E-commerce
August 26, 2026
You spend hours refining your product pages: descriptions, high-resolution photos, customer reviews. Yet, when an AI agent looks for a 15-inch laptop bag under €60 deliverable within three days, it will never see that beautiful page. It only consults structured fields, raw data feeds where each attribute carries a precise label and type. No typed price? No quantified dimension? Your item disappears from the comparison before its quality is even judged. Since Google and Adyen launched their agentic commerce tools in 2026, this requirement has moved from a best practice to a strict prerequisite. This article explains what an agent actually reads, what the four essential data classes are, and how to structure your feeds to stay in the game.
Summary
What an AI agent actually reads instead of your product page
An AI agent never browses your shop like a human customer would. It does not load JavaScript, scan layouts, or interpret hidden intent within a slogan. It consumes structured records, rows of data where every fact has a clear label and a known type: a number declared as a price, a currency coded in ISO 4217, a date in ISO 8601 format, a state chosen from a fixed list of allowed options.
This data arrives through three channels. The first is the product feed in CSV, TSV, or XML format, submitted to a consumer or hosted on an updated URL. The second is structured data within the page itself, tagged in JSON-LD, microdata, or RDFa according to the schema.org vocabulary. The third is a catalog API that responds to a query with records rather than an HTML document.
The result: a single query returns values that are already labeled and typed. The agent then applies constraint rules—budget constraints, size, delivery times, condition. It does not form any subjective impression when reading your description: it simply checks whether a value meets a rule. No value? No test possible, leading to immediate elimination.
Convert over 2,000 customers on average per month with Qstomy.
The world’s 1st Shopify AI dedicated to customer conversion
Empowering 200+ e-commerce merchants
The four classes of fields that determine your visibility
Four categories of data determine whether your product can enter the comparison. Missing just one of them does not place it at the bottom of the list: it simply removes it from the running entirely.
Stable identifier (GTIN, EAN barcode, manufacturer MPN, internal SKU): the agent recognizes the same item across multiple merchants and deduplicates occurrences. Without an identifier, a product appears as several distinct references or does not match any known item.
Price and currency (price and priceCurrency fields): the agent tests the budget constraint. A price displayed only as rendered text, such as "from €39", provides no unique value to compare. The item is discarded.
Availability status (availability field, enumerated value like InStock or OutOfStock): the agent filters what is immediately purchasable. Without this field, available items and stockouts mix until failure at checkout.
Variant structure (item_group_id plus declension axes like size or color): the agent separates a variant from a different product. Without a clear structure, the wrong size is selected or all variants blend together.
A GTIN administered under GS1 rules resolves globally, allowing the agent to compare three offers for the same object. An internal SKU deduplicates nothing. This logic extends to technical specification fields: a vague adjective like "compact" satisfies no quantitative dimension constraint. To learn more about the key data to track, consult our guide E-commerce analytics: what to track and why?
Why a page that is full for a human remains empty for a machine
Your product sheet displays all the necessary information. Yet, a parser sees only a fraction of it. The first reason: images. Dimensions printed on a diagram, a compatibility grid in the form of tables, care instructions photographed on a label: all of this is invisible to a script that reads raw HTML. Alt text written for accessibility rarely repeats key figures.
Next come contents hidden behind an interaction. Specification tabs, accordions, customer review panels arrive via an asynchronous request triggered after the initial load. They exist in the DOM only after the browser has executed the page's JavaScript. A parser that does not render the page receives an empty shell; no drawer is opened.
Finally, free-text specifications fail in another way. A phrase like "fits most standard frames" makes sense to a buyer visualizing their need, but it carries neither a numerical value nor a unit of measurement. The agent cannot test this assertion against a width constraint expressed in centimeters. If you are looking to improve your organic search results and the machine-readability of your product sheets, our article SEO for E-commerce Sites: Strategies That Really Work will help you structure your content.
How an agent tests a price or size constraint
The agent assembles a set of constraints from the user request: budget ceiling, minimum or maximum size, delivery time, brand new or refurbished condition. It then tests each attribute of each item against each constraint. This is not an overall judgment; it is a series of binary checks: does the value comply with the rule, yes or no?
Take the price, for example. If your feed contains a price field with the value 45.00 and a priceCurrency field with EUR, the agent can check if €45.00 is less than €60. The comparison works because both numbers share the same type and unit. On the other hand, if the price only appears in a text description, the agent cannot extract any unique value. The item is discarded before any qualitative evaluation.
The same logic applies to dimensions. A "15-inch laptop bag" constraint requires a width, height, or diagonal expressed in a number and a unit. An additionalProperty field containing "width: 38 cm" directly satisfies the rule. The adjective "spacious" satisfies nothing. For this reason, technical specifications must be carried by typed fields rather than drowned in description paragraphs.
The three channels through which your product data travels
Your product information flows to agents via three distinct routes, each with its own conventions and limitations. The product feed remains the historical channel: you submit a CSV, TSV, or XML file to a platform that ingests it, or you host this feed on an accessible URL according to a fixed schedule. The format often follows conventions established by RSS 2.0 or Atom 1.0 for XML, or a simple tabular schema for delimited files.
The structured data on the page marks the same facts directly within the HTML, thanks to JSON-LD, microdata, or RDFa. You declare a Product type and an Offer type according to the schema.org vocabulary, and you fill in the name, price, priceCurrency, availability, and image properties. A parser loading the page retrieves these tags without executing any script.
Finally, a catalog API responds to an HTTP request with a set of structured records, usually in JSON format. This is the most recent path, favored by agentic commerce tools to query your inventory in real time. If your Shopify platform exposes a compatible product API, the agent can directly query your stock without going through a scheduled export. To understand how Google Shopping leverages these feeds, read our guide What is Google Shopping for e-commerce?
Why GTIN and MPN identifiers are essential
A stable identifier allows the agent to recognize that it is the same physical product, regardless of the merchant offering it. The GTIN, administered under GS1 rules and often represented by an EAN or UPC barcode, resolves globally. An agent reading three offers with the same GTIN knows it is comparing three prices for an identical item. It can then rank the offers by price, delivery time, or shipping costs.
Without a GTIN or manufacturer MPN, the agent treats each occurrence as a distinct product. Three merchants selling the same headphones? The agent sees three unrelated references. It is impossible to detect the best offer, impossible to deduplicate the results. The end customer receives a confusing list, and you lose the opportunity to stand out through price or service.
The internal SKU does not replace the GTIN: it only circulates within your systems, resolves nowhere else, and deduplicates nothing. Use it to link your catalog to your inventory management, but always add a GTIN or MPN in your public feeds. If you manage multiple sales channels, our article What is the real purpose of an order management system in e-commerce? explains how to orchestrate these identifiers at scale.
How to structure variants to avoid confusion
A single product available in several sizes, colors, or finishes must be clearly represented in your feeds. The agent needs to know that these variants belong to a single logical group; otherwise, it treats them as completely different products. The item_group_id field groups all variations under a single parent identifier.
Next, each variant carries axis fields: size for size, color for color, material for material. These fields allow the agent to distinguish a "size M, blue" variant from a "size L, red" variant. If these axes are missing or mixed up in the description, the agent cannot distinguish the correct reference. It selects the first row it encounters, regardless of the requested size.
Even worse, without a variant structure, multiple variations can collapse into a single ambiguous row. Your catalog then displays an unclear price, contradictory availability, and no match with the user's query. Structuring variants correctly from the start spares you these errors and improves the conversion rate, as the agent sends the customer to the correct reference on the first try.
Price and currency: why "starting from" always fails
An AI agent can only test a budget constraint if the price is provided as a single number, accompanied by a standardized currency code. The price field contains 45.00, the priceCurrency field contains EUR: the comparison with a ceiling of €60 works immediately. However, a mention "from €39" or "starting at €39" does not provide any usable value.
The agent does not know whether the final price will be €39, €45, or €60. It cannot resolve the ambiguity, so it eliminates the item from the list of candidates. The same problem occurs if the price only appears on the rendered page, calculated by JavaScript based on chosen options: the parser that does not render the page sees nothing.
Ensure that every row in your feed carries a precise price for the relevant variant. If you sell configurable products with paid options, create one row per price combination or indicate the base price in the feed and the surcharges in dedicated fields. The key is to provide a testable value against a numerical constraint. To optimize your advertising campaigns and ensure your prices are properly distributed, consult our article E-commerce marketing and advertising: what are the differences?
Availability status: filter what is purchasable now
The availability field indicates whether the item can be purchased immediately. The allowed values are taken from the schema.org ItemAvailability enumeration: InStock, OutOfStock, PreOrder, Discontinued, LimitedAvailability. An agent receiving InStock knows that the item is available, so they can offer it in response to an urgent request.
Without this field, the agent cannot distinguish a product in stock from a temporary out-of-stock situation or a pre-order item. All products end up mixed together in the results, and the customer only discovers the shortage at the moment of validating their cart. The abandonment rate skyrockets, and your reputation takes a hit.
Update the availability status as often as your inventory changes. If you use an order management system connected to your feed, automate this synchronization so that the availability field reflects the actual status in real time. This ensures that the agent only suggests items that are actually available, which improves the customer experience and reduces disappointments at checkout.
Technical specifications: moving from free text to typed fields
Product descriptions written in prose are valuable for convincing a human, but they remain opaque to an agent. A phrase like "generous dimensions for all your documents" carries neither width, height, nor unit of measure. The agent looking for a 38 cm wide bag cannot verify this constraint, so the item is discarded.
The solution is to extract each technical fact and place it in a dedicated field. Use the schema.org PropertyValue type or the additionalProperty field to declare width, height, depth, weight with their numerical value and unit. For example: "width: 38, unitCode: CMT" for 38 centimeters. The agent reads these properties directly and tests them against the constraints of the query.
The same logic applies to compatibility, power, capacity, resolution, or any other measurable attribute. If you sell electronics, indicate the screen diagonal in inches, the battery capacity in milliampere-hours, the weight in grams. If you sell textiles, specify the composition in percentages, the dimensions of each size in centimeters. The more typed your specifications are, the more the agent can refine its recommendations. To understand how this data influences your organic visibility, read our guide What is e-commerce SEO and how does it work?
Qstomy and product feeds: from structured data to real-time recommendation
Qstomy is a Shopify AI agent that accompanies your customers from product discovery to parcel tracking and after-sales service. Unlike traditional feed tools that simply export your data to advertising platforms, Qstomy consumes your structured data in real time to guide each visitor towards a purchase. It relies on the same typed fields, GTIN identifiers, price with currency, availability, and variants, to recommend the right product at the right time.
The result: the agent instantly answers availability questions, offers alternatives when an item is out of stock, suggests relevant complementary products, and supports the customer all the way to cart validation. It also manages package tracking, return requests, and questions about your commercial policies, all without leaving the conversation. You reduce the abandonment rate, increase the average cart value, and free up your support team from repetitive tasks.
More than 100 merchants already trust Qstomy to convert their visitors into buyers. By correctly structuring your product feeds, you maximize the agent's ability to compare, recommend, and convert, whether on your store or on the external channels where your products are distributed.
Checklist, in brief, and frequently asked questions
AI Agent-Ready Product Feed Checklist
Stable Identifier: GTIN or MPN filled in for each product
Price and Currency: Typed price and priceCurrency fields, a single value per variant
Availability: Availability field with enumerated value (InStock, OutOfStock…)
Variant Structure: item_group_id and variation axes (size, color) clearly separated
Technical Specifications: Dimensions, weight, capacity carried in typed fields with units
In short
AI agents never read your rendered product pages: they compare typed fields in your feeds and structured data. Four data classes are mandatory to remain visible: stable identifier, price with currency, availability status, and variant structure. Any technical fact locked inside an image, a hidden tab, or a prose description remains invisible. Structure your feeds now to meet the requirements of agentic commerce and guarantee that your products enter the comparison.
Frequently Asked Questions
Can an AI agent read dimensions in a product image?
No. Dimensions printed on a diagram or photo are invisible to a parser. You must declare them in typed fields, for example, additionalProperty with numerical value and unit.
Why is my product disappearing from results when my product page is complete?
If the price, availability, or identifier are missing from your structured feed, the agent eliminates the item before any comparison. A visually complete page does not guarantee the presence of usable typed fields.
Is a GTIN required for all products, even artisanal or handmade ones?
Products without a universal barcode can use a manufacturer MPN or an internal SKU, but they will not benefit from cross-merchant deduplication. If you have access to a GTIN, always add it.
What is the
To go further: How to help a customer blocked by 3D Secure during payment - Qstomy.
Enzo
August 26, 2026
