An AI shopping agent is a system that searches for and compares products on the user’s behalf, shortening the path from question to choice. Instead of a long list of stores, the customer gets a narrower comparison — which raises the value of well-described product data: price, availability, brand, and identifier. For most stores, though, this is still preparation rather than a meaningful source of orders.
What’s real and what’s hype
There’s a lot of noise around agentic shopping, so let’s separate out what can actually be documented.
What’s documented is that it’s being tried. With ChatGPT, there are publicly observable scenarios where it helps compare and choose products; that doesn’t mean a broadly available, established route to purchase. With Perplexity, tests of shopping-oriented output and paths to a product have been publicly visible. Google’s AI experiences surface shopping-oriented answers and product information too — which is why it still makes sense for merchants to work on product data and Merchant Center. There are also concepts and partial attempts at consolidating the purchase steps inside assistants, but none of it has settled into a stable shape.
What can’t be documented: how much sales volume runs through agents, exactly how agents pick and rank products, or that this is an established channel anywhere. The scope and availability of these features also vary by market and over time, and operators don’t publish how the selection works.
From a list to a shortlist
The most significant change isn’t technical, it’s logical.
| Aspekt | Angle | Difference |
|---|---|---|
| Output | List vs. shortlist | Instead of ten links, the customer gets a handful of options |
| Goal | Ranking vs. inclusion | Sixth place was still a chance; missing the shortlist means far less of one |
| Reader | Human vs. machine | An agent doesn't judge a site the way a person does; it leans on machine-readable values |
| Basis | Copy vs. fields | Specific values decide: price, availability, brand, identifier |
| Maturity | Still early | Shopping features are rolling out unevenly by market — this is preparation |
This connects to what the piece on zero-click describes: the more of the decision that happens inside the answer, the smaller the role your site’s appearance plays — and the bigger the role of what a machine can determine about the product.
Which data actually matters
There’s no secret here. These are the fields that let a product be identified and compared unambiguously.
- Name and brand An unambiguous product name and brand — not an internal code or a marketing slogan.
- Identifier (GTIN, MPN) It's what makes it possible to tell this is the same product your competitors sell; brand only sharpens the identification.
- Price and currency A specific amount and currency, not just text on the page.
- Availability In stock, on backorder, sold out — and current above all.
- Image and description A visual and a clear description of what the product is and who it's for.
- Ratings and terms Real ratings, shipping, and returns. Never invented.
In structured data, that mainly means watching the Product and Offer types: on the product, name, brand, and gtin, plus mpn where it applies; on the offer, price, priceCurrency, and availability. Optionally aggregateRating and shipping details via Offer.shippingDetails. How to implement them is covered by product pages for AI and more broadly by structured data for AI. On WooCommerce or Shopify, this is usually handled by the theme or an app.
Your feed and your page have to match
Structured data on the page and a product feed aren’t alternatives — they describe the same product, just for a different consumer. The problem starts when they drift apart.
A price in the feed that differs from the page isn’t a detail: for the customer it’s a letdown, and for any machine it’s a signal that your values can’t be trusted. The same goes for availability.
The most common mistakes
Price only in the copy
Price and availability visible on the page but missing from structured data — or the other way around. The machine then reads something different than the customer does.
Missing identifier
Without a GTIN, or an MPN where it applies, it’s harder to tell that this is the same product your competitors sell; brand only sharpens the identification.
Feed and page drift apart
A different price or availability in the feed than on the site. The mismatch hurts trust and machine processing alike.
Rebuilding the store over hype
Agentic shopping doesn’t yet drive a meaningful number of orders. Investing in clean data makes sense; an expensive rebuild based on speculation doesn’t.
Key takeaways
- The output changes, not just the ranking — a shortlist instead of a list of stores, and outside it your product isn’t visible.
- Data decides, not impressions — name, brand, identifier, price, availability.
- Feed and page have to match — a mismatch hurts customers and machines alike.
- How agents select isn’t public — anyone claiming to know exactly how they pick is speculating.
- This is still preparation — do the things that make sense without agents; for how to measure it, see the AI report in Search Console.
Not sure whether your product data holds up? An AI SEO audit goes through the structured data on your product pages, checks that the feed and the page agree, and finds the fields that are missing. If you’re building a new store, get the product and category templates right up front — with a structure that both people and machines read well.
For transparency: this is based on publicly available information about shopping features in AI systems as of July 11, 2026. How products are actually selected isn’t published, availability of these features varies by market, and the space is moving fast — so this article doesn’t explain how to “win” agents, and it doesn’t guarantee results.
