GEO for E-commerce
For your brand to survive an assistant's product shortlist, the product data has to be complete.
Ask for "wireless headphones under 500 TL" and the assistant produces a short candidate list from data on the product pages it has crawled. A product whose price, connection type, battery life and compatibility are not written plainly never enters the comparison. The brand is not even eliminated — it was never in the pool.
Here the assistant behaves less like a search engine than a buying adviser: it takes constraints such as budget, compatibility and size, eliminates candidates, names three to five. That elimination runs largely on how complete the product data is. Name recognition does not help; with data missing, a large brand misses the list too.
Our work makes catalogue products eligible for that elimination. We set up Product, Offer and AggregateRating schema correctly, put attributes into plain text, rewrite category pages around the comparison question, keep stock and price consistent with the schema, and measure assistant-sourced traffic as its own channel.
What your customer asks the assistant
What we fix in this sector
Product, Offer and AggregateRating schema
Each product page carries brand, model, sku and where available gtin, with price, currency, stock status and price validity date under offers. Every schema value is kept identical to the visible HTML. AggregateRating is filled only with review data shown on the page.
Putting attributes into plain text
Compatibility, dimensions, material, weight, power draw and warranty are written as a table and as full sentences; assistants read tables but quote sentences. Data loaded by JavaScript after a click on the specifications tab moves into the first HTML the server sends.
Turning the category page into a comparison
The category page gains elimination criteria by use case, price-band headings and a "which one suits you" section; it stops being only a product grid. We decide which filter combinations deserve a permanent, indexable address and close the rest to crawling.
Keeping stock and price current
Stock status and price in the schema are wired to real inventory; a schema left on an old price errors in the validator and costs trust. Sold-out pages are not deleted — stock status is updated and links to alternatives added.
Opening reviews to machines
Reviews are rendered server-side rather than buried in a third-party widget, so crawlers running no JavaScript still see them. Recurring themes — sizing, compatibility, durability — are pulled into a section of the product description. Review text is marked up with Review.
Separating and measuring AI traffic
A separate channel group for assistant referrals is defined in GA4 and those sessions tracked through add-to-cart and purchase. Server logs show which category and product pages each bot pulled. Put side by side, the two show which pages actually pay off.
OnlineStoreStructured data for this sector
OnlineStore describes a business selling online and comes from the Organization family. Site-wide it carries the store name, logo, contact details, returns and shipping policy and customer service channels, giving "is this store trustworthy, how do returns work" a single readable source.
The real work on product pages is done by Product: brand, model, sku, attributes, and price, currency and stock status under offers. Because returns and shipping terms can also be defined inside Offer, delivery and returns questions get answered per product.
Review data attaches to aggregateRating and review only with real reviews visible on the page. Invented or bulk-generated ratings break structured data rules and cost the store lasting trust.
What we measure
- A fixed product question list records which brands the assistants recommend and whether yours make the list.
- The structured data validator is scanned regularly for faulty products and price gaps between schema and HTML.
- In GA4, assistant-sourced sessions, add-to-cart rate and completed orders are reported separately.
- Server logs keep which category and product pages AI crawlers pull, and how often.
- Catalogue coverage is measured: how many products are live with complete attributes and valid schema.
- Valid and invalid items in the Search Console product structured data report are compared monthly.
Frequently asked questions
Supplier text sits identically on hundreds of sites, so it is not a distinguishing source. What you add on top gives it distinction: compatibility notes, use cases, frequent questions, customer feedback. No need to delete it; adding your own knowledge over it is enough.
No. Schema and attribute fields are generated from catalogue data by template, fully automatically. Hand-written content stays limited to the products carrying most of the revenue and to category pages. For the long tail, complete data alone makes a real difference.
The schema field is not fixed text but wired to the same data source that renders the page, so when the price changes the schema value changes with it. The price validity date field is also filled, which stops old data looking current.
Two ways. First, putting a fixed question list to the assistants at intervals and recording the answers — that shows whether the brand is named. Second, separating assistant referrals as their own analytics channel — that measures whether visibility turns into visits and orders.
Only if real reviews are visible on the page. Valid markup is possible with a handful; the number need not be large. A rating with no counterpart on the page breaks structured data rules and risks a manual action, so leaving the field empty until reviews accumulate is safer.
E-commerce Sites — let us look at where you stand
We check how AI assistants answer for your brand today, then work out which of the items on this page are missing on your side.
Get in touch