Three major developments in 2025 and 2026 point in one clear direction. Google, OpenAI, and Anthropic are each taking a different path, but their goal is the same: users are increasingly delegating purchase decisions to AI assistants.
If you rely on customers visiting your website, browsing products, and clicking "Buy," you have about two to three years before this traditional behavior begins to decline in the Czech Republic.
Here's a critical fact most e-commerce stores ignore: an AI assistant doesn't read your website. It reads your product feed. Your site's beautiful design, great photos, and compelling brand story don't exist for an AI assistant. Meanwhile, Google Merchant Center defines over 70 fields and parameters. These attributes determine whether an AI will show your product to a customer at all.
What Agentic Commerce is (the trend in a nutshell)
We all know the classic shopping journey: a user goes to an e-commerce store, picks a product, and buys it. In autonomous shopping, an AI assistant enters the process. The user simply delegates a task, and the assistant scans e-commerce stores, compares offers, and picks the best one.
Picture it in practice:
- User: "Find and buy me a new espresso machine for the kitchen, under $350. I want espresso with milk and easy maintenance."
- AI assistant: Scans dozens of e-commerce stores via their APIs and structured data. It filters down to a handful of candidates, compares them against the user's priorities, and presents the best options.
- User: Simply approves the choice, and the AI assistant places the order.
Today, part of the process is still manual and requires user confirmation. But in the coming years, expect greater autonomy and fewer steps for the user.
Three big signals from 2025-2026
1. Google Shopping AI tier (AI Max for Shopping)
In late 2025, Google launched AI recommendations directly in search results. Instead of traditional links, users see an AI-generated carousel with a few products and a brief explanation for the selection. The AI system reads parameters exclusively from the feed in Merchant Center. It prioritizes product titles, descriptions, categories, and key benefits. The content of the e-commerce site itself is skipped at this stage.
2. OpenAI Operator and ChatGPT Shopping
In early 2025, OpenAI demonstrated a tool called Operator. It's an AI assistant that can browse the web on its own, navigate sites, and fill out forms at checkout. This feature later became part of the broader ChatGPT assistant. Operator reads structured data and website architecture. If you have a complex checkout flow with missing element labels, the assistant won't be able to complete the purchase.
3. Anthropic Claude and the MCP protocol
The MCP protocol is an open standard for communication between AI and external systems or databases. E-commerce stores are already experimenting with it, exposing their product catalogs via API directly to the Claude assistant. This gives Claude access to structured data in real time. Major e-commerce platforms are expected to support this protocol natively soon.
This allows assistants to verify price and availability in milliseconds before a purchase. E-commerce stores with legacy systems, slow servers, or no public API won't stand a chance. The system won't include them in its selection because they can't respond in time. Moving to a modern platform with an open API is therefore a strategic decision.
Why classic SEO + PPC does not solve the problem
This is a major shift for PPC managers and SEO specialists. Current optimization is targeted toward generating a human click. But AI assistants work completely differently and ignore many of the elements we typically optimize.
| What you optimize today | What the AI assistant ignores |
|---|---|
| Responsive search ad copy | The AI assistant doesn't see the ad. It goes straight to the feed data. |
| Landing page look, UX, and copy | The assistant doesn't read the website. It needs structured data. |
| Branding and tone of voice | AI has no emotions. Your brand story won't move it. |
| Meta descriptions for higher CTR | AI doesn't care about click-through rate. It cares about a clean parameter match. |
| Product photos in PMax campaigns | AI models analyze images programmatically, but emotion plays no role. |
On the other hand, the AI assistant actively uses the product feed, structured data on the website (Schema.org), APIs, and the store's rating from reviews. Current price and real-time availability are also critical. If that information is missing, the product simply doesn't exist for the assistant.
What the AI assistant reads (and what it does not)
Let's look at an example. A customer is looking for a men's leather handbag for a 15-inch laptop under $200. The AI assistant zeros in on specific concepts in the feed.
| Semantic concept | What AI looks for | Real GMC feed field |
|---|---|---|
| Material | "leather" | material |
| Target audience | "men's" | gender: male |
| Use case | "for 15-inch laptop" | product_highlight or custom_label |
| Price | "under $200" | price |
| Dimensions | width over 38 cm | product_detail or description |
| Trust | reviews and brand strength | Shop and product ratings |
Store A has a beautiful website, great photos, and a strong brand story. But in its product feed, the product title is just Handbag KH-204, with no material or gender specified. The AI assistant will most likely skip it.
Store B has an ordinary website and photos. But in Shopping, its product title is: Picard Sonja leather men's handbag for 15" laptop, brown. Additionally, it has the material, gender, and key benefits filled out correctly.
Store B wins. It may have a worse website, but it's readable by the AI assistant. That's the basic principle of this new form of search.
7 signals that decide your visibility
1. Structured material (material)
Forget generic phrases like "quality textile" or "premium material." Be precise: "100% cotton," "leather," or "Gore-Tex membrane." This information belongs in a dedicated field, not just the product description.
2. Use case (product_highlight + custom_label)
A dedicated field for use case doesn't exist in the Merchant Center help documentation yet. But you can address this with a combination of key benefits and custom labels. Instead of phrases like "for every occasion," use specific terms like "running in the rain," "office work," or "home workout."
3. Target audience (gender + age_group + custom_label)
Here too, there's no single standalone field. You define the audience using existing parameters for gender and age. If you target a specific segment, add a custom label. For example: gender: male, age_group: adult, and a custom label like "beginner runner."
4. Compatibility, or "suitable for"
This attribute is critical for electronics and replacement parts. Again, provide it using key benefits or custom labels. Be precise: "Compatible with iPhone 15" or "Fits BMW E90." Without a clear compatibility match, the AI assistant can't confidently make the purchase.
5. Price-tier signaling (price_tier)
The AI assistant also tracks a product's price level. Based on the user's budget, it estimates whether they are looking for budget-friendly items, mid-tier options, or a premium model. Custom labels work well for this purpose, for example, custom_label_4: premium.
6. E-commerce store trust (Seller Rating and reviews)
An AI agent doesn't just hunt for the lowest price. It also tries to minimize the risk of errors and returns. If a customer wants a reliable product, the assistant will focus on the store's rating (Google Seller Rating), reviews on platforms like Heureka or Trustpilot, and the brand's overall history. An e-commerce store with a low price but no reviews is unlikely to be selected by the AI. Investing in review-collection platforms becomes essential. A lack of trust can also lead to a hard misrepresentation ban in Google Ads, resulting in a complete account shutdown.
7. Conversational attributes (6 new fields straight from Google)
In Merchant Center, Google officially labels a set of six fields as conversational attributes: question_and_answer, document_link, related_product, item_group_title, variant_option, and popularity_rank. These help AI systems understand product details for AI Mode and Agentic Commerce. AI agents prioritize reading them because they contain structured answers to common buying questions ("Does the product have this feature?" "Does it fit a specific model?"). If you want to be included in the AI's selection, filling out these fields is essential.
A hidden threat: returns
Autonomous shopping significantly speeds up the purchase process. But the first warning signs are emerging from other markets: for purchases made by AI agents, return rates are climbing. The cause is often incomplete or inaccurate data in the product feed.
The AI agent picks the wrong size or variant because of bad data. The customer discovers the error upon delivery and returns the item. This increases logistics costs and erodes your margins. Accurate data is therefore your main defense. Focus on a clear sizing system (size_system), precise sizes (size), specific colors, and dimensions in centimeters.
A hidden threat: the attribution crash (GA4, gclid, Meta Pixel)
This is the most technically challenging aspect of this topic. When an AI agent makes a purchase, a web browser is never opened. The entire process runs via API in the background. As a result, traditional conversion measurement fails.
JavaScript doesn't fire, so Google Tag Manager doesn't load and GA4 doesn't record the purchase. Cookies aren't saved, which disables the Meta Pixel. The URL is missing parameters like gclid, so Google Ads never sees the conversion. In analytics, these purchases show up as direct traffic (Direct / none) or don't show up at all. Meanwhile, your campaigns learn from conversion data. If they lose these signals, the algorithm will start to throttle performance and budgets.
And the new protocols for AI communication and payments are already in routine use. They include the Model Context Protocol (MCP) from Anthropic, OpenAI Operator, and payments via the Agent Payments Protocol (AP2) from Google. These aren't plans for the next decade; these technologies are in use today.
Solution: move to server-side measurement
The only reliable solution is to send conversion data directly from your server, not from the user's browser. You can use three main layers to accomplish this:
- Server-side GTM: A measurement endpoint on your own domain that receives data from your e-commerce backend and forwards it to GA4 or ad systems. This allows you to bypass the browser entirely.
- Meta Conversions API (CAPI): The official interface for sending events directly from your server to Meta. This ensures reliable attribution even without the pixel firing on your website.
- Offline conversion imports for Google Ads: Using the API, you can send the conversion value and the original click identifier (
gclid) back into the system, safely preserving attribution.
What you can do today:
- Check in GA4 what percentage of your conversions fall under unassigned sources. If it's more than 10%, you have a data gap.
- Implement server-side GTM on your own subdomain.
- Set up the Meta CAPI with correct event deduplication.
- Prepare your e-commerce backend to store and send the
gclidparameter for completed orders.
E-commerce stores with this setup can measure AI purchases as precisely as human ones. Those who stick with legacy measurement will lose visibility into their campaign performance.
A four-week action plan
Week 1: data audit
Export your product feed to a CSV file. Check how many products have the material, key benefits, gender, or age_group attributes filled out. Identify your hundred best-selling items by revenue over the last year.
Week 2: optimize your top products (a supplemental feed with no developer)
For your top 100 products, fill out the missing material, gender, and age_group attributes. Add three to five clear benefits to each product. For accessories, don't forget to state compatibility. Also, set price tiers using custom labels.
Week 3: check and scale
Check the Diagnostics tab in Merchant Center to confirm the products haven't been disapproved. Monitor for changes in click-through rate (CTR). If you see a positive shift, apply the same edits to the next 500 products in your catalog.
Week 4: structured data and reviews
Deploy structured data (Schema.org) on your product pages. Verify that Google correctly displays your product ratings from reviews. If you aren't using platforms like Heureka or Trustpilot yet, it's time to start.
An outlook on the near future
The window to prepare is closing. We expect AI features to be integrated into shopping in the Czech Republic during 2026. A significant impact on traffic will be apparent by early 2027. Additionally, local players like Heureka are also experimenting with AI integration.
Product data enrichment also delivers immediate results, improving performance in PMax campaigns, standard Shopping campaigns, and AI Overviews. The investment pays off right away. If you wait until 2027 to start, you will struggle to catch up to the competition. By 2028, you'll be at a significant disadvantage.
Summary
- Autonomous shopping is coming. The main wave will hit in 2026 and 2027. The time to prepare is now.
- AI assistants don't read your website. They read your product feed. Your e-commerce store's design means nothing to them.
- Focus on key signals: material, use case, compatibility, audience, price level, store trust, and Google's conversational attributes are what matter.
- The future is direct APIs. Soon, assistants will pull data directly from e-commerce platforms, bypassing slow or closed websites.
- Manage your return risk. Inaccurate data leads to purchasing errors by AI agents. Detailed parameters protect your margins.
- Traditional conversion measurement is becoming obsolete. AI agents bypass browsers. The solution is to move to server-side measurement and the Conversions API.
- New protocols are already in use. Standards like MCP and AP2 aren't science fiction; they are being used today.
- Start with a four-week sprint. Optimizing your top 100 best-selling products will give you a solid head start.
- The investment pays off immediately. Better data improves your PMax and Shopping campaigns as soon as you deploy it.
- An early start is decisive. If you postpone these changes, you'll struggle to close the gap with prepared competitors.
Enriching your feed data isn't just a routine operational task. It's a strategic decision that will determine your visibility in the AI era. If two stores offer the same product at the same price, the one with the better-prepared feed wins.
