How to Structure Product Data for AI Shopping Agents

When a shopper asks ChatGPT, Google's AI Mode or Copilot which running jacket to buy, nothing on your homepage gets read. The agent reads your product data: the category, the attributes, the price, the stock status, the GTIN. If that data is thin or inconsistent, the agent either skips you or describes you wrongly. Most Shopify catalogues were built for a human scrolling a collection page, not for a parser, and it shows.

This is the practical piece on structuring product data for AI shopping agents on Shopify: what to change, in what order, and which parts of the agentic commerce story are still speculative. Some of it is. The data hygiene underneath it is not.

Why structured product data matters now (and why it is not a growth hack)

The traffic is real, even if the checkout standards are not settled. Adobe Analytics reported that AI referral traffic to US retail sites was up 62% year on year in July 2026, and up more than 1,200% since October 2024. Adobe has also reported those visitors converting at a higher rate than non-AI traffic for most of the past year. UK data lags the US, but the direction is the same. You can check your own store: Shopify's sessions by referrer report will show ChatGPT and Perplexity if they are already sending you visits.

Be clear about what you are buying with this work, though. Structured product data will not double your revenue next quarter. It makes your catalogue legible to every machine that reads it: Google Merchant Center, Meta's catalogue, Shopify's own search and recommendations, and now the AI channels. It is foundations work. If your conversion rate is the real bottleneck (and for most stores with steady traffic, it is), conversion rate optimisation comes first. For the search side of the same problem, read our guide to GEO for Shopify. This article is about the catalogue itself.

Start with Shopify's product taxonomy and category metafields

Shopify's Standard Product Taxonomy is an open source classification of categories, attributes and values, published on GitHub and updated several times a year. The February 2026 release alone added over 700 categories and 600 attributes. Every product in your admin should have a category from it, at the most specific level that fits. "Apparel & Accessories" on its own is nearly useless. "Apparel & Accessories > Clothing > Clothing Tops > Shirts" is what unlocks the rest.

Once a category is set, Shopify surfaces category metafields for that product: size, fabric, sleeve length, colour, age group and so on, depending on the category. These use standardised values, which is the point. "Navy", "navy blue" and "Midnight" are three different things to a parser. Fill the category metafields, connect them to your variant options where Shopify allows it, and you get consistent attributes that flow to Google, Meta and the agentic channels without a separate mapping exercise for each.

On a large catalogue this is a bulk job, not a one-by-one job. Export, categorise in a spreadsheet, review the automatic suggestions Shopify makes, and import. Budget a few days for a few thousand SKUs, and expect to find products that were never set up properly in the first place.

Get the variant data complete and consistent

Agents match on identifiers. A product without a GTIN (the barcode field in Shopify) and a brand cannot be reliably matched against the same item sold elsewhere, and Shopify's guidance for the Catalog that feeds its agentic storefronts leans heavily on both being present. Missing barcodes are also one of the most common reasons products get limited in Google Merchant Center, so this is not new advice. It is old advice that now has a second consequence.

Go through every variant and check:

• GTIN or barcode on every variant, or the correct no-GTIN handling for genuinely custom or handmade items.
• Brand and vendor set, and spelled one way across the whole catalogue.
• Price and compare-at price accurate, with inventory tracking on so availability is real.
• Option names and values consistent (Size and Colour, not Size on one product and Sizes on another).
• Weight and dimensions filled in, because shipping estimates depend on them.

Product JSON-LD: the structured data fields agents actually read

Structured data on the product page is how a crawler confirms what your feed says. Most Shopify themes ship with some Product schema, but "some" is the problem. Check what yours outputs with Google's Rich Results Test and compare it against Google's Product structured data documentation. The Offer should carry price, currency, availability, item condition and the GTIN. Google also asks for shipping details and a merchant return policy, either on the Offer or through an Organization-level return policy, and Merchant Center settings take precedence where both exist.

For UK stores, make sure the currency is GBP, the shipping destinations reflect where you actually deliver, and the returns window matches your policy page. Agents cross-check. A mismatch between the schema, the feed and the visible page is worse than a missing field, because it reads as unreliable. This usually needs a theme change rather than another app, and it is the kind of fix we make inside Shopify theme design and development work rather than bolting on a third script that fights the first two.

Put the deciding attributes in metafields, not in prose

Think about what actually decides a purchase in your category. For homewares it is dimensions and materials. For supplements it is ingredients and dose. For accessories it is compatibility. For clothing it is fit and sizing. If those facts live only in a paragraph, or worse, in an image of a size chart, a parser has to guess. Put each one in its own metafield with a proper type (dimension, weight, list of single-line text, and so on), show it on the product page as a spec table, and reference it in the description.

This is also where the overlap with human conversion is strongest. A clear spec table answers the question a shopper had before they hit the back button, and it gives the agent a clean fact to quote. Our guide to Shopify product page best practices that convert covers the page layout side.

Titles and descriptions that work for a person and a parser

Titles should state what the product is, its defining attribute and the variant in a predictable order: brand, product type, key feature, then size or colour. Avoid titles that are only a name ("The Fraser") with the product type buried three paragraphs down. Descriptions should open with a plain sentence that says what the thing is, who it is for and what makes it different, then get into detail. Keep your brand voice, but keep the facts in text, not in graphics.

One test: read your description aloud without the images. If you could not tell what the product is, what it is made of, what sizes it comes in and what it is compatible with, neither can an agent.

Agentic commerce on Shopify: what is settled and what is not

The standards around agent-led checkout are moving fast. OpenAI's Instant Checkout in ChatGPT launched to US users in early 2026 and was reported to have been pulled back within weeks, with only a few dozen merchants live and problems with scraped pricing and stock data. Google's Universal Commerce Protocol and OpenAI's Agentic Commerce Protocol are both still evolving. Shopify's agentic storefronts, managed under Sales channels > Agentic in the admin, currently cover ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Meta, but the terms and behaviour of those channels will keep changing.

Notice what the reported failure was, though: bad product data. Whichever protocol wins, the agent still needs an accurate category, a matching identifier, a real price and honest stock. That part is not speculative. It is the same data Google Merchant Center and Meta have wanted for years, so the work pays back on channels you already run, even if agentic checkout arrives more slowly than the press releases suggest.

Whichever protocol wins, the agent still needs an accurate category, a matching identifier, a real price and honest stock. Fix that and you are ready for all of them.

A sensible order of work

1. Assign a specific taxonomy category to every product and fill the category metafields.
2. Audit variants for GTIN, brand, price, availability, option names and weights.
3. Fix the theme's Product JSON-LD and add shipping and returns detail.
4. Move purchase-deciding attributes into typed metafields and surface them on the page.
5. Rewrite titles and the first sentence of each description for clarity.
6. Turn on Shopify Catalog and the agentic channels, then check the diagnostics for rejected products.

Done properly, this is a few weeks of focused work for a mid-sized catalogue, and most of it is data rather than code. If you want it done alongside the conversion work that usually matters more, it fits naturally into our Conversion Growth Retainer, where product data, page structure and testing are handled by the same senior team from our Glasgow studio. Or get in touch if you just want an honest read on how machine-readable your catalogue is today.