Your Shopware shop is optimised for Google. However, it is practically invisible to the entity that is increasingly acting as an intermediary between your customers and Google.
This is not just a pipe dream. Google AI Overviews appear in around a quarter to just under half of all search queries, depending on the study; BrightEdge estimates this figure will reach almost 50 per cent by early 2026. They answer the question directly, often without the user even needing to click. In addition, millions of people are asking ChatGPT, Perplexity, Gemini and Claude for product recommendations, comparisons and buying advice. These systems no longer automatically direct the user to your product page. They provide an answer whilst citing a few sources. Whoever features on this list of sources wins. Those who aren’t included simply don’t exist for that customer.
And here’s the awkward bit: the list of sources cited by AI isn’t the same as your Google rankings. An analysis published in May 2026 by 5WPR, using data from the GEO provider Brandlight, shows that the overlap between the top Google results and the sources cited by AI response engines has fallen from around 70 per cent to less than 20 per cent. In other words: you can be on page one of Google and still not appear in the AI response. The SEO you’ve spent years building is no guarantee of AI visibility.
GEO isn’t a second SEO. But it isn’t a planet of its own either
First, the good news: you don’t have to throw your SEO away. The bad news: it’s still not enough. Clean technical implementation, fast loading times and relevant content benefit both worlds, and Google itself makes it clear that optimising for AI search is, at its core, still SEO, because AI Overviews are based on the same ranking systems.
The difference lies in the goal. SEO optimises for the click: you want someone to click on your search result. GEO optimises for citation: you want a language model to include your content in its response as a reliable source. This shifts the requirements. An AI can only cite your product page accurately if it clearly understands the content, classifies it as trustworthy and can extract individual sections from their context without the meaning being lost. In an e-commerce context, this is precisely where the work lies. We’ve described how to check, in general terms, whether AI can find your page at all in the article “Can AI even find your website?”. This article goes one step further and asks: How do you integrate this natively and permanently into Shopware?
What Shopware already offers, and where the gap begins
Shopware isn’t resting on its laurels. Since version 6.7.9, the core has included native JSON-LD output, controlled via the feature flag JSON_LD_DATA. When active, the storefront delivers structured data as clean <script type="application/ld+json"> in the <head> instead of the old microdata scattered throughout the HTML. The core thus covers the basics: an Organisation schema with the shop logo, a WebSite schema with SearchAction, ItemList on category and search pages, the product data, and – provided the data has been maintained – also FAQPage and OfferShippingDetails. The old microdata is considered obsolete and will be phased out with Shopware 6.8.
This is an important step, but it is the baseline, not the end goal. Two gaps remain. Firstly, the Core does not natively provide certain schema types that a response engine would like to utilise: HowTo for user and assembly instructions, and MerchantReturnPolicy for returns policies, along with authoritative brand references via sameAs. Secondly, and this is often underestimated: a schema type does not exist simply because the flag is set. An FAQPage without any questions stored in it remains empty, and a GTIN with an incorrect check digit does more harm than good. So it’s not just a matter of implementing more types, but of populating the existing ones completely and correctly. These are precisely the fields that enable a model to uniquely identify a product and trust it.
In short: from version 6.7.9 onwards, Shopware provides the foundation. You can build the GEO-relevant layers above that yourself, or have them built for you.
The four levers on which GEO in the shop really depends
GEO in e-commerce breaks down into four areas, which only come together to form a complete picture. I’ll briefly run through them from the top before we move on to the architecture.
The first lever is structured data. This is the language your shop uses to explain to a machine what it’s doing. If it’s missing or incomplete, the AI will make an educated guess about your price, availability and returns policy – and when in doubt, it’ll get it wrong. It’s important to supplement Shopware’s native output rather than duplicate it: Duplicate or conflicting schema nodes are worse than a gap.
The second lever is often overlooked because it looks like simple robots.txt maintenance: crawler control. AI bots visit your shop long before they can index it. Using user agents such as GPTBot, OAI-SearchBot, PerplexityBot and ClaudeBot, these response engines collect the content they will later index. If you block them across the board, you forfeit any chance of being cited; if you let them in unchecked, you lose track of things. In between lies a conscious decision that belongs within the shop, including a llms.txt, which gives AI systems a curated entry point into the relevant catalogue. (Google Extended is a special case here: it controls whether Google is allowed to use your content for Gemini and AI training, not your search ranking.)
Then there is the citability of the content itself. A product description that begins with ‘This product impresses with its top-quality’ is worthless to an AI: no fact, no figure, no independent core statement. A section becomes quotable when it identifies its own topic, contains specific information and remains accurate even when taken out of context. This can be measured and thus specifically improved.
That leaves the fourth point: measurement. Without measurement, GEO is just gut feeling. You need a repeatable score for each product and category, a history showing whether things are improving, and a list of priorities indicating where the next action will yield the greatest benefit.
The actual architecture decision: Deterministic rather than an LLM per product
Now it gets technical, and this is where the decision lies that determines whether a GEO tool is suitable or unsuitable for a real shop.
The obvious approach would be to run a language model for each product: “Rate this description, suggest a better one.” This works wonderfully in a demo with twelve products. In a catalogue with 50,000 items, however, this becomes a cost and latency issue that nobody wants to deal with. Every audit run costs money in API fees, and an LLM call has no place in a storefront request anyway.
That is why our approach is clear: the GEO evaluation runs as deterministic PHP logic, not as runtime AI. The heuristics that a language model would otherwise apply ‘in its head’ can be mapped as transparent rules. Does the text begin with a proper definition? Does the paragraph name its subject itself, rather than just using ‘it’ and ‘this’? How many specific figures, measurements or sources are there per 500 words? Is the heading hierarchy clear? This results in a composite GEO score across six categories: citability, brand, E-E-A-T, technology, schema and platform.
Put simply: you calculate the score once at your leisure and see it applied millions of times across the shop, not the other way round. The advantage isn’t just the price. A deterministic score is reproducible: the same content yields the same value today and tomorrow, and if the value rises, you know exactly which change caused the increase. An AI assessment fluctuates; a rule does not. And the heavy lifting happens where it should – in the Shopware Message Queue and in scheduled tasks – not with a single click from the customer. Ultimately, the storefront simply displays a pre-calculated result. Anyone who takes GEO seriously wants precisely these characteristics: predictable costs, stable figures, and zero performance risk in the shop.
GEO Booster: What we’re currently building
A word of honesty before you ask: GEO Booster is under development; it’s not a finished product in the store. We’re deliberately publishing this approach now because the architectural decisions – deterministic, channel-specific and queue-based – will still apply even if you implement GEO for your Shopware shop yourself or with another partner.
Based on the analysis above, we’re developing a plugin that brings GEO into the shop not as a one-off agency audit, but as a permanently active feature: four modules on a shared service layer, aligned with the four levers.
The schema enrichment supplements the types that the core does not provide by default (HowTo, MerchantReturnPolicy), and ensures that the existing FAQPage, OfferShippingDetails, and the product identifiers are correctly populated and rendered as valid data, including a checksum verification for the GTIN. If the plugin detects that the native flag is active, it simply supplements what is missing rather than duplicating nodes.
The llms.txt and crawler management generates an llms.txt for each sales channel from your catalogue, turning the control of known AI bots into a conscious, documented decision rather than a hidden robots.txtentry, including a warning if a crawler relevant to citability is currently blocked.
The Citability and Content Scoring system evaluates every product and category deterministically and provides a specific recommendation for action for each gap, rather than simply assigning a score. The audit and reporting dashboard brings everything together: an overall score, a breakdown by category, the ‘quick wins’ with the greatest impact, and a trend graph. Everything can be configured by sales channel right from the start, as a B2B channel has different geographical requirements to an international B2C shop.
What you should do now
You don’t need to wait for a plugin to get started. Three steps will make an immediate difference. Firstly, check whether Shopware’s native JSON_LD_DATA flag is available and active in your version, and whether your product pages return valid JSON-LD. Secondly, check your robots.txt to see if you’re unintentionally blocking AI crawlers such as GPTBot or PerplexityBot – the quickest way to do this is by shop domain in the AI Visibility Check. Thirdly, take a look at your ten best-selling products and read the descriptions through the eyes of a machine: does the opening line contain an independent, factual statement or a marketing cliché?
If you want to know how citable your shop is today, our free AI Visibility Check is the practical place to start. And if you want to set up GEO properly as part of your Shopware strategy, rather than tacking it on as an afterthought, that’s exactly the kind of work we do as a Shopware agency.
If you don’t want to tackle this on your own: the process and price range are outlined on the website of our GEO agency based in North Rhine-Westphalia.
The AI response engine will provide your customers with an answer anyway. The only question that remains is whether your shop features in it.