Agentic commerce: when AI agents shop for the customer, and what Swedish online retailers should do in 2026

Imagine a potential customer asking ChatGPT to find a winter jacket for under SEK 2,000, and the agent comparing the range, choosing a store and completing the purchase without the customer ever visiting a website. That is the promise. The reality in 2026 is considerably more nuanced than the headlines, and that nuance determines what you should actually spend your time on as an online retailer.

What is agentic commerce?

Agentic commerce means an AI agent carries out all or part of a purchase on the user's behalf: it searches, compares, chooses and, in some cases, completes the payment too. What sets it apart from ordinary AI search, often called agentic search, is that the agent does not just recommend. It acts.

In practice, agentic commerce has two parts. First, the agents themselves: services such as ChatGPT, Google AI Mode and Perplexity. Second, the protocols that let agents read product data and complete transactions with merchants in a standardised way. The protocols decide which stores an agent can shop at – and that is where today's tug-of-war is taking place.

The protocols behind it: ACP and UCP

Two protocols dominate the discussion right now.

ACP – Agentic Commerce Protocol

ACP was developed by OpenAI and Stripe and launched on 29 September 2025 alongside Instant Checkout in ChatGPT – buying directly in the chat. It started in the US only, with Etsy as the first merchant and Shopify promised next, according to Stripe. The protocol is open source under the Apache 2.0 licence and is maintained jointly by OpenAI and Stripe on GitHub.

The ecosystem grew quickly. PayPal joined as a payment partner on 28 October 2025, according to TechCrunch. In December 2025, Stripe launched its Agentic Commerce Suite with Wix, WooCommerce, BigCommerce, Squarespace and commercetools as the first partners, according to Stripe.

UCP – Universal Commerce Protocol

UCP is Google and Shopify's answer, with Etsy, Wayfair, Target and Walmart as co-founders and some twenty supporting companies – among them Mastercard, Visa, Zalando and, interestingly, Stripe. Google CEO Sundar Pichai presented the protocol at the NRF retail trade show on 11 January 2026, according to Google.

There are more protocols on the fringes, such as Google's payment protocol AP2 and Klarna's equivalent. For a Swedish store, what matters today is not picking the right protocol. It is understanding what they all have in common: they require structured, machine-readable product data.

What happened to buying directly in ChatGPT?

This is the part of the story that many guides miss, because they were written before March 2026.

In March 2026, OpenAI stopped allowing purchases directly in the chat. OpenAI confirmed in early March that Instant Checkout was moving to apps, according to Digital Commerce 360, and CNBC reported on 20 March that the initiative had run into problems. OpenAI moved instead to an app-based model in which merchants such as Instacart, Target and Expedia build their own apps inside ChatGPT, and presented the new shopping experience on 24 March, according to CNBC. OpenAI's own explanation was that the first version "did not offer the level of flexibility that we aspire to provide".

The numbers behind the decision are telling. Walmart's data showed that purchases through the ChatGPT checkout converted three times worse than on its own website, according to Search Engine Land. Customers want to see images, compare variants and recognise the checkout; a text chat is a poor place to buy most things other than repeat purchases of standard goods.

The takeaway for online retailers: the technology is still experimental, and the buying experience is currently moving back towards the merchant's own environment. That is good news. Your website is not becoming redundant – it is becoming the destination the agent sends the customer to.

When is agentic commerce coming to Swedish e-commerce?

Short answer: none of the transactional features is available in Sweden at the time of writing, and no company has announced a confirmed EU plan.

Where each service stood in August 2026 – a snapshot of a landscape that has been redrawn twice in six months:

• Google: agentic purchases have been live with selected US merchants since Google Marketing Live on 20 May 2026. Google says it wants to "expand globally" but gives no timeline, according to Google.
• Google AI Mode: available in Swedish since 8 October 2025, according to Google, but the buying feature in AI Mode is limited to the US.
• Perplexity: "Buy with Pro" launched on 18 November 2024, US only, with no transaction fee for merchants, according to Perplexity.
• ChatGPT: direct purchases in the chat have not been live anywhere since the change of course in March 2026.

But – and this is the point – the discovery phase is already here. Swedish Shopify merchants' products can already appear in ChatGPT answers, because Shopify sends product data to OpenAI. And traffic from AI services is growing fast: according to Adobe, AI-referred traffic to US e-commerce sites rose by 62 per cent in the year to July 2026, an increase of 1,219 per cent compared with October 2024. The same data shows that AI traffic converts 60 per cent better and generates 53 per cent more revenue per visit. Note that this is US data – there is no Swedish equivalent yet – but the direction is hard to dismiss.

So agents are not shopping for Swedish customers today. But they are already recommending stores to them. That is where the competition will be decided in 2026.

The AEO connection: the same job you should already be doing

This is the most important point in the whole subject: agentic commerce is not a new discipline. It is AEO applied to transactions.

Look at what the protocols actually require of a store: correctly structured product data, schema markup with price and stock status, a well-kept product feed and clear terms in a machine-readable format. That is exactly the groundwork you need to be visible in AI Overviews and AI assistants' answers – and largely the same foundation that classic search engine optimisation rests on. We cover that layer in more detail in our guide to AI Overviews and GEO and in how SEO is affected by AI.

There is also a sharper version of the argument that applies to the moment of purchase. An AI agent that cannot verify your price, your stock status or your returns policy in a structured format does not say "let me get back to you". It silently excludes your store and recommends a competitor whose data it can read. That is not a loss of visibility that shows up in any report – it is a lost sale that never leaves a trace.

So the recommendation is simple: you do not need a new strategy for agentic commerce. You need to cover the fundamentals you should already have in place, and do them properly – they hold whichever protocol or agent ends up winning.

We have gathered all of that groundwork for an online store – category structure, unique product copy, structured data and the technical pitfalls along the way – in our guide to e-commerce SEO.

Do it now or wait?

Not everything is worth doing today. Here is how the trade-off looks for a Swedish store on Shopify or Webflow:

Decision matrix: what a Swedish store should do now – and what can wait (status August 2026)

Action	Effort	Benefit today	Recommendation
Complete, structured product data	Low–medium	Immediate – strengthens SEO, AEO and feeds	Do now
Schema markup on product pages	Low	Immediate – readable by both Google and agents	Do now
Google Merchant Center feed	Low–medium	Immediate – Shopping, AI Mode and future agent flows	Do now
Returns and delivery terms in plain text on the website	Low	Immediate – agents factor in terms when they recommend	Do now
Your own ACP/UCP integration	High	None – no protocol is live for Swedish purchases	Wait
Your own app in ChatGPT	High	Unclear – the model launched in spring 2026 and is US-focused	Wait and watch

The pattern is clear: everything in the "do now" column pays off however agentic commerce develops, because it strengthens your ordinary visibility today. Nothing in the "wait" column has a proven return for a Swedish store yet. Do not panic-invest in integrations.

Checklist: how to prepare your store

Concrete steps, with a rough time estimate for a small to medium-sized catalogue:

1. Clean up your product data (2–8 hours depending on catalogue size). Complete product titles, GTIN/EAN, brand, size and colour as separate attributes – not baked into free text. Unique descriptions, not the supplier's standard copy.
2. Verify your schema markup (30–60 minutes). Product pages should have Product schema with price, currency and stock status. Shopify themes often have this built in – do not take that on trust, test it in Google's Rich Results Test. If you build on Webflow Ecommerce, structured data often has to be added through custom code – check that it is there and correct.
3. Set up or clean up your Merchant Center feed (half a day the first time). On Shopify via the Google & YouTube channel, on Webflow via a feed tool. Rejected products and missing attributes mean invisible products – to agents too.
4. Publish returns and delivery terms in plain text (1–2 hours). A dedicated, indexable page on the website – not just a line in the checkout or a PDF. An agent that cannot find the returns policy cannot factor it in, and may pass you over as a result.
5. Check robots.txt and rendering (30 minutes). Do not block AI services' crawlers, and make sure price and stock status are not rendered solely by scripts that crawlers miss.
6. Watch – but do not integrate yet (ongoing, costs nothing). Keep a particular eye on Shopify's UCP work: if your store is on Shopify, protocol support will most likely come through the platform, not from your own development work.

In total: roughly one to two working days for the fundamentals. That is the whole investment required in 2026 – and the work pays for itself in ordinary SEO and AEO before any agent completes its first Swedish purchase.

Want to know where your store stands? We review product data, schema and feeds as part of our AEO work – get in touch and we will look at it together.