Sooner or later, most of us working in ecommerce have had that electronic mail. A shopper, a colleague, somebody senior, asking why a product seems prefer it’s on sale when it isn’t. Or why an previous product picture is exhibiting up in Procuring. Or why a marketing campaign that appears advantageous on Google is getting rejected someplace else totally.
You verify what you all the time verify. The feed. The schema. The web site. However you possibly can’t work out what’s inflicting the problem.
That’s normally the primary signal you’re coping with Google’s hidden knowledge layer.
Google isn’t solely working with what you’re at the moment telling it. It builds a persistent reminiscence of your merchandise. Costs, photographs, and different product knowledge are gathered over time and generally stitched collectively throughout sources.
Like an elephant, it by no means forgets what it’s beforehand seen by yourself web site, lengthy after you’ve moved on. And like a truffle pig, it generally goes foraging properly past your area totally, into market listings and third-party pages you could not have considered in years.
In the meantime, no one on the web optimization or PPC staff is normally watching what it finds and remembers. Totally different groups personal completely different layers, and audits don’t cross over.
And, you solely discover out one thing is improper when the e-mail lands.
This piece covers what that hidden layer really comprises, the way it manifests in ways in which harm efficiency, and what to search for while you’re already caught.
What The ‘Hidden’ Layer Really Is
The hidden layer is Google’s internally generated knowledge about your merchandise – gathered over time and generally stitched collectively throughout sources.
Once you submit a product feed, you’re telling Google: this product prices £169.99, it’s in inventory, right here’s the picture. That is your reality, proper now.
In comes the elephant.
Google additionally crawls your pages and revisits your feed over time, constructing its personal working historical past. And that historical past retains going lengthy after you’ve modified the underlying actuality in your facet.
Then there’s the truffle pig.
Google seems at your wider digital footprint on websites you don’t personal, like an previous product picture nonetheless stay on a market itemizing. It might additionally lengthen to one thing even much less seen: knowledge that platforms trade immediately with one another behind the scenes.
All that is crunched, verified, and could be considered alongside no matter you’re at the moment submitting.
This isn’t some form of clandestine agenda by Google. Finally, Google is simply attempting to maintain an correct, ongoing image of your merchandise, and the hole opens up when both facet stops updating: You neglect to inform Google one thing has modified, or Google’s personal report hasn’t caught up but. It’s not a simple process, on both finish.
How Google finds and understands your knowledge over time and sources is complicated. There are such a lot of layers to it.
Right here we’ll talk about three elements of the hidden layer I encountered throughout my work and in conversations with different practitioners within the discipline:
- Value Monitoring: Google crawls your product pages, schema, and feed over time and builds a working worth historical past of its personal. In case your schema labels, HTML copy, or historic feed costs indicate a sale, even one which by no means existed as an specific sale_price submission, Google’s worth reminiscence can corroborate these alerts and floor an annotation no one really needed.
- Picture Indexing: Google indexes your product photographs, runs its personal classifiers on them and applies flags or restrictions based mostly on what it finds. With no clear sign that an previous picture is gone, Google can hold utilizing an previous product picture in your advertisements and free listings lengthy after you’ve ‘swapped’ it on the web site and even within the feed.
- Cross-Platform Knowledge Change: Whereas not publicly confirmed, we suspect main marketplaces share product feed knowledge with one another (probably by way of paid API entry). A mistake in your Service provider Heart (MC) feed may cause rejection in Google Adverts and Amazon market concurrently.
See additionally: Why Product Feeds Shouldn’t Be The Most Ignored web optimization System In Ecommerce
Value Monitoring
Many people working in ecommerce have had an electronic mail are available asking why it seems like we’re working a sale when we aren’t.
It’s straightforward for somebody not concerned within the day-to-day administration of advertisements or web optimization to be confused. Google has, to date, three distinct price-related options and it’s very straightforward to conflate them. All of them look related and so they can all be bundled beneath this “sale” umbrella.
In actuality, these are very completely different with utterly completely different mechanisms behind them.
First we’ve got Sale worth annotations (or gross sales badges).
These are the simple ones. You submit sale_price and sale_price_effective_date in your feed. Google validates the low cost is between 5% and 90%, and that the unique base worth has been submitted for a qualifying interval. Within the UK, a minimum of 30 days inside the previous 200 days. For those who meet these situations, the badge exhibits. It exhibits since you advised it to and since the value historical past on report helps it.
The fascinating ingredient of sale worth annotations is that Google doesn’t simply learn your present submission. It validates your base worth in opposition to its personal report of what you’ve submitted traditionally. Which is the primary trace that Google’s worth reminiscence is doing extra work than many people understand.
Then we’ve got Value drop annotations.
These are a really completely different beast than your run-of-the-mill sale worth annotations. You possibly can’t management worth drop annotations. They’re a results of Google crawling your web page. Google compares your present submitted worth in opposition to the common worth it recorded to your product over the previous 60 days and, if the drop is important sufficient in opposition to a steady baseline, it mechanically generates a “Value drop” badge with a “Was” reference worth. You by no means really submit the “Was” determine.
Lastly, there are the Value drop wealthy snippets. These are the natural search equal. To be eligible your Provide schema must have a particular worth and never the AggregateOffer worth with lowPrice and highPrice. However once more, what Google really shows doesn’t simply come out of your markup. As Brodie Clark, who first documented this characteristic in depth, famous: “It isn’t the positioning proprietor that’s specifying this within the Structured Knowledge. It’s really Google stepping in and including what their historic information in regards to the web page have been for the value.”
Google’s personal documentation treats these as separate options throughout separate assist pages which is a part of the confusion. However the sensible penalties throughout all are the identical: Google is holding a worth historical past and surfacing them in ways in which aren’t all the time seen contained in the instruments you utilize to handle your knowledge.
Now let’s have a look at a use case.
You’ve in all probability seen annotations like those beneath within the wild. They make it appear to be the retailer is working a deliberate sale. On the face of it this seems like a Sale Value badge. It has a share lower from the unique worth and never only a Value Drop label, generally seen in worth drop annotations.
Google was displaying “Was £474, now £355” in Procuring and a sale worth drop badge.
But, the shopper was adamant that no sale was working.
So, the place did the “Was £474” come from?
First, we checked the present main feed. The feed carried £355 – the ex-VAT worth, with no sale_price attribute submitted. Not very best since Google must infer VAT, however that’s a separate dialog.
Whereas we had been in MC, we checked the newest crawl knowledge. The final web page crawl returned £426, which is £355 inc-VAT.
Then we seemed on the stay web site and located the phrase “Now” within the worth show!
Once we checked the schema markup, we additionally noticed the positioning used AggregateOffer with a priceSpecification array and two tiers – one tier labeled “Checklist worth” at £395, one labeled “Sale worth” at £355.
That “Checklist worth” of £395 is the place we discovered the mysterious “£474” worth. £395 ex-VAT is £474 inc-VAT.
Since we had an older feed export on file, we checked what the feed worth had been three months earlier to substantiate. On April 15, the value discipline even carried £474.
The product had been repriced sooner or later. Nobody submitted a sale worth within the feeds. But, Google assembled the sale narrative from three unbiased alerts: a schema label it learn as a sale indicator, web page HTML it learn as a current-price marker, and a worth historical past it constructed from feed submissions over time. None of these alerts individually mentioned, “Run a sale badge.” Collectively, they did.
Individuals neglect. Techniques don’t.
Picture Indexing
How web site managers deal with previous product photographs varies firm to firm. It’s typically a kind of boring processes that slips between the cracks and but may cause points if not accomplished correctly.
Some retailers delete the picture file from the server when retiring a product, which forces a 404 on the previous URL and offers Google a clear sign to drop it from the index. Others do a periodic cleanup of orphaned photographs at scale. Whereas there are additionally those that depart the picture file stay on the CDN indefinitely. The latter is the most important problem, however a periodic cleanup can be problematic.
With no 404, the picture file remains to be publicly accessible at its authentic URL, which suggests Google can proceed to floor it. Indefinitely or through the hole between cleanups.
Platforms deal with this in another way too. Shopify CDN URLs are everlasting by design, and deleting a product doesn’t take away its photographs from the CDN. Magento shops photographs in a flat media listing, and until somebody manually purges the file, it stays stay. WooCommerce uploads sit in the usual WordPress media library and are not often cleaned up when merchandise are retired.
The issue compounds when groups use picture renaming conventions. A product will get a brand new way of life shot, the previous filename stays on the server, and the brand new picture is uploaded beneath a special title. Google has now listed two picture URLs for a similar product and has to determine which one to affiliate, and it doesn’t all the time select the present one.
Google maintains its personal picture index to your product photographs independently of what you’ve submitted in your feed’s g:image_link discipline or in any schema markup. When Google crawls a web page, it finds the pictures on that web page, shops them in opposition to that URL, and runs its personal classification on them. That classification occurs server-side. The outcomes don’t seem anyplace in your feed or schema audits.
The elephant by no means forgets.
You discover out about it solely when a shopper involves you and asks: Why do we’ve got an previous product image showing on this advert?
That is precisely what occurred with our shopper.
We might clearly see the offending picture in MC.
So, we ran the same old checks. Appeared on the photographs in belongings, checked the feeds (main and any supplemental), checked the principles, visited the web site, checked the HTML and the schema.
And located nothing.
Solely after we began excited about indexing did we really come near determining what was occurring.
Google had listed the previous picture URLs from earlier crawls. Updating the feed and the web page HTML eliminated the submission-side sign, however Google’s unbiased picture index nonetheless held the previous affiliation.
With out the 404, there was no approach for Google to replace the index, and for some purpose it determined that this picture was the very best one to point out as main, whatever the feed saying one thing completely different.
We found that the shopper had a purge cycle as their technique to cope with previous photographs. Clearly, this wanted a course of tweak since Google was catching previous photographs between the cycle and utilizing them in advertisements as the first picture for the advertisements and the free itemizing.
Issues get much more complicated after we pile on the truffle pig nature of Google. Its foraging can go properly past your web site and into locations like different marketplaces and third-party web sites.
An awesome instance of this was shared with me by Worldwide Structured Knowledge and Semantic web optimization marketing consultant Jarno van Driel throughout a latest catch-up (hope we may have many extra of these because it was good!).
One in every of his purchasers had spent months constructing out their feed, web site, and schema appropriately. But a few of their best-selling merchandise had been persistently exhibiting the improper product picture in search outcomes, and no one might work out why.
Till a easy filename search revealed the picture on an Amazon product element web page that an worker had manually created two years earlier. The picture was by no means up to date and lengthy since forgotten about.
Amazon is a large model with quite a lot of authority behind it. So in a approach it is smart why Google would suppose this picture was essential to floor.
However most ecommerce managers and SEOs gained’t suppose to look there.
Which brings us to the final and probably most fascinating dimension of the hidden layer.
I haven’t encountered this immediately in shopper work, nevertheless it’s in keeping with every part else we learn about how Google builds its product knowledge image.
Cross-Platform Knowledge Change
Every little thing we’ve lined to date has been about Google’s independently constructed knowledge layer to your personal merchandise by yourself web site or Google deciding on knowledge it finds publicly obtainable that you’ve sooner or later supplied and perhaps forgotten about.
However the hidden layer doesn’t essentially cease at this.
Throughout our name, Jarno van Driel talked about a mechanism that almost all ecommerce practitioners have by no means thought-about.
Main marketplaces, Google, Amazon, and others, is likely to be sharing feed knowledge with one another – probably by way of some form of paid API entry. A minimum of for now.
“Huge main marketplaces pay one another for API entry to their product feeds,” he advised me. “So what finally ends up taking place is that while you’ve obtained a mistake in your Service provider Heart feed, that follows by way of all the best way. Then you may get advert campaigns in Amazon rejected, and advert campaigns in Service provider Heart rejected. There’s a lot cross-matching between all these completely different knowledge units.”
He encountered this immediately. The investigation finally led to a discrepancy between the Service provider Heart feed and the Amazon feed.
These are two completely completely different knowledge sources that, on the floor, had nothing to do with one another. And but they had been influencing one another with actual impacts on the advert facet.
Jarno is obvious that there isn’t a public documentation for the precise mechanism he describes. It comes from conversations with engineers quite than printed coverage. He additionally notes these are edge circumstances and don’t occur to many corporations. However they do occur, and understanding about them makes an actual distinction in hours spent attempting to determine this out.
I attempted to seek out any public report of this association and got here up with nothing to substantiate this.
I did discover a publicly documented data-sharing settlement on a platform degree typically between Google and Amazon. The Amazon MCF integration, introduced by Google in 2024, implies that Amazon can now present success and transport knowledge on to the Service provider Heart to energy supply velocity estimates in Procuring.
Clearly, that’s hardly the identical factor, nevertheless it exhibits the infrastructure and industrial relationship between the 2 platforms exists.
Whether or not product feed knowledge flows between them, we will’t know for positive, however it isn’t implausible and suits the broader sample of this text and what we learn about different platforms sharing APIs.
When You’re Caught
We shared three examples of when the hidden layer surfaced unexpectedly. Right here’s methods to examine when it occurs to you.
Value Layer Checks
The best place to start out is in MC: Go to Merchandise > All merchandise, click on into any particular person product, and scroll to the underside of the Product particulars tab.
The “Info discovered in your web site” part exhibits you the value and availability Google final crawled out of your web page HTML, together with the date it checked. That is Google’s unbiased crawl report and never what you submitted by way of the feed. Test the value there.
If the crawled worth differs out of your feed worth by precisely 20%, you nearly actually have a VAT mismatch: your feed submits ex-VAT, your web page renders inc-VAT, and Google’s consistency verify doesn’t know the distinction.
If it differs by extra, or matches a worth you haven’t submitted in months, you may need a worth historical past publicity that could be producing annotations you won’t need.
You possibly can typically see all of the merchandise which have the completely different badges from the MC.
In your Merchandise panel, simply filter the beneath.
It’s undoubtedly a helpful filter, though we discovered that generally the merchandise don’t seem there. We didn’t see this shopper’s product there after we checked, despite the fact that it was displaying a worth annotation in Procuring outcomes.
Do a fast MC verify, however whether or not you discover the product or not, the principle level is to verify your knowledge. Test your schema and the web site (frontend AND code). In case your schema mentions sale anyplace within the markup or your HTML has Sale/Now/Was talked about, and you aren’t really working a sale, change the label. It sounds trivial, nevertheless it’s contributing to how Google assembles its image of your pricing.
Test your present feed, but additionally think about using a extra everlasting resolution to maintain a comparable historical past on file. For me, that’s seemingly going to be testing out a personal repo on Git. We had the previous feed on file this time by luck. Subsequent time we’ll have it by design.
Picture Layer Checks
When a shopper reviews a improper picture showing in Procuring or search outcomes, verify the marketing campaign belongings, the feed, and the stay web site, however then go additional.
Dig deeper into the MC and discover all the pictures Google really has on file for that product. Take that URL, verify the standing code, and all of the attainable locations it could possibly be.
Search the filename in Google Pictures. If it seems on a third-party web site, an previous market itemizing, or a web page you’ve forgotten about that ought to really be a 404, that’s seemingly the supply. Google can pull picture associations from anyplace it has crawled – your personal area or third-party sources.
This is the reason it’s essential to not simply discover the picture in query however to consider what this implies to your operational processes. Take into consideration the way you at the moment handle your product photographs:
- What occurs when a picture is faraway from the web site entrance finish?
- How are you at the moment auditing for orphaned photographs on a web site degree?
- What system are you utilizing, and the way does this influence how you ought to be managing this?
- How are you managing photographs throughout sources? Do you take away it simply out of your web site, or do you verify different third-party web sites?
- How can this be improved, and who owns it?
Cross-Platform Checks
For those who’re working on each Google and Amazon and experiencing disapprovals or efficiency anomalies that you may’t hint again to something in your personal feed or schema, it’s a good suggestion to tug each feed exports and put them subsequent to one another.
Evaluate worth, availability, title, and picture URL for the affected merchandise. You’re searching for any mismatch between what your MC feed says and what your Amazon says. These are two knowledge sources that almost all ecommerce managers and SEOs deal with as totally separate. However they will not be in any case.
Additionally, verify the timestamps to see when your MC feed was final processed in comparison with your Amazon feed. If these dates differ, the 2 platforms could also be working from completely different variations of the identical product knowledge even in case you consider they’re in sync. Date-stamping your feed exports while you audit is such a small behavior, and it will probably prevent a ton of time and stress taking place rabbit holes.
Lastly, Really Firstly: Deal With The Greater Hole
Most ecommerce groups are already working with a fragmented image of their very own product knowledge. That’s earlier than Google provides its unbiased layer on prime.
The hidden layer doesn’t create this downside. It lands inside one which already exists.
SEOs aren’t logging into MC. The feed is handled as a PPC asset. Schema is handled as an web optimization asset. The PPC supervisor optimizing the feed has typically by no means seemed on the structured knowledge on the product web page. The web optimization auditing the schema has typically by no means pulled a feed export. Improvement owns the web page however solutions to neither. Ecommerce operations manages the product catalog and the pictures however sits exterior each channel conversations totally.
I’ve written earlier than in regards to the basic group downside we’ve got. There merely isn’t sufficient co-ownership between PPC and web optimization groups. It nonetheless surprises me what number of SEOs have by no means even logged into the MC!
Every staff audits what they submitted. And since they’re auditing individually, no one has a shared view of what Google is definitely working with throughout all three layers – feed, schema, web page. What I fondly name the unholy trinity of ecommerce.
Google itself is attempting to reconcile all of the layers it makes use of to handle ‘the reality’ in regards to the merchandise. For one, they’ve been desirous to unify schema.org markup and Service provider Heart feed knowledge into one constant product knowledge mannequin.
Then there’s the query of all of the completely different information graphs Google runs because the verification layer for each conventional and AI search. Two of that are key to ecommerce corporations: Data Graph and Procuring Graph.
The Procuring Graph alone now comprises over 50 billion product listings, with greater than 2 billion of these refreshed each hour. The information is pulled from a large set of sources, together with Service provider Heart and Producer Heart feeds, but additionally YouTube movies, producer web sites, product element pages, product testing knowledge, and critiques.
This knowledge is then cross-referenced in opposition to what Google understands about merchandise, manufacturers, and entities extra broadly by way of the Data Graph.
How conflicting alerts are weighted and reconciled once they contradict one another throughout these layers is just not publicly documented.
In the meantime, new AI-led requirements are ballooning and including additional complexity. Agentic commerce is not a future state of affairs. Google has already launched agentic checkout, the place a client can set a goal worth, obtain a worth drop notification, and have Google autonomously full the acquisition on their behalf by way of Google Pay.
For that to work precisely at scale, Google wants a single authoritative reality about your product (the proper worth, picture, availability…) pulled in real-time from every part it is aware of.
Proper now that image is assembled from a number of conflicting sources throughout groups that aren’t speaking to one another. And, as autonomous shopping for turns into the norm, the price of that fragmentation goes up considerably.
All of us working on this house – SEOs, PPC managers, builders, ecommerce operations – are finally working towards the identical factor: a single, correct, constant image of our merchandise that each system can belief. Google is attempting to construct that from its finish. The hidden layer is what occurs within the hole whereas we catch up from ours.
Extra Sources:
Featured Picture: Roman Samborskyi/Shutterstock
