AI in search and shopping

Mark needs a new washing machine. Five years ago he would have typed "washing machine 8 kg" into Google and opened five pages. Today he types into a chat: "Which washing machine for a family of four isn't too loud and costs up to 500 euros?" The assistant replies with two or three suggestions and a few cited sources underneath. Mark often opens none of them.

If your store sells washing machines, the question is no longer whether you are on the first page of Google. It is whether the assistant knows you, trusts you, and describes you correctly. That is what this page is about. First, what has really changed; then what helps people and machines at the same time; and finally, what can actually be measured today.

What has changed

For a merchant this means three things:

• fewer clicks from search at the same ranking, because part of the queries now end with an answer;
• more valuable visitors: those who do arrive are often closer to buying;
• a new opportunity: being mentioned or recommended in an assistant's answer.

We do not know exactly how many people already shop this way; Google AI Overviews are available in many EU countries and languages (status as of autumn 2026). So we do not recommend changing everything. Strengthen the basics that help both worlds: clear content, good data, a fast site and trust.

How to be visible in answers

How assistants choose sources

They search, work out the intent, and then pick from pages, catalogs, reviews, forums and encyclopedias. Sources that are clear, verifiable and consistent are more likely to be chosen.

What helps

• concrete answers to concrete questions (guides, comparisons, FAQs); a summary at the top of the page;
• structured data (JSON-LD, examples below);
• identical company details everywhere (name, address, contact, description);
• external confirmation: reviews, media mentions, comparison sites, forums, Google Business Profile;
• a named author with real experience (E-E-A-T) and a visible last-updated date;
• a fast, accessible site that crawlers can read without heavy rendering in a browser;
• unambiguous product data (brand, model, GTIN).

What to avoid

Keyword stuffing, mass-produced content with no value, fake reviews, content behind a password, and pages that most AI readers see empty or incomplete because they depend on JavaScript.

Crawler access

There are three types of crawlers: for training (e.g. GPTBot, ClaudeBot), for search (e.g. OAI-SearchBot) and for the user (e.g. ChatGPT-User, Perplexity-User). Google-Extended is not a crawler but a robots.txt token for Google's AI training, and it does not affect search. ChatGPT-User and Perplexity-User are triggered by a user, so robots.txt rules may not apply to them; robots.txt is not binding on everyone anyway. The decision is yours: search and user crawlers are usually worth allowing (they bring visits), training crawlers are up to you. Example (adapt it to your own decisions):

# robots.txt (example)
User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: GPTBot
Disallow: /      # decision: whether to allow model training

User-agent: Google-Extended
Disallow: /      # decision; does not affect rankings in Google Search

Measurement

In GA4, look for traffic sources such as chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. Clicks from AI Overviews and AI Mode are counted by GA4 and Google Search Console as ordinary Google (organic) traffic, so they cannot be measured separately. Once a month, manually test 20 key questions from your niche in several assistants and note whether you are mentioned. Do not blindly trust paid tools for "AI answer visibility": their methodology is inconsistent.

Honestly: nobody knows a reliable recipe for getting cited. The most solid asset is good content.

Do it today (15 minutes, no coding)

1. Open ChatGPT, Gemini and Perplexity and type in three questions your customer might ask (not your store's name).
2. Write down which stores and pages appear as sources. Are you among them?
3. Open one of your product pages in the Rich Results Test and check whether it finds the price and availability.
4. Open your robots.txt (your-domain.com/robots.txt) and check whether you have ever made a decision about AI crawlers. If not, that is a decision to make, not a mistake.
5. Record the result in a spreadsheet. Repeat in a month.

Data for machine readers: catalog and structured data

• The Google Merchant Center catalog is the source for free listings and shopping surfaces, and increasingly for AI assistants too. Attributes: title, description, GTIN, brand, price, availability, shipping, returns, key features, reviews. Update as often as you can; prices and stock must match the page.
• Product JSON-LD, plus Organization, BreadcrumbList, Review (real ones), MerchantReturnPolicy and OfferShippingDetails.
• Before/after example (Product): a bad record has the price "from €449", availability "available" and no GTIN; a good record has the price 449.00 EUR, availability InStock and a valid GTIN. What the code says must also be on the page.
• Consistency: the catalog, structured data and visible content must say the same thing; a mismatch is the most common reason for rejection and lost trust. Imagine an old price in the catalog that was never updated: the assistant quotes it, the customer arrives at the store and finds a different price.
• Quality product copy: specifications in a table, intended use ("who it's for"), comparisons, FAQs, dimensions.
• Catalogs for assistants: some assistant providers offer programs for merchants; terms differ by country.

Web add-on: JSON-LD structured data examples

Structured data is markup on the page that tells search engines and AI assistants exactly what the page contains: a product, its price, availability, a review, the manufacturer. It is best written as JSON-LD (a code block in the page head), which Google recommends; the older microdata format works but is harder to maintain. On most platforms this is a module or a setting; if not, ask your developer - you do not need to write the code yourself. The examples below are shortened, the names and addresses are made up, and you should adjust the values to your own store.

The markup must match the visible content of the page exactly (price, stock); differences can be a reason for exclusion from rich results. Structured data does not guarantee display: in May 2026 Google stopped showing rich results for FAQs. An FAQ section on a page is still useful for visitors, but FAQ markup no longer brings visible rich results. Structured data helps search engines and assistants understand the page.

Product with Offer and GTIN

The most important markup for a store: name, brand, GTIN, price, currency and availability. Include a rating (aggregateRating) only if you have real reviews of the product.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Alfa WM-200 washing machine 8 kg",
  "image": "https://www.example.com/images/wm-200.webp",
  "description": "Washing machine 8 kg, 1400 rpm.",
  "sku": "WM200-WHT",
  "gtin13": "3850000000008",
  "brand": { "@type": "Brand", "name": "Alfa" },
  "offers": {
    "@type": "Offer",
    "url": "https://www.example.com/washing-machine-wm-200",
    "priceCurrency": "EUR",
    "price": "449.00",
    "availability": "https://schema.org/InStock",
    "itemCondition": "https://schema.org/NewCondition"
  },
  "aggregateRating": { "@type": "AggregateRating", "ratingValue": "4.6", "reviewCount": "37" }
}
</script>

Organization

Company details that must be identical everywhere (name, address, contact).

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Example Ltd",
  "url": "https://www.example.com",
  "logo": "https://www.example.com/images/logo.png",
  "email": "[email protected]",
  "telephone": "+386 1 000 00 00",
  "address": {
    "@type": "PostalAddress",
    "streetAddress": "Example Street 1",
    "postalCode": "1000",
    "addressLocality": "Ljubljana",
    "addressCountry": "SI"
  },
  "sameAs": [
    "https://www.facebook.com/example",
    "https://www.instagram.com/example"
  ]
}
</script>

BreadcrumbList

Breadcrumbs tell machines where a page sits in the store's structure. The last item completes the path and may have no link.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "BreadcrumbList",
  "itemListElement": [
    { "@type": "ListItem", "position": 1, "name": "Home", "item": "https://www.example.com/" },
    { "@type": "ListItem", "position": 2, "name": "Washing machines", "item": "https://www.example.com/washing-machines" },
    { "@type": "ListItem", "position": 3, "name": "Alfa WM-200 washing machine 8 kg" }
  ]
}
</script>

WebSite

Basic information about the site, optionally with store search.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "WebSite",
  "name": "Example",
  "url": "https://www.example.com/",
  "potentialAction": {
    "@type": "SearchAction",
    "target": "https://www.example.com/search?q={search_term_string}",
    "query-input": "required name=search_term_string"
  }
}
</script>

MerchantReturnPolicy and OfferShippingDetails

Return and shipping terms in machine-readable form, nested inside the offer. The values (return window, shipping cost and time) must match the terms published in your store.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Offer",
  "url": "https://www.example.com/washing-machine-wm-200",
  "priceCurrency": "EUR",
  "price": "449.00",
  "availability": "https://schema.org/InStock",
  "hasMerchantReturnPolicy": {
    "@type": "MerchantReturnPolicy",
    "applicableCountry": "SI",
    "returnPolicyCategory": "https://schema.org/MerchantReturnFiniteReturnWindow",
    "merchantReturnDays": 14,
    "returnMethod": "https://schema.org/ReturnByMail",
    "returnFees": "https://schema.org/ReturnFeesCustomerResponsibility"
  },
  "shippingDetails": {
    "@type": "OfferShippingDetails",
    "shippingRate": { "@type": "MonetaryAmount", "value": "4.90", "currency": "EUR" },
    "shippingDestination": { "@type": "DefinedRegion", "addressCountry": "SI" },
    "deliveryTime": {
      "@type": "ShippingDeliveryTime",
      "handlingTime": { "@type": "QuantitativeValue", "minValue": 0, "maxValue": 1, "unitCode": "DAY" },
      "transitTime": { "@type": "QuantitativeValue", "minValue": 2, "maxValue": 4, "unitCode": "DAY" }
    }
  }
}
</script>

Check your markup with the Rich Results Test and the Schema Markup Validator (Google's Structured Data Testing Tool has been retired). In Google Search Console you can follow the enhancement reports. The same data should also be in your Merchant Center catalog.

Shopping agents

An AI agent is a program that searches, compares and sometimes pays on a customer's behalf, with their consent. To do this it uses shared conventions (protocols) shaped today by the big players: Anthropic, OpenAI, Google, Stripe and the card networks. The names will change; the fundamentals will not. If your product data is clean, price and stock are accurate, and delivery and return terms are machine-readable, you are ready for whatever protocol comes next. If not, no protocol will save you. Do not commission an expensive integration before you have measured whether anyone actually comes.

What this means for a small or mid-sized store

Do not launch an expensive integration. Prepare the basics:

1. a clean catalog via API or feed,
2. real-time stock and prices,
3. delivery and return terms in machine-readable form,
4. guest checkout and payment methods an agent can complete,
5. strong authentication for agent payments: the PSD2 strong customer authentication (SCA) rules are technology-neutral, so they generally apply when an agent initiates the payment too; as of the time of writing (13 October 2026) there is no official position from the European authorities (the European Banking Authority will only analyze agent payments in 2027, and the new European payment rules PSD3 and PSR have not been published yet). How this works in practice (delegated authorization, exemptions such as a trusted beneficiary) is something to agree with your payment provider and, if needed, a lawyer; do not claim that an agent payment is "PSD2 compliant" until your provider confirms it,
6. telling good agents apart from malicious bots (bot protection must not block permitted agents).

Legal note: the right of withdrawal applies whether the purchase was made by the customer or by their assistant; the legal framework for agents is still evolving. You will find more about the rules for stores on the page Online store legislation in the EU.

From our practice: how many agents actually came. We measured this ourselves. Between 1 August and 8 September 2026 (external requests only, excluding our own visits and tests), on our site, which is not among the most visited, no external AI system actually used our MCP tools: not a single tool call, and the few handshakes were our own tests. The UCP manifest was requested 8 times (mostly link previews), the agents.json file 2 times, and there were about 5,700 ordinary page visits. That does not mean MCP is pointless: today it is more useful for internal work, connecting your own tools and data to AI, while on the customer side we are only at the beginning. The volume is small, and that is exactly the point: measure before you invest. If you measure on your own site, your numbers will differ; what matters is that you measure.

The llms.txt file

This is a proposal for a website to publish a summary of its structure for language models. On our site between 1 August and 8 September 2026 (external requests only), no AI crawler requested the file; it was requested 45 times, all from SEO tools. Ordinary pages showed real usage over the same period. Google states in its documentation that its search does not use such files and that no special files or markup are needed to appear in AI features. The file is harmless and cheap, but it is not a priority. It does not replace good content, structured data and a correct robots.txt.

AI in your store

Uses and risks

Uses: descriptions and translations, search, chat and assistants, recommendations, image processing, draft replies for support, review analysis, stock forecasting, fraud detection.

Risks: invented claims (prices, terms, technical data) - you are responsible for what is said; privacy (do not enter personal data into public tools without a contract with the provider); copyright for images and text; bias; vendor dependence; cost.

Law

The AI Act (Regulation (EU) 2024/1689) applies in stages; it was amended by Regulation (EU) 2026/1744. The "literacy" requirement (Article 4) has been softer since 27 July 2026: providers and deployers must take measures to support their staff's AI knowledge, not guarantee a particular level of knowledge. The transparency rules (Article 50) apply from 2 August 2026: a system that interacts directly with people (e.g. a chatbot) must be designed so that it informs them they are interacting with AI (this is the system provider's duty; as a merchant, the easiest way to meet it is a clear label); providers of generative systems must mark their outputs in a machine-readable way (for systems already on the market before 2 August 2026, by 2 December 2026 at the latest); a deployer must disclose deepfakes (Article 50(4)). High-risk rules usually do not apply to an online store, unless you introduce credit scoring or something similar.

AI use policy (a template for your company)

1. We use only approved tools.
2. We do not enter customers' personal data or trade secrets into public tools.
3. Everything a customer sees (copy, replies, translations) is reviewed by a person.
4. Where a customer is talking to AI, we tell them clearly.
5. We do not invent products, reviews or experts; images do not mislead about how a product looks.
6. We keep conversation history and errors for a limited time that we set in advance.
7. Once a year we review which tools we use and why.

Checklist

Start with items 1, 2 and 3: everything else depends on them.

1. It is decided which crawlers are allowed (robots.txt).
2. Product pages have valid structured data (JSON-LD).
3. The catalog (Merchant Center or equivalent) matches the page.
4. Key pages have a summary, FAQs and clear answers.
5. Company details are identical everywhere.
6. Customer reviews are on your own site and on an independent portal.
7. Authors are named and the update date is visible.
8. The site is fast (Core Web Vitals) and accessible.
9. Traffic from assistants is tracked in GA4.
10. 20 key questions are tested once a month.
11. The catalog and terms are also available in machine-readable form (API, feed).
12. An AI use policy exists.

This page is a web add-on to the book "Spletna trgovina zdaj" (2nd edition). Related page: Online store legislation in the EU.