- In its guide from July 2026, Google states explicitly: Optimization for generative AI search is still SEO.
- The AI features take their sources from the Google index and expand a question with related search queries. A page must be indexed and allowed to appear with a snippet.
- According to Google, there is no need for an llms.txt, special markup, texts broken into small sections or artificial mentions.
- What is needed is content based on your own experience that an AI cannot assemble from general knowledge.
Generative AI features such as AI Overviews and AI Mode are now an established part of Google Search. Many website owners are therefore wondering whether classic SEO is losing importance.
Google has published an official guide on this, most recently updated in July 2026. The most important statement up front: SEO remains the foundation; it only shifts its focus. In this article, we summarize what Google says and what it means for your website.
Google creates AI answers from its own index
Anyone who wants to optimize for AI search has to understand how Google generates the answers. According to the guide, the generative AI features rely on the same ranking and quality systems as regular search. Google explicitly describes two methods.
The AI retrieves current sources from the index
The AI does not guess. With retrieval-augmented generation (RAG), which Google also calls grounding, the system retrieves current pages from the Google index for a question and bases the answer on them. The pages the answer is based on appear as links.
The AI expands the question with related search queries
With so-called query fan-out, the model generates additional, related search queries to find more relevant results. Google's example: The question of how to fix a lawn full of weeds can lead to searches for weed killers, for weed removal without chemicals and for prevention.
For SEO, this means: A good position for the main search query remains important. But pages that answer one of the sub-questions particularly well also have a chance.
The language model formulates the answer
From the content found, a language model such as Gemini formulates a new, fluently readable answer. It summarizes, compares and uses links to refer to the pages that support the answer.
Google says clearly: Optimization for AI search is SEO
Many new terms are circulating in the marketing world, such as AEO (Answer Engine Optimization) or GEO (Generative Engine Optimization). Google classifies them soberly: Optimization for generative AI search is optimization for search and therefore still SEO. You should therefore evaluate AEO and GEO service providers by the same standards as any SEO service provider.
According to Google, you can ignore many of the tips going around:
| Assumption | What Google says |
|---|---|
| An llms.txt file or special AI markup is mandatory. | There is no need for new text files, special markup or Markdown for AI. However, such files for other services do not harm visibility in search either. |
| Content must be broken into tiny sections. | Not necessary. There is no ideal page length; pages are made for readers. |
| Texts must be rewritten for AI systems. | Not necessary. The systems understand synonyms and meaning. |
| Artificial mentions on the web improve visibility. | The ranking systems favor quality; the spam systems block manipulation. |
| Structured data is the key to AI search. | It is not required for generative AI search. It remains useful for rich results. |
Google also points out that no external tool has access to the internal ranking or AI systems. You should be skeptical of promises based on that.
Content based on your own experience gets you into AI answers
Anyone who wants to appear in AI answers needs content that an AI cannot easily generate itself. Google speaks of “non-commodity content,” meaning content that goes beyond interchangeable general knowledge.
Google formulates the litmus test as a question you should ask yourself for every page: Would your visitors find this content satisfying? If you can honestly answer yes, you are on the right track.
Your own perspectives
A general guide such as “7 tips for buying your first home” is based on general knowledge that an AI can summarize itself. What is needed is first-hand experience. Google's counterexample is an article about why someone skipped the home inspection and saved money in the process, with a look into the sewer line. No AI can assemble such reports from other sources.
Clear structure
Write for people. Paragraphs, clear subheadings and a logical structure help your readers, and ultimately also the AI systems, to understand your page correctly.
High-quality images and videos
Besides text links, AI search also includes images and videos. So complement your texts with relevant, high-quality images and videos and optimize them according to the rules of image and video SEO.
Quality over quantity
It is tempting to create a separate page for every variant of a search query. Google explicitly warns against this: Anyone who mass-produces pages primarily to influence rankings or AI answers violates the spam policy on scaled content abuse. More pages do not make a website higher quality. Content created with AI assistance must meet the same requirements as all other content.
Clean technology remains the prerequisite
For Google to use your content for AI answers, the technical foundation has to be right. The familiar SEO fundamentals remain valid.
- Ensure indexing: Only pages that are indexed and allowed to appear with a snippet are eligible for the AI features. In addition, your website must be eligible for the generative AI features of Search in Search Console. Even then, there is no guarantee of being shown.
- Enable crawling: Keep your site clean, avoid duplicate content and, for very large websites, pay attention to crawl budget.
- Keep JavaScript under control: If you use modern JavaScript frameworks, you should follow the recommendations for JavaScript SEO. Many other AI crawlers do not execute JavaScript at all; more on this in our article on the JavaScript problem of AI crawlers.
- Use semantic HTML: It helps readability and assistive tools such as screen readers. According to Google, however, perfect code is not necessary.
- Maintain shop and location data: AI answers also show products and local businesses. Feeds in Merchant Center and the Google Business Profile ensure that your products and services appear there correctly.
You measure your visibility in AI search in Search Console
Google provides a dedicated report on performance in generative AI features in Search Console. It shows how your content performs in AI search and in Discover, and gives an impression of how people discover your content there.
We analyze this report together with visits from ChatGPT, Perplexity and other AI sources. This shows you which pages work in AI search and what these visitors contribute to inquiries and revenue.
Prepare your website for AI agents now
The guide offers an outlook on AI agents. These systems act on behalf of users: They compare products and take over tasks such as reservations.
According to Google, browser agents analyze screenshots, the DOM structure and the accessibility tree of a page for this purpose. Google recommends following the guidance for agent-friendly websites and mentions new protocols such as the Universal Commerce Protocol (UCP), which will allow agents to do more in Search in the future. What this means for online shops is shown in our guide to agentic commerce.
You implement the guide in four steps
- Bring your own experience into the content: Encourage your authors to build personal expertise, their own data and first-hand accounts into their articles.
- Expand images and videos: Optimize image and video SEO so that your content also appears in the visual parts of AI answers.
- Fix technical errors: Check in Google Search Console whether your pages are crawled and indexed without problems, and track performance in the AI features there.
- Focus on fundamentals: Ignore questionable GEO tricks and concentrate on a clean SEO strategy that starts from the user.
