Chapter 15: Commercial Data and Shopping Agents

Last updated: November 1, 2026.

In Chapter 1 I wrote that for purchase queries, users still have to come to your website, and I added “at least for now.” This chapter is about how long “for now” is likely to last. As of November 2026, buying from inside an AI assistant exists in the US and is spreading to a few other countries, but it isn’t available in Turkey yet. The details are messier, because a lot was announced in this area over the past year and some of it has already been pulled back.

It gets simpler if you split the topic into two layers. The first is commercial data: price, stock, shipping, returns, variants and product attributes. AI uses this data when it recommends a product, and this layer already works in Turkey. The second is checkout: AI adding items to a cart and completing payment on the user’s behalf, plus the commerce protocols that make that possible. This layer is moving fast but hasn’t reached Turkey.

Almost everything an e-commerce site should be doing today sits in the first layer. For the second, keeping an eye on it is enough; it makes more sense to plan around it once it actually arrives in Turkey.

1. Two routes for product information

In Chapter 3 we saw that Google’s AI Mode draws on the Shopping Graph, its product graph. Back then, Robby Stein said the catalog held about 50 billion products and was refreshed about 2 billion times an hour. A Google post from October 2026 describes it as more than 60 billion constantly refreshed product listings (Google).

Much of this data comes from what merchants submit themselves. According to Google’s AI guide, generative AI answers can include product listings, product information and local business information where relevant, and Merchant Center feeds and Google Business Profile can help your products and services appear in AI answers and other search results (Google Search Central).

ChatGPT works a little differently. For the shopping research feature it announced in November 2025, OpenAI says results are organic and based on publicly available retail sites, and it describes an application process for merchants who want to be included (OpenAI). Since March 2026, merchants can also share product feeds and promotions through the Agentic Commerce Protocol (OpenAI). I’ll come back to that protocol below.

So AI learns about your products in two ways. One is your website: your product pages get crawled, which is where the crawling and indexing foundation from Chapter 2 and the readable page from Chapter 10 come in. The other is the product feed you send to systems like Merchant Center. That route is faster and more structured.

The two routes need to say the same thing. A product listed at 4,990 TL in the feed and 5,490 TL on the page runs into Google’s rules, and on top of that it gives AI two different prices.

2. Old prices and sold-out stock

In Chapter 3 we saw that AI can state an old price, a closed branch or a discontinued product with the same confidence as anything else. In commerce, that mistake costs money directly.

A hypothetical scenario: a sneaker retailer sells out of a model. The product page says “out of stock,” but the feed updates a day late. That day, AI recommends the shoe as “on sale and in stock right now.” The user comes to the site, can’t find it and leaves. There’s no sale, and the user trusts the brand a little less.

That’s why Google’s Merchant Center rules are strict. For price, it requires the following (Google Merchant Center Help):

• The price and currency in the feed must match the price on the landing page and at checkout.
• The price must be shown prominently on the page, in the target country’s currency.
• Practices like “see price in cart” aren’t accepted.
• The price on the product page must not change based on the user’s location.

Products that break these rules get disapproved. But the hard part isn’t the rules, it’s speed. The more often price and stock change, the faster your site and feed need to update.

Here’s a related example from our own work. On a news site, the sitemap was cached for about 10 minutes; we brought that down to seconds. Product data works the same way. If a price change reaches the feed, the sitemap or the structured data late, AI gets the wrong information in the meantime.

Google has a safety net here. Automatic item updates in Merchant Center can use the structured data on your site to update price, sale price, availability and condition (Google Merchant Center Help). In Google’s own example, a product listed at $4 in the feed and $3 on the product page gets updated to $3 in ads and free listings. For this to work, your product pages need to provide price and availability as structured data. In Chapter 2 I called structured data “not required, but useful”; in commerce, the use is very concrete.

3. The Merchant Center feed

For e-commerce sites in Turkey, the first step is still Google Merchant Center. Turkey is on Google’s list of supported countries, products can be shown in Turkish lira and the Shopping tab is available to users in Turkey (Google Merchant Center Help).

The feed you send to Merchant Center is the most orderly way for AI to understand your products. In a good feed, the title is descriptive: brand, model, product type and the key attribute. “X Brand Men’s Volleyball Shoe, High-Top, White” rather than “Shoe 4521”. The description is complete and covers what the product does, who it’s for, its materials and dimensions. GTIN (barcode), brand and manufacturer part number are filled in correctly, which makes it easier to match the same product across sellers. Variants such as color and size are defined within the same item group. Delivery time, shipping cost and return terms are spelled out. Images show the product from different angles and in use.

This list is also the foundation of classic Shopping ads. AI, though, can use these fields when answering questions like “how long does shipping take?”, “can I return it?” or “will this size fit me?” A missing field is an unanswered question.

Don’t treat Merchant Center as just an ads tool. Your products can also appear in free listings, and Google’s AI guide ties Merchant Center feeds directly to visibility in AI answers. Even for a small e-commerce site with no ad budget, I’d say building a feed is one of the most productive things you can do today.

Setting up a feed once and leaving it isn’t enough. New products, price changes and promotions should flow into it regularly. You can do this with your e-commerce platform’s automatic feed feature or a feed management tool. The goal is for the feed to be an up-to-date copy of your site at all times.

4. When many sellers carry the same product

In e-commerce, most products aren’t sold on just one site. The same shoe might be sold on the brand’s own site, on three marketplaces and by ten dealers. Which seller AI shows the user in that case is a separate question.

OpenAI has given the clearest explanation. In its September 2025 announcement, it wrote that product results in ChatGPT are organic and unsponsored, ranked purely on relevance to the user’s question. When several merchants sell the same product, it said it considers availability, price, quality, whether the merchant is the primary seller and whether Instant Checkout is enabled (OpenAI).

Instant Checkout no longer carries the weight it did, but the rest of that list works as a checklist for any platform. Is the product available right now? Is the price competitive and the same everywhere? What do the seller’s reviews, return terms and delivery times look like? Who’s selling it: the brand itself, an authorized dealer or an unknown middleman?

The last question matters especially for brands. If you also sell your own product on your own site, your product page and feed need to be more complete, more current and more trustworthy than the copies on marketplaces. Otherwise AI may recommend your product and send the user to another seller.

There’s also the ads side. Google shows ads in AI Mode in some countries and has started a pilot called Direct Offers for merchants offering discounts (Google). These appear separately, labeled “Sponsored.” This book is about the organic side, but both run on the same data, meaning the same feed.

5. New conversational fields in Merchant Center

In January 2026, Google announced dozens of new data attributes in Merchant Center for conversational shopping. According to the announcement, they go beyond classic keywords and include answers to common product questions, compatible accessories and substitutes for the product (Google). In May 2026, Google wrote that these conversational attributes had opened up to merchants worldwide (Google).

The most interesting of these is the question_and_answer attribute. According to the Merchant Center help page, it’s designed primarily for conversational experiences like AI Mode and is available in all countries (Google Merchant Center Help). You can add up to 30 Q&As per product; each question and answer can be up to 1,000 characters, and all Q&As for one product can total up to 10,000 characters. Questions can come from users, the merchant or the manufacturer.

Other new attributes include related products (related_product), links to documents such as user manuals (document_link) and a popularity rank showing the product’s sales performance within the catalog (popularity_rank).

That means customer questions now belong in your feed, not just on your page. You can take the questions from Chapter 12, the ones that come up most often from your sales team, your call center and your product reviews, and add them with answers to each product’s feed entry. For a volleyball shoe, they might look like this:

• “Does it run narrow?” “True to size; people with wide feet should go one size up.”
• “Can I use it outdoors?” “The sole is made for indoor courts and wears quickly outside.”
• “How’s the ankle support?” “It’s a high-top model designed for people with a history of ankle sprains.”

These answers are just examples. Your real feed should hold real customer questions with accurate answers. The answers also mustn’t contradict what’s on the page; the rule from the first section applies here too, and the site and feed have to say the same thing.

6. A year of announcements

The second layer, checkout, has moved fast over the past year and changed direction more than once. Here are the main announcements in date order:

Sources: OpenAI, September 2025; Google, November 2025; OpenAI, November 2025; PayPal, November 2025; Axios, January 2026; Google, January 2026; OpenAI, March 2026; Google, May 2026; Google, October 2026.

All the big platforms are moving into this space, and almost everything starts in the US before spreading to selected countries. None of the checkout features in the table has an announced date for Turkey.

7. ACP and UCP

“Protocol” sounds technical, but the idea is simple. For an AI agent to shop at hundreds of different stores, it has to speak the same language with each of them: what the product is, what it costs, whether it’s in stock, how to add it to the cart, how to pay and how to track the order. Commerce protocols define that shared language. Two stand out today.

The Agentic Commerce Protocol (ACP) is the protocol OpenAI built with Stripe and released as open source. It was first used for Instant Checkout in ChatGPT (OpenAI). Since March 2026 it has been used mainly for product discovery: merchants send their product feeds and promotions to ChatGPT through it. According to OpenAI, retailers such as Target, Sephora, Nordstrom, Lowe’s, Best Buy, The Home Depot and Wayfair have connected to ACP for discovery, and Shopify merchants are included through Shopify Catalog with no extra work (OpenAI). OpenAI’s developer documentation says the standard feed upload targets the US; the shared format lists the US, Canada and Mexico as target countries, and other markets need OpenAI’s approval (OpenAI Developers).

The Universal Commerce Protocol (UCP) is an open standard Google announced in January 2026 and developed with Shopify, Etsy, Wayfair, Target and Walmart. More than 20 companies, including Adyen, American Express, Best Buy, Mastercard, Stripe, Visa and Zalando, were announced as supporters. Google says UCP covers the whole shopping journey from discovery to purchase and post-purchase support, and works with other protocols such as Agent2Agent (A2A), the Agent Payments Protocol (AP2) and the Model Context Protocol (MCP) (Google). A March 2026 update added multi-item carts, real-time price and inventory from the merchant’s catalog and linking of loyalty program memberships (Google).

According to the Merchant Center help page, UCP-powered checkout currently applies to eligible products from participating merchants in the US, Canada and Australia (Google Merchant Center Help). The merchant stays the seller of record, and the user pays through Google Pay with the payment and shipping details saved in Google Wallet.

An e-commerce manager doesn’t need to know the technical details. Most merchants won’t build these protocols themselves; they’ll use them through e-commerce and payment platforms such as Shopify, Stripe and Salesforce. Google has also said partners like Salesforce and Stripe will add UCP to their platforms. The question to ask is when, and in which countries, the platform you use will support it.

8. ChatGPT’s step back

The most interesting row in the table is March 2026. Six months after announcing Instant Checkout in September 2025, OpenAI stepped back from it. In the company’s words: “We’ve found that the initial version of Instant Checkout did not offer the level of flexibility that we aspire to provide, so we’re allowing merchants to use their own checkout experiences while we focus our efforts on product discovery.” Merchants will use their own checkout pages, and OpenAI will focus on product discovery (OpenAI).

According to TechCrunch, OpenAI said it wouldn’t prioritize developing Instant Checkout as a standalone feature and that merchants could keep using it for a while through apps inside ChatGPT. The report also cites a source speaking to The Information who said users weren’t really using the chatbot to make purchases (TechCrunch). In the same announcement, OpenAI said Walmart had launched a ChatGPT app supporting account linking, its loyalty program and Walmart payments.

If that source is right, people use AI to research, compare and decide on products, but still seem to prefer paying on a site they know. That fits what we saw in Chapter 1: visitors who arrive from AI are close to a decision. In this picture, discovery matters more than checkout.

Your checkout is still your responsibility. At İmza’s clients, AI traffic is usually under 10% of the total, but it converts at around 6–7%. Meeting a user who has decided, or nearly decided, with a slow checkout or a price that contradicts the product page means losing your most valuable visitor.

An announcement and widespread use are also two different things. Not every announcement in this area means everything changes tomorrow. The direction is clear, though: the platforms are pulling discovery inside their own products, and they’re doing it with data the merchants send them.

9. What’s available in Turkey today

Based on the sources as of November 1, 2026:

Sources: Merchant Center country list, question_and_answer, UCP checkout, Business Agent, Universal Cart, Ads in AI Mode, OpenAI feed documentation.

This table will change quickly, but for a Turkish e-commerce site the priority today is clear. The checkout layer isn’t here yet, while the data layer already works. A site that gets its Merchant Center feed, product pages and structured data in order now will be among the best prepared when checkout features reach Turkey.

10. When agents visit your site

Even when the purchase doesn’t happen inside the AI, agents can come to websites. Google’s AI guide says browser agents can visit your site to complete a task, and do so by analyzing screenshots of the page, inspecting the DOM and interpreting the accessibility tree (Google Search Central). The guide points to web.dev’s guide on agent-friendly sites.

Most of the advice there is the basics of a good website (web.dev):

• Use real button and link elements for clickable items instead of boxes styled to look like buttons, and connect form fields to their labels.
• Avoid transparent overlays on top of interactive elements; an agent may ignore the element underneath.
• Keep the layout stable. An “Add to cart” button that sits in a different spot on every category page confuses agents that work from screenshots.
• Check the accessibility tree. The structure screen readers use is also an agent’s map of the page.

The guide also mentions WebMCP, a proposed web standard that would let sites declare the actions agents can take on them. It’s still in development: worth following, but not a priority today.

In Chapter 1 I wrote that sites hiding prices in PDFs, specs in images and contact details behind complicated forms would be at a disadvantage both in AI answers and with agents. This guide is the technical version of that sentence.

Keeping the site up is part of this too. After one client’s site was down for about an hour, we started uptime monitoring for every client. Agents, bots and feeds may visit your site at different hours and frequencies than people do. If the site is down when an agent arrives, as far as that agent is concerned, the product doesn’t exist.

11. B2B and quote-based sales

Everything so far has mostly been about products you add to a cart and pay for. B2B products sold by quote, with prices that vary by customer, look different.

Back to the filling machine example: a buyer who needs to fill 3,000 bottles of a foaming product a day, with a budget of around €25,000. I don’t expect that buyer to tell an AI agent “add to cart and pay” any time soon. But an agent finding suitable manufacturers, comparing specs, working out price ranges and requesting quotes from a few of them looks much closer.

We have a B2B client with a similar profile: 30 to 40 visits a day, selling a €20,000 machine. On sites like this, every visitor counts. If an agent or an AI can’t get accurate information from the site, that’s less a lost visit than a lost sales lead.

A few things make a difference on B2B sites. Even if you can’t give an exact price, a line like “machines in this capacity range usually cost between X and Y” lets AI assess you against a budget; a site with no price information at all starts behind on budget questions. Capacity, dimensions, energy use and suitable product types should be written as text on the page, not buried in a catalog PDF. Delivery time and service area should be stated clearly; in Chapter 1 these were among the follow-up questions buyers ask when deciding. The quote form should be simple: a few clearly labeled fields that people, and later agents, can fill in.

12. Where to start

If I turn this chapter into a preparation list, the order looks like this:

1. Keep price, stock and basic information on product pages accurate and current. Everything else builds on this.
2. Add structured data for price and availability to product pages; Merchant Center’s automatic updates depend on it.
3. Complete your Merchant Center feed: title, description, GTIN, variants, shipping and returns.
4. Starting with your best sellers, fill in the question_and_answer attribute with real customer questions.
5. Bring marketplace and comparison site information in line with your own site. As we saw in Chapter 14, AI can pull product information from those sites too.
6. Check your site for agents: semantic HTML, a stable layout, the accessibility tree.
7. Follow your e-commerce platform’s UCP and ACP roadmap, so you’re among the first ready when these features reach Turkey.

On measurement, Google announced in May 2026 that it would add a tool to Merchant Center (AI performance insights) comparing a brand’s performance on AI surfaces with similar brands. It’s launching first in Australia, Canada, India, New Zealand and the US (Google). In Turkey, for now, you’ll need the methods from Chapters 7 and 8: regular prompt sets and AI-referred traffic in Analytics.

13. Data first

Checkout is the part of AI commerce that makes headlines, but data is the part that works today. AI can only recommend your product as accurately as the data you give it. A site that keeps price, stock and product information identical and current on the page, in the feed and on marketplaces stands out in today’s answers, and will be ready from day one when checkout features reach Turkey.

This chapter focused on e-commerce. AI search doesn’t affect every industry the same way, and priorities shift for B2B companies, local businesses, news sites and businesses selling abroad.

Checklist

Sources

• Google Search Central: Optimizing your website for generative AI features on Google Search (last updated July 10, 2026)
• Google Merchant Center Help: Supported languages and currencies (accessed November 1, 2026)
• Google Merchant Center Help: Price [price] (accessed November 1, 2026)
• Google Merchant Center Help: Allow Merchant Center to update product information automatically (accessed November 1, 2026)
• Google Merchant Center Help: Question and answer [question_and_answer] (accessed November 1, 2026)
• Google Merchant Center Help: About the Universal Commerce Protocol (UCP) and UCP-powered checkout feature on Google (accessed November 1, 2026)
• Google: Let AI do the hard parts of your holiday shopping (November 13, 2025)
• Google: New tech and tools for retailers to succeed in an agentic shopping era (January 11, 2026)
• Google: Universal Commerce Protocol updates improve AI shopping for retailers (March 19, 2026)
• Google: How we’re helping retailers thrive with new Universal Commerce Protocol features and AI tools on Google (May 20, 2026)
• Google: An intelligent new shopping cart (October 15, 2026)
• Google: Ads in AI Mode expand to new markets (October 28, 2026)
• OpenAI: Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol (September 29, 2025)
• OpenAI: Introducing shopping research in ChatGPT (November 24, 2025)
• OpenAI: Powering product discovery in ChatGPT (March 24, 2026)
• OpenAI Developers: Product feed spec (accessed November 1, 2026)
• TechCrunch: OpenAI’s plans to make ChatGPT more like Amazon aren’t going so well (March 24, 2026)
• PayPal: PayPal Launches Agentic Shopping with Perplexity (November 25, 2025)
• Axios: Microsoft turns Copilot chats into a checkout lane (January 8, 2026)
• web.dev: Build agent-friendly websites (last updated April 1, 2026)

Previous: Chapter 14, Third-Party Sources · Table of contents · Next: Chapter 16, AI Search by Industry