When someone asks Google or Gemini "how much does X cost" or "where to buy Y", the AI doesn't go to your site: it pulls from a specific product catalog. If you haven't configured your listing in that catalog with an up-to-date price, availability and reviews, the AI cites your competitor's price — not yours. You're losing customers who are ready to buy at the exact moment they're searching. You set the configuration up once and then it works for you.
Gemini answers “how much does X cost” using Google Merchant. If you don’t have a well-configured product feed, the price cited in the AI answer is a competitor’s — and the click goes to them.
Let me describe a scene I’ve watched play out dozens of times over the past few months. A business owner opens Gemini and types “how much does a men’s made-in-Italy linen shirt under 120 euros cost”. Back comes an answer with three or four products, price, availability, sometimes the size. They click, they land on the competitor’s site. Their own e-commerce, which sells exactly that shirt, isn’t there. Not because it’s more expensive or worse: it doesn’t have a Google Merchant Center feed configured the way AI shopping needs.
This is where the battle for visibility in AI answers is shifting for anyone selling online. On transactional queries, grounding doesn’t happen on the open web, it happens on the structured product feed. And the feed is no longer a nice-to-have for Shopping Ads: it’s the prerequisite for existing inside Gemini’s answers and inside Google’s AI Overviews shopping modules.
How AI engines handle a price query
Put yourself in the shoes of the model that receives “best Apulian extra virgin olive oil under 25 euros”. Pulling only from the open web, it would find blogs and generic guides, none with an up-to-date price, real availability, or a certified image. The answer would come out vague or wrong on the numbers.
Google has set up a different flow for high-intent transactional queries. The model identifies that it’s a price or purchase query and reroutes grounding to a structured corpus it has known for ten years: Google Merchant Center. The same feed that powers Shopping Ads and the organic Shopping listing now also powers generative answers.
In the public documentation (Merchant Center Help, support.google.com/merchants) the feed is the structured set of product attributes Google uses to show products in Shopping ads and in free organic listings. From this principle I draw an operational consequence: if your product isn’t in the feed, it doesn’t exist in the grounding layer that powers Gemini’s transactional answers. No amount of open-web SEO compensates for absence from the feed on these queries.
The feed attributes that carry the most weight for AI
The AI model prefers structured data for three concrete reasons. Numerical precision: a price inside a `price` field with an explicit currency doesn’t turn into 98 or 890 through a hallucination. Synchronization: the feed is updated hourly, so the availability cited matches the real one. Disambiguation via GTIN, the international barcode, which lets the system know whether your “Oxford Regular Fit Shirt” and the competitor’s are the same product or not.
Inside Merchant Center, not all attributes carry the same weight. Three groups. The first is mandatory: `id`, `title`, `description`, `link`, `image_link`, `price`, `availability`, `condition`, `brand`, `gtin`. If one of these is missing, the product risks not being distributed at all. The second covers the trust signals that increase the probability of citation: `shipping`, `product_highlight` (key features in 100 characters), `product_detail` (section/attribute/value specs), `material`, `color`, `size`. The third comprises external authority signals: product reviews via Product Ratings and seller reviews via Google Customer Reviews. A product with a 4.6 rating across 340 reviews has a much higher chance of entering the AI selection than an identical one with no rating.
The thread is simple: visibility in AI shopping answers is built by filling in these fields well, not by writing longer product descriptions.
Many Italian SMEs, especially tailoring shops and small producers, don’t buy official GS1 codes and leave the field empty or put in random strings.
A concrete case: a clothing e-commerce in Lecce
Let me give you a real example, without a specific brand for confidentiality. A men’s clothing e-commerce in Lecce, mid-range, around 1,200 SKUs across shirts, trousers and knitwear in linen and cotton, produced by Apulian tailoring shops. Target: men aged 35-55 looking for “made in Puglia” pieces between 60 and 180 euros.
The initial problem. Technically clean site, product descriptions in rich prose, Product schema on the page. The Merchant feed, on the other hand, was minimal: only the mandatory fields, GTIN present on a quarter of the items, `availability` updated by hand once a week, zero `product_highlight`, no reviews program connected. On Gemini, for the query “men’s linen shirt made in Italy under 100 euros”, three northern competitors and a generic marketplace showed up. The Lecce shop was absent, even though it had eight shirts that matched perfectly.
The feed intervention over six weeks. Populating the GTIN on all products (where it was missing, buying GS1 codes). Switching to an automatic hourly feed from the ERP. Adding `product_highlight` with three points per piece (composition, tailoring origin, workmanship) plus normalized `material`, `color`, `size`. Connecting Google Customer Reviews and activating Product Ratings with Trustpilot.
The result at three months. On price queries for Apulian linen and men’s shirts under 100 euros, the brand entered the AI Overviews selection in a repeatable way, cited with the correct price and availability. Merchant impressions in Search Console roughly doubled. A single case, not a study: I’ve seen the mechanism replicate on two other e-commerce stores in food and cosmetics with similar interventions.
Populate the GTIN on all products; where it doesn’t exist from the manufacturer, buy it officially through GS1 Italy (gs1it.org).
The test you can run in fifteen minutes
Before working on the feed you need to understand where you stand today. Three steps.
- Open Gemini and run five transactional queries from your sector (“how much does X cost”, “best Y under N euros”, “where to buy Z with free shipping”). Note which brands it cites and whether you show up.
- Open your Google Merchant Center and go to the diagnostics section. Count two numbers: active products and products with errors or warnings. If more than 10% have warnings, you have a data quality problem that the AI perceives.
- Check GTIN coverage: filter for products without a GTIN. If the percentage exceeds 20%, you’re in a heavy penalization zone on AI shopping queries.
A three-way decision threshold for your feed: 0-5% warnings and over 90% GTIN coverage means you’re competitive; between 5 and 20% warnings or GTIN between 70 and 90% puts you in the gray zone; above these thresholds the feed is your real bottleneck, before any other action.
As always, this is an entry-level check. Serious analysis — correlation between populated attributes and AI citation rate, monitoring across multiple engines, GTIN matching between competitors — requires professional shopping intelligence tools.
The mistakes I see most often on Merchant feeds
Four recurring patterns in the feeds of Italian e-commerce stores that don’t make it into AI answers.
Static feeds updated by hand. A CSV uploaded once a month. Real availability changes every hour, the feed doesn’t. When the AI finds discrepancies between the feed and the live site, it penalizes the source. Better an automatic feed via ERP or a native connector (Shopify, WooCommerce, Magento all have one).
Missing or invented GTIN. Many Italian SMEs, especially tailoring shops and small producers, don’t buy official GS1 codes and leave the field empty or put in random strings. Without a GTIN the system can’t disambiguate your product from the competitor’s and favors whoever has the clean identifier.
Product description as a marketing piece. “Discover the timeless elegance of our spring collection.” From a description like that the AI model extracts nothing useful. Composition, country of production, washing, fit, workmanship details work much better.
No connected reviews. Without a Product Ratings feed your product competes on price alone. With aggregated reviews of 4+ and at least a few hundred reviews, the probability of AI citation grows noticeably.
What to actually do on the feed
In order of operational priority.
- Connect the feed to your ERP or platform with automatic updates at least hourly on price and availability; eliminate manual CSVs.
- Populate the GTIN on all products; where it doesn’t exist from the manufacturer, buy it officially through GS1 Italy (gs1it.org).
- Rewrite `title` and `description` in a data-first key: brand + type + distinctive feature in the title; composition, origin, materials, measurements in the description. No emotional prose in the structured fields.
- Fill in `product_highlight` and `product_detail` with normalized technical specs: these are the fields the AI reads to justify why it should cite your product.
- Activate Product Ratings by connecting Trustpilot, Feedaty or Recensioni Verificate. Goal: at least 50 reviews per main SKU with an average rating above 4.3.
- Compare your feed with the three to five competitors that Gemini cites today on your business queries. Look at the attributes visible on the Shopping listing, see where they’re richer, close the gap.
Merchant Center on its own is not a magic factor. I’ll say it without FUD: if your brand has no web authority and the product pages on your site are thin, a well-configured feed helps you but won’t take you to first place. It’s one of the levers — on transactional queries it’s the one that carries the most weight.
Where it fits in the AI visibility picture
In previous articles I explained how AI engines work on informational queries, how much backlinks as a citation proxy matter and why E-E-A-T for AI remains decisive in getting you recognized as a reliable source. Merchant Center closes the loop on transactional queries: it’s the channel where structured data replaces the editorial as the primary signal.
In the coming articles I’ll get into other pieces of the same puzzle: how Gemini integrates the Knowledge Graph on brand searches, how AI Overviews choose snippets among competing sources, how the Perspectives module behaves on discussion queries. The thread stays the same: in 2026 your visibility in AI answers comes down to the overall quality of the signals — editorial, authority, and, on price queries, structured feed.
If you sell online and you haven’t yet touched Merchant Center with AI in mind, today is the right day to open it.
