How to optimize your product pages so AI recommends them

A product page AI can recommend has the data people ask about written in its HTML: specific attributes with their units (weight, dimensions, material, what it's for), with the same values across the category, a title that says what it is and who it's for, the GTIN and product structured data. Nice prose helps little. Assistants compare data, and if a detail is missing, your product doesn't make the comparison.

What does AI look at on a product page?

When someone asks an assistant for "lightweight waterproof trail running shoes for women", the assistant looks for products that match every part of the sentence. Your page only makes it if it says, in those words or with an equivalent detail, that the shoe is for trail, that it's waterproof, that it's a women's model and how much it weighs.

For that it needs three things from the page:

• To be able to read it. Text and data have to be in the HTML, not loaded only with JavaScript or inside an image. AI crawlers don't execute JavaScript.
• To find the detail. Every attribute written out, with its unit, on every product in the category.
• To know which product it is. Brand, model and GTIN, which help link your page to what other sites say about that product.

What does a product page AI can recommend look like?

Here's an example with a made-up product, the Trail Runner 2 GTX shoe. The left column is the typical page that arrives from the supplier. The right one has the same plus the details people ask about.

The "before" page isn't bad: it has a photo, a price and decent copy. But for the question in the example it only makes "waterproof" clear. The "after" page answers all four parts, and the follow-up questions too: "and with a low drop?", "which one is lightest?".

Which attributes should every product page have?

The ones people use to decide in that category. They vary a lot from one category to another, and that's where it shows who knows their product.

Some attributes hide nuances that machine translation or a writer without context will miss. In Spanish wine, "Crianza", "Reserva" and "Gran Reserva" aren't degrees of "aged": they're legal categories with a minimum ageing time, part of it in barrel and the rest in bottle. Getting them right is what makes the page answer "a Rioja Reserva to give as a gift".

Why do you need to normalize values?

Because a detail written five different ways works like five different details. If half your shoes say "245g", others "0.245 kg" and others "Weight: light", neither a filter nor an assistant can compare them.

The practical way to get there is to define, for each attribute, a closed list of allowed values, and have every product page pick from that list. "Water-resistant" and "waterproof" aren't the same thing, and the list forces you to decide which one each product is.

What structured data does a product page need?

As well as being written on the page, the key data should go in schema.org Product markup: name, brand, GTIN and an offer (Offer) with price and availability. It's how you tell any machine without it having to interpret your design. A minimal example:

{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Trail Runner 2 GTX Women's",
  "brand": { "@type": "Brand", "name": "Altamonte" },
  "gtin13": "2012345678903",
  "offers": {
    "@type": "Offer",
    "price": "139.90",
    "priceCurrency": "EUR",
    "availability": "https://schema.org/InStock"
  }
}

Check the GTIN above all: it helps link your product to the reviews and comparisons that talk about it on other sites.

Does the description matter, or only the attributes?

Both matter, for different things. Attributes let you compare: weight with weight, drop with drop. The description explains what the product is for and who it's for, which is exactly what people ask in natural language: "to start running in the mountains", "for wide feet".

What doesn't help is adjectives. "Exceptional comfort" doesn't answer any question, and you can't filter on an adjective. A good description turns attributes into sentences: what it has, what terrain it's for, what kind of runner.

How do you do it across thousands of product pages?

Not one page at a time. In a catalog of several thousand products, filling ten attributes per product by hand never ends. What works:

1. Start with the categories where you lose questions, not in alphabetical order.
2. Define each attribute once, with its allowed values, and apply it to the whole category.
3. Fill only empty fields, without touching content that's already fine.
4. Review before publishing. An invented value is worse than an empty field.
5. Translate the attributes too if you sell in more than one language.

That's what the Semantico catalog builder does: from the EAN, SKU or name, it finds reliable sources for the product, places it in your category tree, fills the attributes you define across the category with your allowed values, and translates it. Your team reviews it before publishing. Vinoselección went from 20–30 minutes per wine to under 30 seconds, with its product pages translated into five languages.

Where do I start this week?

1. Pick your most important category and write down the ten buying questions people ask most.
2. Pull out of those questions the data they ask for. That's your list of attributes.
3. Export the category's product pages and count how many have each attribute, and in how many different ways it's written.
4. Open a product page, view the source and check that the attributes and the GTIN are in the HTML.
5. Fix first the attribute that's asked about most and that you have least.

Common questions

Keep reading

• How to get ChatGPT to recommend your online store's products Product pages are one part. Here's the rest: crawlers, buying guides and review sites.
• Free AI visibility audit for online stores Which questions you lose and whether product data is the cause, in your inbox within a couple of hours.
• GEO for ecommerce: a company to get your store into ChatGPT If you'd rather we measure and fix it: what Semantico does and what it costs.
• GTIN Validator Check for free whether your EAN and UPC codes are valid.
• The Semantico catalog builder How we fill attributes, descriptions and translations across a whole category.