Generative engine optimisation

A customer asks Claude: “I’m looking for an office chair under €600 for long working days, I often have back pain. Which one is right, and where can I get it quickly?” She doesn’t get a list of ten links, but a finished answer with three models and two stores. Whoever is in that answer has the customer. Whoever is missing does not exist for her at that moment.

Generative Engine Optimization refers to optimising content for search systems that generate their answers with large language models. These include ChatGPT, Claude, Gemini, Perplexity and Microsoft Copilot, as well as AI Overviews and AI Mode in Google Search. The term goes back to a 2023 research paper involving Princeton University, among others. It examined which properties of content increase the chance of being included in generated answers. Related terms are answer engine optimisation (AEO), LLM optimisation or AI SEO. At its core, they mean the same thing.

How an AI answer is created

To understand GEO, you need to know what happens behind an answer:

• The question is broken down: The system turns one long question into several sub-queries, such as “ergonomic office chair test”, “office chair back pain” and “office chair under 600 euros”. Experts call this query fan-out.

• The index is searched: These sub-queries run against a search index, at Google against its own, at other providers against Bing, among others.

• Passages are selected: From the pages found, the system takes individual sections that answer a sub-question as directly as possible.

• The answer is written: Only then does the language model formulate the text and attach the sources used.

Two things follow from this that decide success or failure. First, classic search remains the foundation: whatever is not in the index cannot be cited. Second, it is not whole pages that are evaluated, but sections. A paragraph that answers a question precisely beats a long text in which the answer is hidden in the third subsection.

SEO and GEO: what stays, what changes

GEO does not replace search engine optimisation; it builds on it. Technical accessibility, clean structure and trust remain the prerequisites. What is new is the weighting:

• Being named instead of being clicked: Success means being part of the answer, even if the customer only comes to the store later or via the brand.

• Sections instead of pages: Every important paragraph must be understandable on its own and fully answer a question.

• Questions instead of keywords: Content is aligned more closely with the complete questions customers ask. More on this under search queries.

• Reputation instead of self-description: Language models derive a brand’s credibility from many mentions across the web, not just from its own website.

What AI systems say about your store

GEO has a second side that many merchants underestimate. AI assistants talk about your store, whether you steer it or not. They name outdated prices, recommend a discontinued model or send the customer to a competitor because its information is clearer.

The answer itself cannot be edited. Only the sources it is created from can be influenced: your own website, your product data and what others write about you. That is why GEO starts with an honest assessment: what do the assistants answer today to your customers’ questions, and who is named there?

The levers of Generative Engine Optimization

• Citable answers: Every important question is answered briefly and completely in a clearly recognisable place. Short definitions at the beginning of a section, meaningful subheadings, lists and question-and-answer blocks make it easier for the system to find the right passage.

• Evidence instead of claims: According to the original GEO study, concrete figures, dimensions, sources and expert statements significantly increase the chance of being named. Advertising language such as “high-quality” or “innovative”, on the other hand, hardly helps.

• Clear entities: Brand, products and contact persons are named and described the same way everywhere. Structured data make relationships machine-readable, and entries in knowledge bases such as Wikidata confirm them. More on this under entity SEO.

• Presence on third-party sites: Specialist portals, review platforms, comparison sites and industry directories give language models confirmation from outside. Recognisable expertise in the sense of E-E-A-T counts double here.

• Technical access: The AI providers’ crawlers must be allowed to retrieve your pages. It is worth taking a close look at the robots.txt, because many providers distinguish between crawlers for training and crawlers for search, at OpenAI for example GPTBot and OAI-SearchBot. Block everything across the board, and you disappear from the answers. Important content also belongs in the delivered HTML and not only in scripts loaded later. Which AI crawlers your robots.txt allows is shown by the robots.txt Checker.

• Freshness: AI systems prefer current sources. Visible dates, a recognisable expert reviewer and regular revisions are therefore more important than before.

GEO in an online store

For online stores, part of the buying advice is moving into the chat. Customers ask about the right product for a use case, about comparisons and experiences, and get a recommendation before they visit a store. The foundation for this lies in your data:

• Complete product data: Material, dimensions, weight, compatibility, area of use. Language models evaluate exactly these details when comparing products.

• A well-maintained product feed: It feeds not only Google Shopping, but increasingly AI systems as well.

• Category pages that answer questions: Which chair for which body height, which shoe for which terrain. This turns the category into a citable source.

• Reviews with substance: Real customer experiences flow into the answers.

The next step is already visible: AI agents that not only search for customers but also compare and buy. Whether your store is readable for them is checked by the free Agentic Commerce Scanner. How well your store is prepared for AI search overall is shown by the AI Readiness Check.

Measuring success

Classic ranking tools do not capture AI answers. That is why measurement relies on regular prompt monitoring: a fixed set of customer questions is put to the relevant AI systems again and again, and the answers are evaluated for whether, how often and in what context your brand is named. That is exactly what arboro.PULSE does. In addition, web analytics show the visitors who come to your store via chatgpt.com, claude.ai, perplexity.ai or gemini.google.com, and rising search volume for your store name is often an indirect sign of growing mentions.

Common misconceptions

• “GEO replaces SEO.” No. Without visibility in the search index, there is no mention in the AI answer either.

• “An llms.txt file is enough.” The file can’t hurt, but so far none of the major providers demonstrably uses it as a basis for its answers.

• “Blocking AI crawlers protects my content.” Above all, it protects you from being recommended.

• “This only affects big brands.” Especially for concrete, specific questions, AI systems often name smaller specialist retailers when they answer the question best.

How GEO is implemented in practice for your store is shown on GEO for online stores.

The results page used to decide who gets found. Today the answer decides who gets recommended. A store the AI doesn’t know loses customers before they even know it exists.