What is agentic commerce? How to make your website agent-ready

Agentic commerce is a way of buying in which an artificial intelligence agent handles part or all of the journey for you: it understands what you need, finds options, compares prices and terms and, if you give permission, completes the order and the payment. You delegate; the agent acts within the limits you set.

What separates it from a search engine or a chatbot is action. Typical agentic commerce examples are an agent filling a basket, booking a table or requesting a quote. The word agentic refers to agency: the capacity to act on someone's behalf.

Think of it as a ladder. As of October 2026, almost all real activity sits on the first two rungs; the third exists in pilots and in specific markets. Last reviewed: 1 October 2026; third-party figures carry their date and source.

• Assist: AI answers questions about products and summarises reviews and terms.
• Compare: the agent visits several sites and proposes a reasoned shortlist.
• Act: the agent books or pays with the user's express authorisation. This is the delegated purchase.

Today, AI shopping agents influence far more than they execute. People use AI to discover and compare before they visit the shop, and they still pay themselves. The consequence is immediate: if the agent cannot find you or cannot understand you, you never make the shortlist.

A McKinsey survey published in March 2026, covering 749 consumers in France, Germany and the UK, found that 38% of respondents use AI in their buying decisions, mostly to compare, learn and discover before they reach the retailer's site. A separate McKinsey analysis with EuroCommerce, from June 2026, puts at 61% the share of European consumers who already use AI to discover or evaluate products. The two figures come from different sources and methods, and Spain was not in the survey sample: read them as a direction of travel, not a portrait of the Spanish market.

For Spain, the reference figure dates from 2025. According to IAB Spain's eCommerce Study 2025, 13% of online shoppers already consulted AI tools for product information, and interest in AI advice was highest for technology, followed by travel. That is why hospitality and tourism is one of the first sectors to feel it.

Today's contest is about visibility inside assistants' answers, the territory of GEO, explained in what is GEO. The contest over execution comes later.

An agentic commerce protocol is a shared language that lets an AI agent and a merchant understand each other: what is for sale, at what price, how it goes into the basket and how the user's authorisation to pay is evidenced. In a short time, five acronyms have appeared that are worth recognising.

For a small or mid-sized business, none demands an urgent integration: availability varies by country. The sensible course is to watch for their arrival in your market and work now on what they all share: complete product data, clear policies and a purchase flow a machine can follow.

• UCP (Universal Commerce Protocol): the protocol Google announced on 11 January 2026. It powers checkout in AI Mode and the Gemini app with eligible US retailers.
• ACP (Agentic Commerce Protocol): the protocol backed by OpenAI for connecting a merchant's catalogue and checkout to an assistant.
• AP2 (Agent Payments Protocol): the payments piece that sits alongside UCP: how an agent pays on a person's behalf with their authorisation.
• A2A (Agent2Agent): a protocol that lets agents from different providers talk to one another.
• MCP (Model Context Protocol): connects AI models to external tools and data. Anthropic donated it in December 2025 to the Agentic AI Foundation, formed under the Linux Foundation.

In Spain, as of October 2026, agentic commerce is real at the discovery stage and experimental at the payment stage. Genuine agent-initiated purchases have taken place with Spanish banks, but within a pilot. Google's checkout in AI Mode and Gemini works with US retailers, and OpenAI withdrew its instant purchase feature.

What has already happened. In July 2026, BBVA, CaixaBank, Bankinter and Abanca completed real purchases initiated entirely by AI agents, with Visa and within the Visa Agentic Ready programme, as El Español reported on 2 July 2026. Visa had launched the programme in Europe on 17 March 2026. That is where agentic commerce with Visa stands in Spain today: a test with real money, not a widely available service.

What does not exist here yet. Buying through UCP in AI Mode and Gemini is live with US retailers; as of October 2026 we have found no confirmation that it is available to Spanish merchants. And OpenAI retired Instant Checkout from ChatGPT, as reported in March 2026: purchases are completed on the merchant's website.

What it means for you. Assistants work today as discovery surfaces: they recommend, compare and send the user to your site, which still closes the sale and now has to convince two readers: the person and their agent.

A browsing AI agent can read your website in three ways: the DOM, meaning the structure of the code; the accessibility tree, the version screen readers use; and screenshots it interprets visually. That is how Google's guide to AI search, updated in July 2026, describes it.

The conclusion is useful: whatever helps a person using a screen reader also helps an agent. Buttons that are real buttons, labelled form fields, ordered headings and visible states (sold out, selected, error) let a machine work through the interface without guessing. The guide points to web.dev's recommendations for agent-friendly websites.

In e-commerce, accessibility is also mandatory. Spain's Law 11/2023 has required consumer e-commerce services to be accessible since 28 June 2025, with micro-enterprises providing services exempt. Good UX/UI and accessibility work serves the legal requirement and agent readability at the same time.

The most common obstacles can almost always be fixed without rebuilding the site:

• Prices or availability that only appear after a complex interaction or inside an image.
• Menus, filters and selectors with no accessible name or role.
• Pop-ups that cover the content and cannot be closed from the keyboard.
• Checkouts that force account creation before showing delivery costs.

An agent can only recommend or buy what it can read without ambiguity: what you sell, what it costs including tax, whether it is in stock, what delivery takes and costs, and how returns work. If that data is complete and consistent across your site, feed and profiles, you compete; if not, you drop out of the comparison.

Google's guide states that Merchant Center feeds and Google Business Profiles help products and services appear in its AI responses as well as in classic results. Keeping that feed complete and consistent with the website is the job of marketplaces and retail media.

Structured data is still useful, with realistic expectations: Google makes clear it is not a requirement for its AI features. The law helps too: Article 10 of Spain's LSSI-CE requires published prices to be clear and to state whether they include taxes and delivery costs. If your shop does not yet provide that foundation, the starting point is the e-commerce build itself.

• Product: name, brand, variant and standard identifiers.
• Price: final amount including tax, and how long offers run.
• Availability: real stock for each variant.
• Delivery: areas served, lead times and shipping costs.
• Returns: time limit, cost and procedure, in text and not only in a PDF.

Before optimising anything, check that agents can get in. The main AI platforms run several bots with different jobs (training, search and visits made on a user's behalf), each managed separately in robots.txt. On top of that, your CDN or firewall may be blocking them without your knowledge.

At OpenAI, OAI-SearchBot surfaces your site in ChatGPT search and GPTBot collects training data; both respect robots.txt. Blocking GPTBot does not remove you from ChatGPT search; blocking OAI-SearchBot does.

User bots are the closest thing to agentic commerce: they visit a page when a person asks their assistant to. OpenAI notes that robots.txt rules may not apply to ChatGPT-User, and Perplexity says Perplexity-User generally ignores them; Anthropic's Claude-User does honour them.

With Google, blocking Google-Extended does not take you out of AI Overviews: it controls whether your content is used to train and ground Gemini, not your presence in Search. Infrastructure matters too: since July 2025, Cloudflare has blocked known AI crawlers by default and asks new domains whether to allow them, so a shop can be invisible to assistants even with a flawless robots.txt. Reviewing both layers is part of a technical SEO audit.

Trust is the main brake on agentic commerce today. According to McKinsey's March 2026 survey in France, Germany and the UK, consumers trust AI's judgement when it compares and recommends, but far less when it acts alone, and they ask for three things: reversible actions, clear accountability and explicit permission.

For a merchant, these are design and commercial-policy decisions, not exotic technology. What is new is that a machine now reads those terms first: a confusing policy can leave you out of the comparison.

If you also deploy your own AI agent to serve customers, Article 50 of the EU AI Act has required since 2 August 2026 that people be told they are interacting with an AI, unless it is obvious.

• Reversibility: simple cancellation and returns, with time limits and costs in plain text.
• Explicit permission: a visible confirmation before payment, with the full summary: product, final price, delivery cost and delivery date.
• Accountability: who the seller is, how to reach a person and what happens if something goes wrong.

Traffic arriving from AI assistants can be measured, within limits. Some of it shows up in your analytics as referral visits from the assistants; what comes from Google's AI features is blended into organic search. The sensible approach is to combine three sources: web analytics, Search Console and Bing Webmaster Tools.

Since 31 August 2026, Search Console has offered every site generative AI performance reports for AI Overviews and AI Mode: they show impressions, not clicks or CTR, so they are a visibility signal rather than sales attribution. Bing Webmaster Tools adds an AI Performance report, in public preview since February 2026, showing citations in Copilot.

On its value, one figure deserves a careful reading: according to Adobe Analytics data, published by a software vendor and covering US retail sites only, AI-referred traffic converted 54% better than other traffic in May 2026. It suggests those visits arrive with firmer intent; it is not a forecast for your shop.

In practice, create a segment for visits arriving from assistants and measure their conversions separately, with consent-respecting analytics and tracking.

Getting ready for agents does not mean waiting for any protocol: it means making your offer readable, comparable and actionable by a machine. Order matters: access first, then data, then the purchase or booking flow and, finally, measurement.

Online shops are the most exposed, because the agent compares product by product; for consumer brands and e-commerce, the feed and the product page come first. Booking-based businesses, such as hotels, restaurants and clinics, need availability and terms to be readable without a phone call. In B2B, the advantage goes to whoever documents their offer on open pages rather than only in PDF catalogues.

An agent-ready commerce audit checks these points against your actual website and ranks them by impact. And to see this change alongside the others, read the future of digital marketing.

• Everyone: review robots.txt and your CDN and decide which AI bots get in.
• Everyone: check that price, availability and terms can be read in the HTML, and fix the accessibility of menus and forms.
• Online shops: complete your Merchant Center feed and make it match the website.
• Online shops: publish delivery and returns terms in text and show the total cost before requiring an account.
• Booking-based businesses: publish availability, prices and cancellation terms, and keep your Google Business Profile current.
• B2B: publish specifications, lead times and terms on crawlable pages and offer a clear route to request a quote.
• Everyone: create an AI traffic segment in your analytics and review each quarter what has changed.