Agentic commerce: when AI does the shopping, will it find your store?

Ask ChatGPT to find a good sunscreen for sensitive skin and it will name products, compare them and point to stores where you can buy. Try it with something you sell. The uncomfortable question: is your store in that answer, or does the AI not even know you exist?

The shift under way

For twenty years, being findable online meant one thing: rank in Google, get the click, close the sale on your own site. That funnel is quietly growing a bypass. More and more buying journeys start inside a chat window, where an AI reads the web for you and comes back with a short answer instead of ten blue links.

And the next step is already being built at full speed: AI that doesn't just recommend, but acts. In the industry this is called agentic commerce: an AI agent that researches, compares and eventually orders on behalf of the customer. As AI strategist Paul Krauss puts it in an interview with Shopware: "In agentic commerce, the machine acts on behalf of the customer; the human only formulates the intent."

That makes AI assistants less like a search engine and more like a new sales channel, the way marketplaces were fifteen years ago. Shopware is building dedicated tooling for it: product feeds into ChatGPT, Gemini, Perplexity and PayPal's shopping app, plus support for the emerging standard protocols that let agents browse and buy. According to Shopware's numbers, AI-driven traffic to merchant stores grew 15x comparing early 2026 with a year earlier. You can argue with any single stat, but the overall trend is very clear.

GEO vs SEO: same goal, different reader

And that's where GEO comes in, Generative Engine Optimization: shaping your store's content and data so AI answer engines can find it, understand it and cite it. Not a replacement for SEO, but complementary to it.

There's a very large overlap with classic SEO. Crawlability, clean structure and structured data help both. But the "reader" changed, and that changes the job:

• SEO optimizes for a ranking algorithm that shows a list; a human clicks and judges your site with their own eyes.
• GEO optimizes for a language model that reads your actual content, synthesizes one answer, and cites maybe two or three sources.

Krauss is blunt about the gap: "Those who used to do SEO are, at best, only half-visible in AI Overviews." Ranking third in Google used to mean traffic. In a synthesized AI answer there is no third place: it's all or nothing.

What an agentic readiness check actually looks at

"Being ready for AI" sounds vague. When we assess a store, these are the checks that decide whether an AI agent can do anything with it:

• Can AI crawlers get in at all? Robots.txt rules and bot protection regularly block GPTBot, ClaudeBot or PerplexityBot, sometimes deliberately, often as a leftover default from a CDN or firewall set up in earlier days. If the crawler is blocked at the door, nothing else matters.
• Does your content exist without JavaScript? Many AI crawlers read the raw HTML and don't execute scripts. If your prices, descriptions and availability only appear after JavaScript runs, an AI agent sometimes sees an empty shell.
• Structured data. Product, Offer, price, availability, reviews in schema.org markup: this is the difference between an AI parsing your catalogue reliably and guessing at it. The same markup that powers rich results in Google, double duty, and hopefully something you've had in order for years already.
• Product content that answers questions. Agents match intent ("sensitive skin", "fits a 2019 model") to your descriptions. Plain-language specs and honest answers beat marketing copy that says nothing.
• Feeds and the technology behind it. A current sitemap, a clean product feed, and recently also the new agent protocols. If you're on Shopware, part of this arrives as platform features; it still needs your data to be in order. For projects that have been running a while, a check-up is always worth it: updates or heavy custom styling work on your webshop can make these standard features quietly disappear.

Notice how much of this list overlaps with the technical SEO checklist we run in audits. That's the good news: solid technical foundations pay out twice.

What a real store scores today

Theory is cheap, so we ran the Agentic Readiness Scanner, a free tool from Shopware's Agentic Commerce Alliance, against a real Belgian webshop. One with its technical SEO in decent shape, the kind of store that does well in classic Google and still gets the majority of its traffic there today.

The score: 51 out of 100.

The interesting part is the split. Almost everything that overlaps with classic SEO passed: robots.txt allows the major AI crawlers in, the sitemap lists all products, canonicals are clean, breadcrumbs are marked up, and product pages carry Product markup with prices and GTINs.

Almost everything agent-specific failed:

• No product feed anywhere an agent could discover it. The scanner weighs this as the single biggest gap: five separate checks, all red.
• No llms.txt, a plain markdown index of your site written for language models, the AI equivalent of a sitemap.
• None of the agent protocols. No AGENTS.md, no agent card, no UCP endpoint (the Universal Commerce Protocol that lets agents browse and buy in a standard way).
• No review or FAQ markup, and the product schema is served as older microdata instead of the JSON-LD that AI parsers prefer.
• Policies invisible to machines. Shipping, returns and privacy pages the scanner couldn't find, and no machine-readable statement of what agents may do on the site.

And this is the story we see on a lot of webshops: the SEO half of the work is done, the agent half doesn't exist yet. We'd bet most stores look exactly like this. Which also means that whoever gets moving early now will stand out disproportionately.

The honest hype check

In all honesty though: take all of this with a grain of salt. Shopware's own hype-vs-reality piece cites research that 95% of organisations haven't achieved measurable ROI from generative AI initiatives. Most "AI strategies" fail because they start with technology instead of a business problem. Nobody should rebuild their store this quarter because a keynote said agents are coming.

But that's exactly why readiness is the right frame. Krauss argues merchants have two viable paths: be so relevant that customers ask for your brand by name, or be so efficient, with flawless data and clean interfaces, that agents prefer you. His warning for the middle: "Everything in between will be sorted out by the algorithm and the customer alike."

Running the readiness check and seeing whether AI can read your store is simple and cheap. Being unreadable while your competitor is already cited in every answer leaves money on the table.

What you can do right now

You can check the basics yourself, today:

• Run your store through the free Agentic Readiness Scanner and look at the score split, not just the number.
• Open your robots.txt and your CDN's bot settings; see who you're blocking.
• View your product page source with JavaScript off. Is the price there?
• Run a product URL through a structured data validator.
• And the simplest test: ask ChatGPT or Perplexity what it knows about your store and your products. Read the answer as if it were your storefront, because for a growing group of buyers, it is.

A scanner score tells you where you stand. It doesn't tell you which of those red rows actually matter for your store, what they cost to fix in your Shopware store, or in what order. That's exactly what the AI-readiness part of our SEO & GEO Audit covers: we check your store the way an AI agent sees it, from crawler access to structured data to how your products actually show up in AI answers, and hand you a prioritised report with a walkthrough call.

Have you asked an AI assistant about your own products lately? What did it say? I'm genuinely curious which stores are already showing up. Is there anything we can help you with? Let me know!