A customer no longer types "best running shoes wide feet" into Google and then opens seven tabs. He asks ChatGPT, gets three options with price and stock, and clicks buy. That conversation contains three brands. Not forty. If you are not among them, your product simply does not exist for that customer. That is agentic commerce: an AI assistant that does not just advise, but also compares, fills a cart and pays.
The question is no longer whether this channel becomes relevant. The question is whether your product data is already in it.
From exception to standard in four months
Evertune tracked how often ChatGPT shows a shopping experience instead of a wall of text when asked a commercial question. In October 2025 that happened in 8.3% of relevant answers. In February 2026 it was 87%.
From 8.3% to 87% in four months: shopping mode is no longer an experiment, it is the default view for purchase questions.
OpenAI pushed that line further in March 2026 by extending the Agentic Commerce Protocol to product discovery: visual product displays and comparisons inside ChatGPT itself. The assistant is no longer a referrer to your category page, it is a storefront that replaces your category page.
Agentic commerce delivers more valuable traffic than any other channel
The easy counterargument is volume: AI traffic is still small compared to organic search. That is true, but it is the wrong number to steer on.
Adobe measured in September 2026 that visitors arriving via an AI assistant generate 83% more revenue per visit and have a 90% higher conversion rate than non-AI visitors. Volume grew by 71% in that same year.
The why is at least as interesting. That same Adobe data shows AI visitors stay 73% longer on the site, view 25% more pages and add something to the cart 53% more often. That makes sense: the orientation work has already been done in the conversation. Whoever clicks through has already had the comparison and arrives with intent.
And the curve is not flattening. In July 2026, AI referral traffic to US retail sites was 62% higher than a year earlier, and 1,219% higher than in October 2024.
You do not get in by waiting for a crawler
This is where most webshops go wrong. SEO thinking assumes: we publish, they crawl, we appear. In agentic commerce that does not work. You supply a feed yourself.
ChatGPT accepts product feeds in CSV, TSV, XML or JSON and processes updates as often as every 15 minutes. That last point is not a detail: an assistant that recommends a product that is sold out or quotes the wrong price loses trust and will be slower to show you again. Stock and price discipline have become a visibility factor here.
The Agentic Commerce Protocol, maintained by OpenAI and Stripe, standardises four things: the product feed, checkout endpoints at the merchant, order webhooks and a delegated payment. Read that list again. Three of the four components are backend. Agentic commerce is therefore first and foremost a technical project, not a marketing campaign. Anyone who hands this to the content team without developers will get no further than an intention.
Google took its own route. The Universal Commerce Protocol update of March 2026 added cart and catalogue access, with onboarding via Merchant Center. At I/O 2026 Google added the Universal Cart on top of that: one shopping cart that works across Search, Gemini, YouTube and Gmail, combined with UCP and the AP2 payment protocol. Two ecosystems, two implementations. You cannot pick one and hope the other follows.
Riding along on free crawling is a shrinking channel
At the same time, the old route is getting more expensive. Since 15 September 2026, Cloudflare blocks AI crawlers on ad-supported pages by default and pays publishers via Pay Per Use when their content shapes an AI answer. The idea that assistants will simply keep reading the web and pick up your product pages by themselves does not hold. What remains are two routes you control yourself: a feed you actively supply, and brand mentions in places models do keep using.
That second route is underestimated. An assistant rarely chooses on feed data alone. It leans on reviews, comparison articles, forum threads and editorial pieces to decide which of the five matching products it names. That is exactly the territory where you can improve AI visibility without waiting for a protocol implementation.
The Netherlands: checkout is not possible yet, the recommendation is
An important nuance for Dutch webshops. Instant Checkout launched in September 2025 in the United States with Etsy, and Dutch merchants could not join at the time. So the payment side is still limited in availability here.
That is no reason to wait, because the order in a conversation is: first get mentioned, then get compared, then get paid. You can already influence steps one and two. Anyone who only starts when Instant Checkout goes live in the Netherlands begins with zero brand mentions against competitors who have been working on it for a year by then.
How to measure whether you are in there at all
You cannot steer this on gut feeling. The practical order:
- Define your purchase questions. Not your brand name, but the questions a customer without a brand preference asks: "which X is best for Y", "alternative to brand Z under 200 euros".
- Measure how often you are mentioned. Per assistant, because ChatGPT, Gemini and Claude draw on different sources and arrive at different shortlists. In our case study with three assistants the outcomes for the same brand differed considerably.
- Look at the sources behind the answer. Which review site, which comparison article, which forum is cited? That is your to-do list.
- Check your feed data. Are price, stock, variants, GTIN and shipping information correct? A feed with gaps gets ignored.
- Repeat monthly. One measurement is a snapshot, not a trend.
Steps 2 and 3 are barely sustainable by hand across dozens of prompts and three assistants. That is what measuring AI visibility is for: the same questions, repeated structurally, with the sources included. How we arrive at those scores is written out in the measurement method.
What you can do this month
Start with a baseline measurement on twenty purchase questions from your category. Put your developers to work on the feed side as well, because ACP and UCP both require work that takes weeks, not days. And meanwhile let marketing tackle the sources that assistants turn out to cite.
Shopping mode went from 8.3% to 87% in four months. That speed is the real argument: there is no room to wait and see.