Google Merchant Centre Stopped Being an Ads Tool. Most Fashion Brands Haven't Noticed.

I was doing a feed audit for a brand about three months ago. Solid products, decent site, reasonable organic traffic. When I looked up their hero product in a Google AI Overview query for their category, what came back was thin. A title, a price, a truncated description. No material composition. No fit information. No product highlights. Just enough data for a listing, not enough for a recommendation.

The feed was technically clean. No disapprovals, no warnings. From an advertising perspective it was fine. But it was sitting next to competitor product cards that told a complete story, and in that comparison it looked like a placeholder. The AI had nothing to work with beyond the basics, so it gave the basics back.

That's the conversation most fashion brands aren't having about Google Merchant Centre (GMC).

For years GMC had a reputation problem, and honestly it earned it. Most ecommerce teams only opened it when an ad got disapproved. It lived in the gap between paid media and technical SEO, so in practice nobody really owned it, and the work that should have been done on it quietly didn't get done. Feeding a product title, a price, and a category was enough to run Shopping ads, and that was the job.

That job has changed.

Google's Shopping Graph is the data layer that powers AI Overviews, Gemini's product recommendations, and Shopping surfaces across Google's ecosystem. It's how Google's AI understands what your products are, what they contain, who they're for, and whether they match what someone is asking for. And the primary way you feed information into the Shopping Graph is through Google Merchant Centre.

If your GMC data is thin, Google's AI understanding of your products is thin. That's not a theory. It's the mechanism.

The product card is the clearest way to see this in practice.

When a user asks an AI shopping agent for a recommendation, or when Google generates an AI Overview for a product category query, the output includes product cards. These aren't just links. They're structured summaries that pull directly from your product feed: title, price, availability, images, material, fit, key features, reviews. The AI builds its recommendation around what the card contains.

A complete product card can answer the specific question a customer is asking. A thin one can't. And the AI, which is trying to give a confident answer, will default toward the brand whose data answers the question over the brand whose data is vague.

I'd run that test I mentioned at the start for a few brands now and the pattern is consistent. The brands winning the product card placement in AI Overviews for competitive fashion queries are not always the biggest names. They're the ones whose feeds are the most complete. That's a structural advantage that's available to any brand willing to do the work, regardless of size or ad budget.

The gap between a thin product card and a complete one almost always comes down to the same set of fields.

Every GMC feed has required fields and optional fields. The required fields get filled because they have to be. The optional fields don't get filled because nothing breaks if they're empty, and the team moving on to the next thing doesn't notice the missed opportunity.

The fields that matter most for AI recommendation quality are the ones that answer the questions customers are actually asking. 

Material composition: not "cotton blend" but the actual breakdown, 320 gsm, 100 percent cotton, preshrunk. Size type and system: whether you're running true to size, whether your large is cut slim or oversized, whether your sizing follows UK, EU, or US conventions. Product highlights: the two or three things about this product that are genuinely worth knowing, written in plain language, not marketing copy. Age group and gender are required fields for apparel products and without them your products are likely to be disapproved before a card is ever generated.

These aren't decorative fields. They're the data points an AI agent uses to decide whether your product matches what someone is looking for, and whether it's worth recommending with confidence. That is the shift from keyword based to context based search. The AI engines need that extra information so they can understand whether that product is the best option for their user.

An AI agent that can't answer a question about your product with confidence won't surface it. That's not a penalty. It's just how the resolution works.

Most fashion brand feeds I look at have maybe 40 to 50 percent of these populated. They are usually inconsistently completed across the catalogue.

Hero products will have more detail than older lines. New season stock gets more attention than previous seasons. The result is a patchy product card record that the AI can't rely on, so it doesn't.

The failure mode worth understanding here is not disapprovals.

Disapprovals are visible and get fixed. The real damage is invisible: products that are approved, running, and generating impressions, but losing AI recommendation placement to competitors with more complete data. You won't see this in your GMC dashboard. You'll see it over time as AI-driven discovery traffic shifts toward brands whose feeds tell a better story.

The practical problem is usually upstream from GMC itself. The feed reflects the quality of the data being pushed into it. If your product information lives in a system that wasn't built to carry material composition, detailed sizing, or product highlights, those fields will always be empty regardless of how many times someone tries to fill them manually. Manual feed management at catalogue scale doesn't hold. The fields get filled once and then slip as new products come in, old products update, and the team's attention moves elsewhere.

The argument isn't about which system you use. It's about whether the system feeding your GMC is structured to carry the full range of product attributes that the AI era now requires. Most weren't built with that in mind, because two years ago it didn't matter. It matters now.

The connection to the Agentic Trust Layer.

This connects to something I've written about across this series, which is the idea of the Agentic Trust Layer. The entity signals, schema markup, and product data that AI agents use to evaluate and recommend your brand aren't separate projects. They're layers of the same infrastructure. Organisation schema tells the AI who you are as a business. GTINs tell it which products are definitively yours. Person schema connects your founders and creative leads to the brand entity. And GMC product data tells it exactly what each product is, in enough detail to recommend it with confidence.

GMC completeness is one layer of that infrastructure, and in some ways it's the most directly commercial one because it operates at the product level. Every incomplete product card is a recommendation that didn't happen. Every thin feed is a product that can't be confidently cited.

I'm still working out exactly how much weight Google places on each individual attribute versus the overall completeness of the feed, and I don't think anyone outside Google knows that precisely. But the directional argument is clear enough to act on. More complete data produces better product cards. Better product cards produce more confident AI recommendations. That's the mechanism.

How to run the test.

The starting point is the same test I'd run with any client: go and look at your products in a Google AI Overview for a category query you should be winning. Note what the product card contains and what it's missing. Then pull up a competitor whose card looks more complete and compare the fields they've populated against yours.

That gap is your GMC roadmap.

The practical method is straightforward. Search Google for a descriptive category query rather than your brand name. Something a customer would actually type: "heavyweight cotton hoodie" or "slim fit merino trousers" or whatever the honest description of your hero product is. Mobile tends to give a cleaner view of how Shopping surfaces are presenting product cards, so worth checking there as well as desktop. Look at the AI Overview if one appears, and look at the Shopping panel. Click through to the full product listing on Google Shopping for your product and a competitor whose card looked more complete. Google shows every attribute it holds on record there including material, size system, product highlights, certifications, reviews. The difference between a well-populated listing and a thin one is immediately visible at that level.

For a systematic read across your whole catalogue, go into Google Merchant Centre directly. The Products tab shows attribute coverage by category, and the Diagnostics section flags missing recommended attributes at scale. This gives you a catalogue-level read rather than product by product, which is useful if you want to understand where the gaps are concentrated. In my experience the gaps tend to cluster in the same places: older lines that were set up before anyone was paying attention to feed quality, carryover stock from previous seasons, and anything that came in through a bulk import that didn't carry optional fields across.

You can't access a competitor's Merchant Centre directly, but the Google Shopping listing is enough for a meaningful comparison. Search their product by name, look at what Google has on record, and note the fields they've populated that yours haven't. That's the actionable version of the gap.

From there it's a question of whether you can fill those fields systematically, at catalogue scale, through whatever system manages your product data, or whether you're looking at a manual effort that won't hold over time. If it's the latter, that's a structural conversation worth having, because patching a feed manually is a short-term fix to a long-term problem.

The brands that treat their product data with the same attention they give their product development will have an increasingly large advantage in AI recommendation surfaces over the next couple of years. The ones that treat GMC as a box-ticking exercise are going to watch that advantage accumulate somewhere else.