AI shopping: what should your webshop do now?

The shortlist is made in AI

"Find black running shoes in size 42 under 900 DKK, deliverable by Friday."

That's the kind of question more and more customers are asking ChatGPT, Gemini or Copilot today. The answer isn't ten blue links — it's a short list of a few products. As SEO Tech Lead at Searchmind, I work with the technical foundation that determines whether a webshop makes it onto that list.

AI and e-commerce are getting a lot of attention right now, and it can be hard to distinguish between what works today and what's still on the drawing board. That's why I've broken it down into three parts: what you should do now, what's coming soon, and what can wait.

Now
Get found and understood

Crawler access, content in HTML, schema and a complete product feed.
Soon
Get used by agents

Buttons with proper names and a page that stays stable.
Later
Sell directly in AI

Available in the US. Not yet possible in Denmark.

AI shopping and agentic commerce are not the same thing

AI shopping is when a customer uses an AI assistant to research, compare and choose products. The purchase typically still takes place on the webshop.

Agentic commerce goes one step further. Here, an AI agent — a programme that can act on behalf of the user — carries out the actual tasks: navigating, filling in forms and, in some cases, completing the purchase.

What the numbers show — and what they don't

McKinsey has just published a study among approximately 21,000 consumers in 11 markets, including the US, the UK, Germany and France. It was conducted in June 2026.

53% have on average used AI for shopping within the past three months.
Almost one in three AI shoppers says that AI chat or AI search had the greatest influence on their most recent purchase.
In the US, the share of AI users who have purchased via AI rose from 10% to 17% in one year.
Consumers are using multiple AI tools side by side.

Two things are important to keep in mind. Denmark is not included, so the figures indicate the direction, not the Danish level. And the vast majority of usage is research and inspiration. AI already influences what customers choose, but it is still mostly the customer themselves who makes the purchase.

That is the reason for my prioritisation. What matters today is being found and understood. Being used by agents and selling directly within AI comes later.

Now: get found and understood

Before AI can recommend your products, it needs to be able to access your pages and understand them. Click through the three checks I always start with, shown here using a running shoe priced at 799 kr.

Can AI find your pages?

Check robots.txt. At OpenAI, search (OAI-SearchBot) and training (GPTBot) are controlled separately, so you can opt out of training and still appear in ChatGPT (OpenAI). The name and price must also be present in the page's HTML, rather than only appearing via JavaScript.

Can AI understand your products?

Structured data and a good product feed remove the guesswork. Microsoft recommends schema, while Google states that it is not a requirement. The most important thing is that price and stock are consistent on the page, in the feed and in the schema.

Three checks before AI can answer with your product

Access

Can the AI crawler access the page?

eksempelshoppen.dk/robots.txt

# Search in ChatGPT
User-agent: OAI-SearchBot
Allow: /

# Model training
User-agent: GPTBot
Disallow: /

Sitemap: https://eksempelshoppen.dk/sitemap.xml

Content in HTML

Are the name and price in the code before JavaScript runs?

view-source:eksempelshoppen.dk/loebesko/loebesko-sort

12 <link rel="canonical" href="https://eksempel…">
48 <h1 class="product-title">Black Running Shoes</h1>
49 <span class="price">799,00 kr.</span>
50 <s class="compare">999,00 kr.</s>

Consistent data everywhere

Do the page, feed and schema say the same thing?

Product page

799,00 kr.
● In stock

feed.xml

<g:price>799.00 DKK
<g:availability>in_stock

JSON-LD

"price": "799.00"
"availability": "InStock"

The answer

When the three checks are in order, AI can answer with your product.

AI assistant

Find black running shoes in size 42 under 800 kr.

I found one that matches:

Black Running Shoes

799 kr. ~~999 kr.~~

Eksempelshoppen · In stock in size 42

1 / 4

AI can only recommend what it can read. That is why everything starts with the technical foundation.

Alexander Achenza, SEO Tech Lead

Coming soon: Being used by agents

Agents that browse and click on behalf of the user already exist, including in ChatGPT. But even in the US, only a minority of AI users have made a purchase via AI. This is not where you should start. That said, much of the work also improves the experience for your customers today, so it is worth including when you are working on the site anyway.

Agents read, among other things, the page's accessibility tree: a structure where every button and every field has a role and a name. Google mentions it in its guide to AI optimisation, and OpenAI writes that ChatGPT's agent in Atlas uses ARIA tags. A button like is unambiguous. An empty

with an icon is not. This also helps customers who use screen readers.

Layout stability counts too. The video below is from Google's web.dev and shows a classic problem: a banner loads late, the buttons shift down, and the click hits "Yes" instead of "No". This hurts the user experience today, and an agent clicking based on screenshots can fall into the same trap.

New standards are also on the way. WebMCP is a proposal where the page itself describes its actions — e.g. "search", "select size" or "book appointment" — as tools that a browser agent can call directly. It is currently being tested in Chrome, and there is no official confirmation that ChatGPT, Gemini or Claude use it.7 The demo below shows the difference on the same booking widget with and without WebMCP.

Later: sell directly in AI

Protocols for actual agentic checkout already exist: ACP from OpenAI and Stripe, and UCP, which Google and Shopify support together with several others. But availability is limited.

That does not mean you should wait before doing anything. What prepares you for checkout later is the same thing that makes you visible now: clean product data, consistent price and stock everywhere, and a checkout you understand in detail.

You do not need to start with this

llms.txt and markdown versions of pages are often mentioned. I include llms.txt in my technical review because it is cheap to create. But Google states that Google Search ignores llms.txt, and that you do not need to create new machine-readable files to appear in AI features. Markdown can be useful for coding agents and documentation. Neither should be the first thing a webshop spends time on.

Buying directly in AI: where is it possible today?

As of 30.10.2026

ChatGPT

ACP · OpenAI and Stripe

Live in the US

Shopping in ChatGPT. Checkout within ChatGPT requires approval as a partner.

US
Denmark

AI Mode and Gemini

UCP · Google, Shopify and others

Selected retailers

Checkout for selected retailers and products.

US
Canada
Australia
Denmark

Denmark

Danish market

Not announced

Neither has been announced for Danish customers yet.

Do this nowKeep price, stock and delivery consistent on the product page, in the feed and in schema.

Sources: OpenAI, Google Merchant Center.

How to get started

Four checks you can do this week. The first three are about "now".

Check access

robots.txt, status 200 and content in raw HTML on product and category pages.

Follow one product

Are price, stock and delivery consistent on the page, in the feed and in schema?

Ask AI yourself

Ask your customers' questions in ChatGPT and Gemini, and see what comes up.

Factor in agents

Do buttons and fields have names, and does the page remain still during loading?

How we help

If you want to go deeper into the technical side, I have written a walkthrough of the disciplines I measure, in an article on LinkedIn. At Searchmind, we work across all three stages, from access and data to action.

We test in

ChatGPT

Gemini

Copilot

Claude

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

Now, soon and later

Agentic Commerce

We make your webshop readable and usable for AI agents: from assessment and planning to feed, website and checkout when UCP becomes available in your market.

See the service →

Searchmind team at screens in the office

Now · Get found

GEO & AEO

Visibility in AI answers: schema and data architecture, crawler access and ongoing measurement of how AI mentions your brand.

See the service →

Now · Get understood

Prism

Our tool for product feeds. AI fills in missing attributes based on your own data, and each channel gets its own feed, including Google Shopping, Meta and ChatGPT.

Test a product →

Now · The foundation

SEO

Technical SEO, content and analysis: the foundation for being found, both in traditional search and in AI.

See the service →

Where does your webshop stand today?

We've built a tool that gives you a first answer. Enter your homepage, pick a category and a product page, and see how well AI agents can read and understand your website. Rather talk it through right away? Book a no-obligation meeting.