AI

Agentic Commerce

When AI helps the customer make a choice, your webshop must be able to respond

Prepare your webshop for AI agents.

Imagine a customer asking their AI assistant to find a running jacket in the right size and purchase it with delivery before Friday.

The agent needs to be able to read your product data and determine whether the item meets the customer's requirements. If the information is missing, your webshop may be skipped; if access to checkout is unavailable, the process stops at the recommendation.

The AI assistants already visiting your website

  • ChatGPT
  • Gemini
  • Copilot
  • Claude
  • + more, added continuously

From product recommendation to purchase on the customer's behalf

The agent researches products and provides a recommendation. No transaction takes place.

This requires: A clear product foundation that the agent can use to assess relevance.

The agent takes the recommendation through to a transaction.

This requires: The same foundation as browsing, plus access to complete the purchase.

Your website has gained a new type of user

A website used to be built for the human eye. Today, AI agents open it too, not to look at it, but to use it on the customer's behalf. If the agent cannot find what it needs, it moves on to the next shop.

That is why we work with the interface itself:

  • Clearly defined actions. Search, select a variant, add to cart and pay must be real actions, not elements that merely look like buttons.
  • Content that does not hide. Sizes, stock and delivery times must be available as text and data, not only in an image or behind a click.
  • WebMCP. A proposal from W3C's Web Machine Learning Community Group, where the site itself exposes its functions to agents. Chrome is testing it, but it is not yet a finished standard.

We make the site usable for both, without changing how it looks to humans.

What the customer sees

Dark green trail running shoe with lime green sole viewed from the side

Running / Running shoes / Trail

Trail running shoe X3

128 reviews

899 kr.

In stock. Delivery in 1 to 2 business days

What the agent can read

GET /loebesko/trail-loebesko-x3

Product
Trail running shoe X3
Price
899.00 DKK
Sizes
Not found. Buttons only, no data
Stock
Not found. Loaded with JavaScript
Delivery
Not found. Only shown in an icon
Add to cart
Not an action the agent can call

Result: The agent cannot see whether size 42 is in stock or will arrive by Friday. It chooses another shop.

Make the products readable so that a purchase is possible

An AI agent can only recommend and purchase what it can read. Therefore, product data, feeds, and product pages must convey consistent information regarding price, variants, stock, and delivery—and each variant must have its own unique ID.

Once this foundation is in place, the agent can move from making recommendations to executing purchases. This requires a checkout process that the agent can access, as well as terms for delivery, returns, and payment that it can read. These are the tasks we outline in the “How We Work” section below.

How we work

We start by assessing where your webshop stands. We then agree on which changes we implement and which tasks require your developers.

01

Agent readiness assessment

We investigate whether an agent can understand your products and what is missing to support a purchase. You receive: A prioritised assessment of the gaps with a recommended order of work.
02

Data layer and feed

We translate the data layer assessment into agreed corrections. Tasks are divided between us and your technical team. You receive: A concrete correction plan with ownership of each task and the agreed changes to the feed.
03

Website and structured data

We review the visible product text and the structured markup. We fix the agreed areas and describe changes that require development. You receive: Updated content and markup within the agreed scope, plus a task list for your developers.
04

Integration when available

When Universal Commerce Protocol (UCP) becomes available for your market and platform, we clarify the integration requirements. You receive: A plan for integrating and testing the purchase flow with a clear division of responsibilities.

What we can help you with

We prepare your products for AI agents, from the initial assessment to a checkout they can use.

Analysis and plan

Agent readiness assessment

Find out where your webshop stands.

We review product data, feed, product pages and checkout to see what an AI agent can read and what is missing.

Example in practice

A selection of your most important products is assessed, and you receive a prioritised list in the order we recommend.

Agent readiness · webshop.dk
Product dataSizes missing
FeedVariant ID differs
Product pagesDelivery only shown in an icon
PriceConsistent on the page and in the feed
CheckoutNo agent access yet

The findings become a prioritised plan.

Test in AI assistants

See yourselves as the agents see you.

We ask specific purchase questions in tools such as ChatGPT and Gemini and see whether your products are found and recommended.

Example in practice

A question about a running jacket shows that competitors are mentioned because your sizes cannot be read.

AI assistant · test
K
Test question

Waterproof running jacket in size M under 1.200 kr.?

AI
Answer

Mentions two competitors. Your jacket is missing because size and stock cannot be read.

We test with real purchase questions and see who gets selected.

Feed and product data

Product feed optimisation

Complete and correct product fields.

We complete and correct the fields in your product feed so titles, attributes, GTINs and variants are complete.

Example in practice

Missing colours and GTINs are added so each variant can be found and compared.

Feed · running jacket
TitleComplete
GTINMissing
SizeM, L, XL
ColourNot completed

The fields are completed and corrected at the source.

Merchant Center and catalogues

Feed and webshop based on the same data.

We set up and fix errors in Google Merchant Center and the Meta catalogue so the feed and webshop are based on the same data.

Example in practice

Variant IDs and prices are aligned so the feed matches the webshop.

ProductProblemStatus
Running jacket, MVariant ID does not matchError
Tights, LPrice differs from pageError
Running vestNo issuesOK

Diagnostics in Merchant Center and the Meta catalogue are resolved continuously.

Website and markup

Structured data

Markup that machines can read.

We implement or correct the markup so price, variants, stock and delivery can be read by machines.

Example in practice

Product and Offer markup is expanded with size, stock and delivery time.

# Product · JSON-LD price: 899.00 DKK availability: InStock size: 42 deliveryTime: 1 to 2 days

The markup must match what is visibly shown on the page.

Product pages for people and agents

The same page, readable by both.

We structure the content so sizes, stock and delivery are shown as text, and the buttons are real actions.

Example in practice

Delivery time that was only shown in an icon becomes text without changing how the page looks.

webshop.dk/loebejakke
SizesShown as text
StockVisible without clicking
DeliveryOnly shown in an icon
Add to cartReal button

The page looks the same to people and can be used by agents.

Checkout and operations

Price and stock alignment

The same figures in all systems.

We identify where price and stock are delayed or differ between your systems.

Example in practice

Stock shown as sold out in the feed but available in the webshop is corrected at the source.

SKUWebshop · feedStatus
JAK-M4 pcs. · 0 pcs.Differs
X3-4212 pcs. · 12 pcs.OK
TGT-L899 kr. · 949 kr.Price error

Discrepancies are typically why the agent chooses another shop.

Ready for agentic checkout

Prepare checkout for AI agents.

We map your checkout and clarify the requirements for e.g. Universal Commerce Protocol when it is available for your platform and market.

Example in practice

Delivery, return and payment terms are made readable as data so they are ready for integration.

Checkout · requirements
Delivery termsReadable as data
Right of returnOnly as PDF
PaymentMapped
UCPAwaiting platform and market

We prepare everything so the integration is ready when the protocol is.

Ongoing monitoring

Catch errors before they cost sales.

We monitor feed errors, discrepancies in price and stock, and missing fields.

Example in practice

New products without a size in the feed are caught and flagged before they are lost.

#webshop-data
S
Monitoringtoday

3 new products are missing a size in the feed, and 1 price differs from the page.

Errors are caught before they cost sales.

Is your business ready for the next generation of search and commerce?

Book a no-obligation meeting where we skip the sales pitch and focus on your data. We have an honest conversation about your potential and whether a partnership can genuinely make a difference to your bottom line.

Frequently Asked Questions about Agentic Commerce

Agentic AI is artificial intelligence that can independently plan and execute multiple steps in a task to reach a goal — as opposed to generative AI, which primarily answers one question or produces one piece of content at a time. An agentic AI can, for example, research, compare, and act on behalf of a user without being prompted at every individual step along the way. We use the term as an umbrella for both Agentic Browsing, where the agent finds and recommends, and Agentic Commerce, where the agent also completes the actual purchase.

Agentic Commerce is when an AI agent completes the actual purchase on behalf of the user — from product discovery to payment — instead of the user clicking through a cart and checkout themselves. This requires the store to provide structured data, inventory, pricing, and a checkout function that the agent can read and act on. The concept differs from Agentic Browsing, where the agent only researches and recommends without completing a transaction.

Universal Commerce Protocol (UCP) is an open standard from Google that enables an AI agent to complete a purchase directly within the AI interface, while the store remains Merchant of Record and continues to own the customer relationship. The protocol covers areas such as identity linking, checkout and order placement, and is transported via APIs, MCP and A2A — with AP2 for the actual payment.

WebMCP is a proposal under development in the W3C Web Machine Learning Community Group, aimed at enabling a webpage to expose its own features — such as search, cart and checkout — as tools that an AI agent can call directly. No browser currently supports WebMCP in production, so it is not an adopted standard but a direction still being explored.

Generative AI generates text, images or other content based on an instruction — the response is delivered and the task ends there. Agentic AI goes a step further: it plans multiple steps, makes decisions along the way and executes the task itself — for example by comparing products, selecting one and completing a purchase, without being prompted for each individual step. The two are not mutually exclusive — many agentic AI systems use generative AI as part of how they reason and formulate responses along the way.

AI Visibility is about whether and how a company's website, products and content are found, understood and recommended by AI systems — including AI agents, chatbots and AI overviews in search engines. Where traditional SEO is primarily about ranking in search results for humans, AI Visibility is about ensuring that data and content are structured and readable enough for an AI to correctly cite, compare and recommend the company. This requires, among other things, semantic structure, schema markup and machine-readable product data.