Your next customer is an AI agent. Can it use your website?

Agentic commerce explained: what it is, where it already happens, and what your website needs so AI agents can find, compare and choose you.

Two purchases, one week apart.

On Monday evening, Anna asks the AI assistant on her phone to find running shoes: size 39, under 150 euros, delivered by Friday. She doesn't open a single shop. Ten minutes later she gets three options with prices and delivery dates. She taps one and confirms.

On Thursday morning, Markus, a buyer at a food-processing company, asks his AI assistant to shortlist three valve suppliers with a valid hygiene certificate and delivery within six weeks. He gets a comparison table. One supplier has two empty cells: certificate and delivery time. Both facts are on that supplier's website, one inside a PDF, one in a sentence on a contact page. Markus forwards only the complete rows.

Neither Anna nor Markus visited the websites they bought from or rejected. Their agents did. That is agentic commerce, and it changes who your website has to convince.

What agentic commerce is

Agentic commerce means an AI agent searches, compares and, with permission, buys on behalf of a person or a company. The human sets the goal and the limits. The agent does the legwork: it opens websites, reads product pages, fills in forms and reports back.

The difference to what came before is agency. A search engine lists links. A chatbot answers questions. An agent completes a task.

Table comparing a search engine, a chatbot and an AI agent. Input: keywords; a question; a goal such as find, compare, book. Output: a list of links; an answer; a completed task or a prepared decision. Who visits your website: the person; nobody, or a crawler; the agent, on the person's behalf. What it needs from you: relevance; accurate, citable content; complete data and forms it can operate.

The agents doing this work are consumer apps, browser extensions and enterprise tools: personal assistants like Meta's Muse, browser agents like Claude in Chrome, and procurement agents inside purchasing software.

Where it already happens

In little more than a year, agentic commerce has moved from demos to live products in both consumer and business buying.

Timeline table: agentic commerce in practice, from demo to live product. Nov 2026, Intershop (B2B): German commerce vendor positions itself as a provider of agentic B2B commerce for manufacturers and wholesalers. Oct 2026, Walmart (B2C): announces Agent Profiles, an Agent Gateway and terms for outside agents, with spending limits set by the customer. Sep 2026, Meta Muse (B2C): personal AI agent for consumers that browses websites, fills in forms, books and buys after approval. Jul 2026, Visa and Lianlian (B2B): first live B2B transaction executed by an AI agent in Greater China. Mar 2026, Shopify (B2C): Agentic Storefronts, product catalogs feed AI assistants and checkout stays with the merchant. Sep 2025, Etsy and ChatGPT (B2C): first purchase completed directly inside ChatGPT.

On the buyer side, the shift is already measurable in B2B: according to Deloitte's 2026 B2B commerce research, nearly 40% of US B2B buyers use agentic AI in purchasing, for example to evaluate products, configure orders and benchmark suppliers.

What it changes

When an agent stands between you and your customer, your website is judged on what software can read and use, not on what people see.

Table: what changes when an agent stands between you and your customer, in B2C and in B2B. The visitor is software: the agent ignores your hero image and reads size, price and availability; in B2B it skips the brochure and looks for specs, certificates and lead times. Visibility means making the shortlist: three options reach the shopper and the rest never appear; incomplete rows get dropped before anyone calls you. Your brand loses the interface: the customer sees Muse or ChatGPT, not your shop; the buyer sees a comparison table, not your website. Data beats design: complete variant and stock data wins; specs in HTML win over specs in PDFs. Your analytics go blind: agent visits look like ordinary page views; lost deals leave no trace in your funnel.

The most dangerous outcome is the quiet one: the agent reaches your site, extracts half the facts and moves on. Nobody tells you that you were in the running.

Reality check: research now, buying later

Autonomous checkout is arriving slower than the headlines suggest, while agent-driven research is already everyday behavior.

  • Checkout is still rare. OpenAI discontinued Instant Checkout in March 2026 because adoption stayed low. A Checkout.com report from June 2026 found that 89% of merchants are preparing for agentic commerce, but only 3% have seen actual agent transactions.
  • Research is mainstream. In a Deloitte survey of 13,500 European consumers (May 2026), 52% of Germans had already used AI for shopping. Asked what they would hand to an agent, 34% named finding the best price and 27% comparing products, only 8% checkout and payment.
  • People keep the final decision. In B2B, Gartner found that 45% of buyers used AI in a recent purchase, mainly to research vendors, and 69% prefer to validate AI results with a sales rep.

For your business this means: the agent may not place the order yet, but it already decides who makes the shortlist. A person can check what a comparison contains. Nobody can see what it left out.

What agents need from your website

In a controlled experiment, a shop rebuilt for agents raised fully completed tasks from 49% to 89%, with the same products, prices and checkout.

Researchers Said Elnaffar and Farzad Rashidi published the study in July 2026. Browser agents based on GPT-4.1, Gemini 2.5 Flash and Grok 4 Fast completed five shopping tasks, 300 runs in total, on two versions of the same prototype shop. On the ordinary version, agents reached the main goal in 78% of runs, but met every requirement in only 49%. The gap was incomplete answers: the right product, with facts missing. The agent-ready version cut those partial runs from 43 to 3. Its changes were plain: product data also published as structured data, explicit labels on controls, product IDs and stock on action buttons, and dated evidence for claims. No agent protocol was involved.

It is one prototype shop, and the authors call it a proof of concept. But the pattern matches what we see on real websites. These are the gaps that cost the most:

Table: the website gaps that cost the most with AI agents, with a typical B2C example, a typical B2B example and the fix. Facts that live in files or images: size chart as an image; specs and certificates only in PDF data sheets; fix: publish key facts as HTML text and structured data, keep the file as a download. Selections that work only visually: chosen color shown as a highlighted tile only; configurator with the part number drawn into an image; fix: state the selected option in text, give each variant its own URL. Forms without labels: checkout field with placeholder text only; quote form with an unlabeled icon button; fix: a real label for every field, real buttons with clear names. Overlays that block: cookie banner and discount popup cover the Add to cart button; chat widget covers the contact form; fix: let agents close them with one clearly labeled button. Claims without proof: Bestseller with no basis; Certified supplier without issuer, number or expiry; fix: add source, date and validity.

Most of these fixes also help people using screen readers. Agent readiness and accessibility largely overlap.

What to fix first

Start with the data and the forms on your most important pages. Protocols can follow once agents actually ask for them.

  1. Product and offer data in plain HTML. Prices, variants, availability, specs, certificates, lead times. Add structured data on top.
  2. Operable forms and flows. Labeled fields, real buttons, overlays that close, no CAPTCHA on the core path, a guest option.
  3. Claims with evidence. Source, date and validity for certificates, ratings and promises.
  4. Know your agent traffic. Identify agent visits in analytics, decide which agents you allow, and keep a human confirmation step before anything is ordered or sent.
  5. Interfaces for agents. APIs or MCP endpoints make sense where customers or tools need direct, programmatic access. Proposed standards like WebMCP are still in testing and worth a pilot, not a rebuild.

Agent readiness builds on the work you may already be doing:

Table comparing SEO, GEO and agent readiness. SEO: goal, your page is found in search; question it answers, can people find us? GEO: goal, AI answers describe you correctly and cite you; question it answers, does AI name us correctly? Agent readiness: goal, an agent completes a task on your site; question it answers, can an agent shortlist, compare and act on our offer?

Find out how agents see your website: the JustRelate Agent Readiness Audit

Our Agent Readiness Audit shows whether AI agents can complete your customers' most important tasks on your website, where they fail, and what to fix first.

Many scanners check technical signals. We test what matters: real tasks, run by real agents.

Table: the JustRelate Agent Readiness Audit in four steps. Step 1, define tasks: together we pick 3 to 5 tasks your customers care about, such as find a product and check availability, request a quote or start a return; you get a task catalog with clear success criteria. Step 2, technical check: we test forms, overlays, consent banners, bot protection, URL structure and interfaces; you get findings with evidence for every check. Step 3, agent tests: several AI agents run each task multiple times and we log where they succeed, stop or return incomplete data; you get the success rate per task and screenshots of every failure point. Step 4, roadmap: we rank the fixes by effort and impact; you get quick wins, medium-term measures and a strategic view.

The audit follows the same proven approach as our GEO Audit. GEO shows how AI describes you. Agent readiness shows whether AI can act on your offer. Together they cover both sides of AI-driven buying, for B2C shops and B2B websites alike.

Frequently asked questions

What is agentic commerce?

Agentic commerce is buying and selling in which an AI agent searches, compares and, with the user's permission, purchases on behalf of a person or company. The human sets the goal and limits, the agent carries out the steps.

Is agentic commerce relevant for B2B or only for consumers?

Both. Consumer agents like Meta's Muse handle shopping tasks, while business buyers use agents to research suppliers and compare offers. In B2B, research by agents is already common, while autonomous ordering is still at an early stage.

What is an agent-ready website?

A website whose content, forms and data an AI agent can read, verify and use to complete a task for a person, such as comparing products or requesting a quote.

Do I need an MCP server or llms.txt to be agent-ready?

Not to start. In the Elnaffar and Rashidi experiment, the gains came from clear data, labels and evidence, without any agent protocol. Interfaces like MCP become relevant where customers or tools need direct, programmatic access.

How is agent readiness different from GEO?

GEO makes sure AI answers describe your company correctly. Agent readiness makes sure an agent can complete a task on your website. One gets you mentioned, the other gets you chosen.

Sources

Our top article
10 successful Welcome Emails examples: Tips and best practices to inspire your Campaigns
Read Article
Artificial intelligence and emails: How to speed up the creation and design processes of your campaigns?
Read Article
Trigger emails strategy: a weapon for engagement and retention
Read Article
Want to chat about the JustRelate platform?