC squad is normally observing what it finds and remembers. Different teams own distinct layers, and audits don’t cross over.
And, you lone discover out item is incorrect whenever the email lands.
This part covers what that hidden tier really contains, how it manifests in ways that hurt performance, and what to appearance for whenever you’re already stuck.
What The ‘Hidden’ Layer Actually Is
The hidden tier is Google’s internally generated data concerning your products – gathered complete period and sometimes stitched together throughout sources.
When you present a product feed, you’re telling Google: this merchandise expenses £169.99, it’s in stock, here’s the image. This is your truth, correct now.
In comes the elephant.
Google additionally crawls your pages and revisits your nourish complete time, construction its own operating history. And that former keeps going lengthy following you’ve changed the underlying actuality on your side.
Then there is the truffle pig.
Google looks at your wider digital footprint on sites you don’t own, akin an old merchandise depiction motionless live on a market listing. It may additionally broaden to item equal small visible: data that platforms toggle immediately alongside all another rearward the scenes.
All this is crunched, verified, and can be taken into deliberation alongside any you’re currently submitting.
This is not several benevolent of clandestine agenda by Google. Ultimately, Google is fair trying to keep an accurate, ongoing image of your products, and the gap opens up whenever either flank stops updating: You ignore to inform Google item has changed, or Google’s own document hasn’t caught up yet. It’s not an uncomplicated task, on either end.
How Google finds and understands your data complete period and sources is complex. There are so many layers to it.
Here we volition conversation three components of the hidden tier I encountered during my activity and in conversations alongside another practitioners in the field:
- Price Tracking: Google crawls your merchandise pages, schema, and nourish complete period and builds a operating cost former of its own. If your schema labels, HTML copy, or historic nourish prices connote a sale, equal one that never existed as an definitive sale_price submission, Google’s cost recollection can confirm those signals and exterior an note nobody really wanted.
- Image Indexing: Google indexes your merchandise images, runs its own classifiers on them and applies flags or restrictions according to what it finds. Without a spotless indication that an old depiction is gone, Google can keep using an old merchandise depiction in your ads and liberated listings lengthy following you’ve ‘swapped’ it on the website or equal in the feed.
- Cross-Platform Data Exchange: While not publically confirmed, we doubtful important marketplaces portion merchandise nourish data alongside all another (possibly via paid API access). A error in your Merchant Center (MC) nourish can logic rejection in Google Ads and Amazon market simultaneously.
See also: Why Product Feeds Shouldn’t Be The Most Ignored SEO System In Ecommerce
Price Tracking
Many of us operating in ecommerce have had an email arrive in asking why it looks akin we are operating a transaction whenever we are not.
It’s uncomplicated for person not engaged in the day-to-day administration of ads or SEO to be confused. Google has, so far, three distinct price-related features and it’s extremely uncomplicated to conflate them. They all appearance akin and they can all be bundled under this “sale” umbrella.
In reality, these are extremely distinct alongside entirely distinct mechanisms rearward them.
First we have Sale cost annotations (or revenue badges).
These are the straightforward ones. You present sale_price and sale_price_effective_date in your feed. Google validates the discount is between 5% and 90%, and that the first basis cost has been submitted for a qualifying period. In the UK, at smallest 30 days inside the former 200 days. If you encounter those conditions, the sign shows. It shows since you told it to and since the cost former on document supports it.
The engaging component of transaction cost annotations is that Google doesn’t fair peruse your current submission. It validates your basis cost against its own document of what you’ve submitted historically. Which is the archetypal clue that Google’s cost recollection is doing additional activity than many of us realize.
Then we have Price autumn annotations.
These are a extremely distinct beast than your run-of-the-mill transaction cost annotations. You can’t authority cost autumn annotations. They are a outcome of Google crawling your page. Google compares your current submitted cost against the average cost it recorded for your merchandise complete the former 60 days and, if the autumn is important adequate against a stable baseline, it automatically generates a “Price drop” sign alongside a “Was” citation price. You never really present the “Was” figure.
Lastly, there are the Price autumn affluent snippets. These are the integrated hunt equivalent. To be eligible your Offer schema needs to have a particular cost and not the AggregateOffer cost alongside lowPrice and highPrice. But again, what Google really displays doesn’t fair arrive from your markup. As Brodie Clark, who first documented this feature in depth, noted: “It isn’t the location owner that is specifying this in the Structured Data. It is really Google strolling in and adding what their historic records concerning the leaf have been for the price.”
Google’s own records treats these as distinct features throughout distinct assistance pages which is part of the confusion. But the applicable consequences throughout all are the same: Google is holding a cost former and surfacing them in ways that aren’t continually apparent inner the tools you use to oversee your data.
Now let’s appearance at a use case.
You’ve likely seen annotations akin the ones below in the wild. They create it appearance akin the retailer is operating a deliberate sale. On the visage of it this looks akin a Sale Price badge. It has a percent decrease from the first cost and not fair a Price Drop label, commonly seen in cost autumn annotations.
Google was displaying “Was £474, now £355” in Shopping and a transaction cost autumn badge.
Yet, the client was adamant that no transaction was running.
So, anywhere did the “Was £474” arrive from?
First, we checked the current primary feed. The nourish carried £355 – the ex-VAT price, alongside no sale_price trait submitted. Not ideal since Google needs to infer VAT, but that’s a distinct conversation.
While we were in MC, we checked the latest crawl data. The final leaf crawl returned £426, which is £355 inc-VAT.
Then we looked at the live website and established the term “Now” in the cost display!
When we checked the schema markup, we additionally saw the location used AggregateOffer alongside a priceSpecification gathering and two tiers – one tier tagged “List price” at £395, one tagged “Sale price” at £355.
That “List price” of £395 is anywhere we established the mysterious “£474” price. £395 ex-VAT is £474 inc-VAT.
Since we had an older nourish export on file, we checked what the nourish cost had been three months before to confirm. On April 15, the cost site equal carried £474.
The merchandise had been repriced at several point. No one submitted a transaction cost in the feeds. Yet, Google assembled the transaction narrative from three autonomous signals: a schema tag it peruse as a transaction indicator, leaf HTML it peruse as a current-price marker, and a cost former it built from nourish submissions complete time. None of those signals individually said, “Run a transaction badge.” Together, they did.
People forget. Systems don’t.
Image Indexing
How website managers grip old merchandise images varies business to company. It’s frequently one of those boring processes that slips between the cracks and yet can logic issues if not done properly.
Some retailers delete the depiction document from the server whenever retiring a product, which forces a 404 on the old URL and gives Google a spotless indication to autumn it from the index. Others do a periodic cleanup of orphaned images at scale. While there are additionally those who depart the depiction document live on the CDN indefinitely. The second is the biggest issue, but a periodic cleanup is additionally problematic.
Without a 404, the depiction document is motionless publically accessible at its first URL, which method Google can continue to exterior it. Indefinitely or during the gap between cleanups.
Platforms grip this alternatively too. Shopify CDN URLs are imperishable by design, and deleting a merchandise doesn’t eliminate its images from the CDN. Magento stores images in a flat media directory, and unless person manually purges the file, it stays live. WooCommerce uploads sit in the norm WordPress media archive and are rarely cleaned up whenever products are retired.
The issue compounds whenever teams use depiction renaming conventions. A merchandise gets a new manner shot, the old filename stays on the server, and the new depiction is uploaded under a distinct name. Google has now indexed two depiction URLs for the identical merchandise and has to decide which one to associate, and it doesn’t continually choose the current one.
Google maintains its own depiction indicator for your merchandise images independently of what you’ve submitted in your feed’s g:image_link site or in any schema markup. When Google crawls a page, it finds the images on that page, stores them against that URL, and runs its own classification on them. That classification happens server-side. The results don’t appear anyplace in your nourish or schema audits.
The elephant never forgets.
You discover out concerning it lone whenever a client comes to you and asks: Why do we have an old merchandise image appearing on this ad?
This is exactly what happened alongside our client.
We could plainly see the offending depiction in MC.
So, we ran the customary checks. Looked at the images in assets, checked the feeds (primary and any supplemental), checked the rules, visited the website, checked the HTML and the schema.
And established nothing.
Only whenever we started thinking concerning indexing did we really arrive near to figuring out what was going on.
Google had indexed the old depiction URLs from former crawls. Updating the nourish and the leaf HTML removed the submission-side signal, but Google’s autonomous depiction indicator motionless held the old association.
Without the 404, there was no way for Google to update the index, and for several logic it decided that this depiction was the finest one to display as primary, despite of the nourish saying item different.
We discovered that the client had a purge sequence as their scheme to agreement alongside old images. Obviously, this needed a procedure tweak since Google was catching old images between the sequence and using them in ads as the chief depiction for the ads and the liberated listing.
Things get equal additional complex whenever we heap on the truffle pig nature of Google. Its foraging can go fine beyond your website and into places akin another marketplaces and third-party websites.
A awesome example of this was shared alongside me by International Structured Data and Semantic SEO adviser Jarno van Driel during a latest catch-up (hope we volition have many additional of those since it was brilliant!).
One of his clients had spent months construction out their feed, website, and schema correctly. Yet several of their best-selling products were consistently showing the incorrect merchandise depiction in hunt results, and nobody could activity out why.
Until a uncomplicated filename hunt revealed the depiction on an Amazon merchandise item leaf that an employee had manually created two years earlier. The depiction was never updated and lengthy since forgotten about.
Amazon is a huge brand alongside a lot of authority rearward it. So in a way it makes awareness why Google would think this depiction was crucial to surface.
But most ecommerce managers and SEOs won’t think to appearance there.
Which brings us to the final and perchance most engaging size of the hidden layer.
I haven’t encountered this immediately in client work, but it’s accordant alongside everything alternatively we cognize concerning how Google builds its merchandise data picture.
Cross-Platform Data Exchange
Everything we’ve covered so far has been concerning Google’s independently built data tier for your own products on your own location or Google selecting data it finds publically accessible that you have at several item provided and perchance forgotten about.
But the hidden tier doesn’t necessarily halt at this.
During our call, Jarno van Driel mentioned a scheme that most ecommerce practitioners have never considered.
Major marketplaces, Google, Amazon, and others, power be sharing nourish data alongside all another – perchance via several benevolent of paid API access. At smallest for now.
“Big important marketplaces pay all another for API admission to their merchandise feeds,” he told me. “So what ends up happening is that whenever you’ve got a error in your Merchant Center feed, that follows through all the way. Then you can get ad campaigns in Amazon rejected, and ad campaigns in Merchant Center rejected. There’s so much cross-matching between all those distinct data sets.”
He encountered this directly. The inquiry eventually led to a discrepancy between the Merchant Center nourish and the Amazon feed.
These are two totally distinct data sources that, on the surface, had nothing to do alongside all other. And yet they were influencing all another alongside genuine impacts on the ad side.
Jarno is apparent that there is no community records for the particular scheme he describes. It comes from conversations alongside engineers fairly than published policy. He additionally notes these are border cases and don’t happen to many companies. But they do happen, and knowing concerning them makes a genuine difference in hours spent trying to fig this out.
I tried to discover any community document of this institution and came up alongside nothing to verify this.
I did discover a publically documented data-sharing accord on a phase flat in broad between Google and Amazon. The Amazon MCF integration, announced by Google in 2024, method that Amazon can now provision fulfillment and shipping data immediately to the Merchant Center to power shipment speed estimates in Shopping.
Obviously, that’s barely the identical thing, but it shows the infrastructure and business association between the two platforms exists.
Whether merchandise nourish data flows between them, we can’t cognize for sure, but it is not implausible and fits the wider form of this part and what we cognize concerning another platforms sharing APIs.
When You’re Stuck
We shared three examples of whenever the hidden tier surfaced unexpectedly. Here’s how to examine whenever it happens to you.
Price Layer Checks
The easiest location to commencement is in MC: Go to Products > All products, click into any idiosyncratic product, and scroll to the base of the Product particulars tab.
The “Information established on your site” division shows you the cost and preparedness Google final crawled from your leaf HTML, alongside alongside the date it checked. This is Google’s autonomous crawl document and not what you submitted via the feed. Check the cost there.
If the crawled cost differs from your nourish price by exactly 20%, you nearly certainly have a VAT mismatch: your nourish submits ex-VAT, your leaf renders inc-VAT, and Google’s consistency inspect doesn’t cognize the difference.
If it differs by more, or matches a cost you haven’t submitted in months, you power have a cost former visibility that may be generating annotations you power not want.
You can frequently see all the products that have the distinct badges from the MC.
In your Products panel, fair display the below.
It’s certainly a helpful filter, although we established that sometimes the products don’t appear there. We didn’t see this client’s merchandise there whenever we checked, equal although it was displaying a cost note in Shopping results.
Do a quick MC check, but whether you discover the merchandise or not, the chief item is to inspect your data. Check your schema and the website (frontend AND code). If your schema mentions transaction anyplace in the markup or your HTML has Sale/Now/Was mentioned, and you are not really operating a sale, alter the label. It appears trivial, but it’s contributing to how Google assembles its image of your pricing.
Check your current feed, but additionally regard using a additional imperishable resolution to keep a comparable former on file. For me, that’s apt going to be evaluation out a personal repo on Git. We had the old nourish on document this period by luck. Next period we’ll have it by design.
Image Layer Checks
When a client reports a incorrect depiction appearing in Shopping or hunt results, inspect the run assets, the feed, and the live site, but afterward go further.
Dig deeper into the MC and discover all the images Google really has on document for that product. Take that URL, inspect the position code, and all the imaginable places it could be.
Search the filename in Google Images. If it appears on a third-party site, an old market listing, or a leaf you’ve forgotten concerning that should really be a 404, that’s apt the source. Google can drag depiction associations from anyplace it has crawled – your own domain or third-party sources.
This is why it’s crucial to not fair discover the depiction in inquiry but to think concerning what this method for your operational processes. Think concerning how you currently oversee your merchandise images:
- What happens whenever an depiction is removed from the website forefront end?
- How are you currently auditing for orphaned images on a location level?
- What scheme are you using, and how does this effect how you have to be managing this?
- How are you managing images throughout sources? Do you eliminate it fair from your website, or do you inspect another third-party websites?
- How can this be improved, and who owns it?
Cross-Platform Checks
If you’re operating on the two Google and Amazon and experiencing disapprovals or achievement anomalies that you can’t trace rear to item in your own nourish or schema, it’s a fine idea to drag the two nourish exports and put them next to all other.
Compare price, availability, title, and depiction URL for the affected products. You’re looking for any mismatch between what your MC nourish says and what your Amazon says. These are two data sources that most ecommerce managers and SEOs treat as entirely separate. But they may not be following all.
Also, inspect the timestamps to see whenever your MC nourish was final processed compared to your Amazon feed. If those dates differ, the two platforms may be operating from distinct versions of the identical merchandise data equal if you accept they’re in sync. Date-stamping your nourish exports whenever you audit is specified a small habit, and it can preserve you a ton of period and emphasis going downward rabbit holes.
Lastly, Actually Firstly: Deal With The Bigger Gap
Most ecommerce teams are already functioning alongside a fragmented image of their own merchandise data. That’s before Google adds its autonomous tier on top.
The hidden tier doesn’t create this problem. It lands inner one that already exists.
SEOs aren’t logging into MC. The nourish is treated as a PPC asset. Schema is treated as an SEO asset. The PPC director optimizing the feed has frequently never looked at the organized data on the merchandise page. The SEO auditing the schema has frequently never pulled a nourish export. Development owns the leaf but answers to neither. Ecommerce operations manages the merchandise catalog and the images but sits exterior the two conduit conversations entirely.
I’ve written before concerning the broad institution issue we have. There merely isn’t adequate co-ownership between PPC and SEO teams. It motionless surprises me how many SEOs have never equal logged into the MC!
Each squad audits what they submitted. And since they’re auditing separately, nobody has a shared perspective of what Google is really operating alongside throughout all three layers – feed, schema, page. What I fondly call the unholy trinity of ecommerce.
Google itself is trying to reconcile all the layers it uses to oversee ‘the truth’ concerning the products. For one, they’ve been wanting to unify schema.org markup and Merchant Center nourish data into one accordant merchandise data model.
Then there is the inquiry of all the distinct cognition graphs Google runs as the verification tier for the two traditional and AI search. Two of which are key to ecommerce companies: Knowledge Graph and Shopping Graph.
The Shopping Graph solitary now contains complete 50 milliard merchandise listings, alongside additional than 2 milliard of those refreshed all hour. The data is pulled from a broad set of sources, including Merchant Center and Manufacturer Center feeds, but additionally YouTube videos, manufacturer websites, merchandise item pages, merchandise evaluation data, and reviews.
This data is afterward cross-referenced against what Google understands concerning products, brands, and entities additional broadly via the Knowledge Graph.
How conflicting signals are valued and reconciled whenever they refuse all another throughout those layers is not publically documented.
Meanwhile, new AI-led standards are ballooning and adding additional complexity. Agentic commerce is no longer a forthcoming scenario. Google has already launched agentic checkout, anywhere a shopper can set a mark price, obtain a cost autumn notification, and have Google autonomously complete the acquisition on their behalf via Google Pay.
For that to activity accurately at scale, Google needs a sole authoritative fact concerning your merchandise (the correct price, image, availability…) pulled in real-time from everything it knows.
Right now that image is assembled from multiple conflicting sources throughout teams that aren’t talking to all other. And, as autonomous purchasing becomes the norm, the disbursal of that fragmentation goes up significantly.
All of us operating in this area – SEOs, PPC managers, developers, ecommerce operations – are ultimately operating toward the identical thing: a single, accurate, accordant image of our products that all scheme can trust. Google is trying to build that from its end. The hidden tier is what happens in the gap during we capture up from ours.
More Resources:
- The Web Is Growing A Second Layer – Almost A Third Head
- The Technical Signals AI Search Uses That Most SEOs Still Aren’t Optimizing
- Stop Treating AI Visibility As One Problem. It’s Actually Three, On Three Different Layers
Featured Image: Roman Samborskyi/Shutterstock
