For an online shop to take part in AI-assisted shopping experiences, it needs a reliable catalogue before a flashy integration. Price, availability, variants, images and terms must match between your website and your product sources. Google is expanding its agentic commerce tools, but several announced features still have limited geographic availability.
What does it mean for an AI agent to help someone buy?
A person might search for "a stringed instrument for a beginner with little space". They do not know the product name, its reference or the exact category. A conversational search or recommendation system can try to interpret needs, compare alternatives and show items that fit those conditions.
For the shop, the challenge is less futuristic than it seems: does it have enough data to answer that question without confusing products? If a product page only contains "premium set", a photograph and an internal reference, the system has little information to explain why that item fits.
So-called agentic commerce covers experiences in which AI systems take part in discovery, comparison and, in some environments, purchase steps. It does not mean any shop can switch on universal autonomous selling tomorrow. Compatibility depends on platforms, markets, permissions, integrations and specific features.
In 2026 Google introduced the Universal Commerce Protocol (UCP) as a standard for connecting commerce systems with assisted shopping experiences. In September it announced extensions to its integration. It is worth telling apart the standard, the announced capabilities and the ones a company can actually use in its own country.
A hypothetical example: a fine-art supplies shop sells brushes of several hardnesses. If the catalogue does not distinguish hair type, recommended technique and size, a recommendation may be of little use. Improving those attributes benefits people who buy on the website, even if they never arrive from an AI assistant.
Source: Google, agentic commerce updates, 16/09/2026. Accessed: 09/10/2026.
Which Merchant Center updates matter in October 2026?
On 16 September Google announced changes related to product visibility in conversational experiences, UCP capabilities and recommendations for improving feeds. The note distinguishes between markets: the new AI performance insights reports were announced as available to businesses in Australia, Canada, India, New Zealand and the United States.
So a shop in Spain, or in any other country not on that list, should not plan its strategy as if that report were available to it. If your shop operates in one of the listed markets, check whether the report appears in your account. The same caution applies to UCP-based direct purchase features and to the Business Agent betas in YouTube ads, whose initial rollout was described for the United States.
What does have practical use is the advice to look after your product feed. Merchant Center lets you share commercial information with Google. That information can include title, description, price, availability, images and other relevant attributes. It is not enough to upload a list once and forget it.
Google also mentioned conversational attributes to add context to certain product listings. Their rollout and acceptance must be checked in the documentation and in the merchant's account. It is not correct to claim that an experimental feature guarantees an assistant will recommend a brand.
The business question is simple: if the way products are presented changes tomorrow, will your catalogue be able to provide consistent data? That preparation is more durable than adapting the whole shop to a screen that may not even exist in your market yet.
Source: Google, Rethink 2026 and Merchant Center, 16/09/2026. Accessed: 09/10/2026.
Why do catalogues fail when stock and variants change?
A shop may have the same item represented in several places: product page, warehouse system, ERP, advertising feed and marketplace. When a size sells out or a price changes, each system needs to reflect it consistently. If one updates later, discrepancies appear.
Google expressly warns about price and availability mismatches between the site and Merchant Center. It recommends updating the product source regularly and explains that automatic updates based on structured data can help correct detected differences. They do not replace reliable synchronisation.
Imagine a small clothing brand with three colours and five sizes of a garment. The shop system sells the last unit of one variant, but the feed still says it is available. A person may reach a product page that no longer lets them buy that combination. The problem is not artificial intelligence: it is data governance.
Descriptions fail too: "Model B17" does not explain what is being sold. Describe verifiable characteristics, without inventing qualities.
For technical products, compatibility deserves special attention. Saying that a part "works with everything" can lead to returns and queries. If there are compatible measurements, materials or models, they must be documented. An agent cannot safely deduce characteristics that the shop has not defined.
Source: Google Search Central, share your product data. Accessed: 09/10/2026.
Which data should a shop sort out before automating?
The first decision is to establish a source of truth for each piece of data. Stock may come from the warehouse; the commercial description from the content manager; the price from the sales platform. If nobody knows which one rules, automation will only spread contradictions faster.
Start with a representative sample of the catalogue and review:
- Identification. A stable reference, an understandable name and the relationship between product and variant.
- Attributes. Materials, dimensions, compatibilities or uses, whenever they are substantiated.
- Price and availability. The value visible on the product page and its synchronisation with the commercial source.
- Images. Photographs faithful to the item, without showing accessories that are not included.
- Delivery and returns. Conditions that correspond to the product and the market.
- Structured data. Product and Offer markup consistent with the visible information, where appropriate.
Google explains that Product structured data can help it understand product pages and qualify for certain rich experiences. It also says that the information in the markup must match what the buyer sees. Do not turn JSON-LD into a second catalogue that nobody updates.
Hypothetical example: an instrument shop sells guitars in different sizes and materials. Grouping them under one generic availability can cause errors; the catalogue must reflect how they are sold.
Tools can check empty fields, compare amounts, detect unexpected variations and warn about inconsistent references. They should not invent technical specifications from a photograph or publish safety claims without review.
Source: Google, structured data for product listings. Accessed: 09/10/2026.
How do you prepare an ecommerce shop without depending on beta features?
Prioritise what you can control today: quality of product pages, synchronisation, speed of correction and clarity of the purchase process. Before hiring a solution presented as "AI commerce", ask them to explain which platform it uses, in which countries it is available and which part of the process it really carries out.
If you work with Shopify, WooCommerce or another platform, check the official connection available with Merchant Center and how it handles inventory changes. Google notes that connected platforms can update information automatically, although changes do not always propagate instantly.
Then create a review routine. You do not need to open every product each morning. You can select references with recent changes, items without essential attributes and listings with issues. A report that says "these variants do not match" has more operational value than a screen full of decorative indicators.
Separate discovery, consideration and purchase. A recommendation in a search engine is not the same as a confirmed order. An integrated checkout experience also needs payment terms, availability and commercial policies that are consistent. Do not design a payment process based on a capability announced for another country.
It is also worth reviewing the shop's language. Product information should serve real customers: explain what is included, what is not, which problem it solves and when it is not the right option. That clarity helps reduce avoidable queries and makes it easier for other systems to interpret the offer.
The goal is not to "please the AI". It is for the business to publish accurate, maintainable information. If new shopping experiences are enabled later, that foundation will make it easier to evaluate their integration without rebuilding the catalogue from scratch.
How have we approached it at meiika?
At meiika we work on websites and custom systems with attention to the relationship between data and operations. For an ecommerce shop, that means starting by understanding where each piece of data is created and who maintains it, before adding another tool.
We do not claim to have our own UCP integration or sales results through shopping agents. We can study the quality of your catalogue and the existing connections, and assess which technical improvements make sense given the business and the features that are really available.