Optimize Your Website for AI Search: What Google Says and What 96,418 URLs Show
Dominik is the founder and managing director of taismo, an SEO and GEO agency from Munich. He has worked in search marketing since 2010 and focuses on ongoing SEO support and visibility in AI answers, for companies in Germany and for international firms that want to be found in the German market.
To optimize your website for AI search, work through four levels: Make your pages accessible to search engines and AI crawlers, publish content with first-hand value, structure it clearly, and make your brand unambiguous as an entity. Google states in its guide to AI features in Search that this work is still SEO, and it calls llms.txt files, special markup and chopping text into chunks unnecessary. A SISTRIX study of 96,418 URLs shows what cited pages have in common: A table of contents with jump links (factor 1.7), Article schema and at least five H2 headings (1.6 each), FAQ schema and tables (1.5 each).
👉 Want to know where your website stands today? Run the free AI visibility check
Google’s guide: Optimizing for AI search is still SEO
Google collects its advice for AI features in Search in a dedicated guide, “Optimizing your website for generative AI features on Google Search,” last updated on July 10, 2026. Its core message: Optimizing for generative AI search means optimizing for search, and that is still SEO. Google explicitly files AEO (answer engine optimization) and GEO under SEO. If you want the full comparison of the two disciplines, read our guide GEO vs. SEO.
The reason is technical. AI Overviews and AI Mode run on the core systems of Google Search. They use retrieval-augmented generation: The system retrieves matching pages from the search index and writes an answer from them. Through query fan-out, it runs several related searches on subtopics in parallel. A page that is missing from the index or does not rank for those subqueries is rarely picked as a source.
Three foundations Google names for AI search
- Non-commodity content: Google asks for content that brings first-hand experience, original data and a perspective of its own. Producing pages at scale mainly to move rankings can violate Google’s spam policy on scaled content abuse.
- A clean technical setup: Pages must be crawlable and indexable, eligible to appear with a snippet, and the site must be included in generative AI features in Search Console. Google recommends semantic HTML and says perfection is not required.
- Maintained business data: Local businesses keep their Google Business Profile current, online stores their product data in Merchant Center. According to Google, both can improve visibility in AI answers.
For AI Overviews and AI Mode, Google lists no additional technical requirements. A page can appear as a supporting link once it is indexed and eligible to show with a snippet. Google also states that crawling, indexing and serving are never guaranteed.
Five AEO tactics Google says you can skip
A mythbusting section of the guide lists what you don’t need to do. It names five tactics that do nothing for AI features in Google Search, and all five are being sold as AEO services right now:
| Tactic | What Google’s guide says |
|---|---|
| llms.txt and other AI files | You don’t need new machine-readable files, AI text files, markup or Markdown to appear in Google Search. |
| Breaking content into chunks | There is no requirement to split content into tiny pieces for AI. Google sets no page length. |
| Rewriting text for AI | You don’t need to write in a special way for generative AI search, and you don’t need more long-tail keywords. |
| Chasing mentions | Seeking inauthentic mentions across the web isn’t as helpful as it might seem. |
| Schema as an AI requirement | Structured data isn’t required for generative AI search, and there is no special schema.org markup for it. |
Two caveats come with this list. First, Google speaks only for its own search. ChatGPT, Perplexity and Claude use their own crawlers, and their providers disclose little about the signals they weigh. Several popular guides still tell you to write in “chunks,” which shows how far advice and evidence can drift apart. Second, Google keeps recommending structured data because it enables rich results and describes a page precisely. For AI features, Google stresses one rule: Your structured data has to match the visible text on the page.
Ask two questions about any AI tactic someone pitches you: Which system is it built for, and which metric will show that it works? A tactic without a target system and without a metric is a bet.
What 96,418 URLs show about pages AI systems cite
SISTRIX, a search analytics company based in Germany, published a correlation study on October 22, 2026. The team around Johannes Beus and Julia Weißbach analyzed 96,418 URLs from 17,380 domains against 43 on-page features. The base is 48,527 pages that ChatGPT, Google AI Mode and AI Overviews cited in the second quarter of 2026, about 16,000 per system. Each cited page was compared with a page from the same domain that was not cited. A factor of 1.0 means no difference. A factor of 1.7 means: Pages with this feature are cited about 1.7 times as often.
The data comes from Germany, Austria and Switzerland. The Google guide applies worldwide, so the study is better read as a strong signal from one large market and less as a US benchmark.
| Feature | Citation factor | Note from the study |
|---|---|---|
| Table of contents with jump links | 1.7 | only 14% of pages have one |
| Article schema | 1.6 | 2.0 in AI Overviews |
| At least five H2 headings | 1.6 | 1.8 in AI Mode |
| FAQPage schema with real questions | 1.5 | smaller test runs pointed in opposite directions |
| Table with a header row | 1.5 | 1.8 in ChatGPT |
| At least 1,000 words | 1.5 | per SISTRIX more a result of better structure |
| Author box with photo | 1.4 | author box in general 1.3 |
| Visible last-updated date | 1.3 | 1.9 on editorial sites |
| At least ten external links | 1.3 | |
| Short summary at the top | 1.3 |
Features with no measurable effect
- llms.txt: 11.7% of cited URLs come from domains with an llms.txt file, compared with 14.0% across all domains in the study.
- Flesch readability: Factor 1.03, not statistically significant.
- Breadcrumb, HowTo and QAPage schema: Factor 1.0 each.
- Product schema: Factor 0.8 overall and 0.6 for online stores.
How solid the numbers are
The study measures correlation and proves no cause. A table of contents often marks a page that was built with editorial care, and the study leaves open whether the table itself or the care behind it makes the difference. SISTRIX names further limits: Perplexity, Claude and the Gemini app are not included, some title and H1 features were added after the fact, and the authors consider studies with fewer than 5,000 URLs unreliable. Pages that meet many of the features are cited 2.1 times as often as barely optimized pages. Only 10% of cited pages reach that level.
Results by site type and AI platform
The overall factors hide large differences. SISTRIX broke the results down by domain type and by platform, and the breakdown changes the priorities for many sites:
| Site type | What the study found | What follows |
|---|---|---|
| Brand and company sites | Article schema and FAQ schema 1.6 each | Mark up advice content as articles and answer real customer questions on the page |
| Editorial sites and publishers | Table of contents 2.5, last-updated date 1.9 | Add jump links to long articles and show the date you actually revised them |
| Online stores | Last-updated date 0.7, Product schema 0.6 | Build an advice section with Article schema next to the product pages |
| SaaS providers | Product schema 1.7 | Describe the software as a product with features and pricing |
| Encyclopedias | The optimizations show no effect in total | Authority comes from elsewhere |
The platforms also weigh features differently. ChatGPT cites pages with a question-style title (1.9), a table with a header row (1.8) and a question as an H2 (1.6) more often. AI Mode favors a table of contents and at least five H2 headings, 1.8 each. AI Overviews show the strongest link to Article schema (2.0) and the weakest to Product schema (0.5). According to SISTRIX, all three platforms still land close to each other overall. For local businesses, Google adds a point the study does not cover: A complete Google Business Profile feeds the answers that AI features give for local searches.
Where Google’s guide and the study data meet
At first glance the two sources contradict each other. Google calls structured data optional for AI search, while SISTRIX measures a factor of 1.6 for Article schema. Both statements hold: Google describes what its systems require. SISTRIX describes which pages actually get cited. Article schema, a table of contents and an author box mark pages that someone maintains with effort, and those are the pages Google means when it asks for non-commodity content.
On llms.txt, both sources agree. Google does not use the file for Search, and SISTRIX finds no effect. We still run a bilingual llms.txt on taismo.de. It costs little to maintain, other AI systems can read it, and it forces us to describe our services clearly in a few lines. We never present it as a visibility lever. How the file is built and what it can do is covered in our llms.txt guide.
Both sources point to the same order of work. First, the page has to be reachable for search engines and AI crawlers. Then content decides, then the structure that makes it readable. The study features act as the last layer: They amplify pages whose content already holds up.
AEO checklist: Optimize your website for AI search in four levels
This checklist assigns every task to one of four levels and names its evidence from the guide and the study. Work through the levels from top to bottom, because each one builds on the previous one.
Level 1: Access for search engines and AI crawlers
- Indexing: Use the URL Inspection tool in Search Console to confirm the page is in the index. Without indexing, a page appears in none of the AI features of Google Search.
- Snippet eligibility: According to Google,
nosnippet,data-nosnippet,max-snippetandnoindexalso limit what appears in AI Overviews and AI Mode. - Search Console setting: Check that your property is included in generative AI features in Search. The guide names this inclusion as a prerequisite.
- robots.txt: OpenAI separates OAI-SearchBot, which powers search in ChatGPT, from GPTBot, which collects training data. Sites that block OAI-SearchBot do not appear in ChatGPT search answers. How to allow each bot separately is explained in our guide to GPTBot and the other AI crawlers.
- Text without JavaScript: According to an analysis by Vercel, the AI crawlers of OpenAI, Anthropic, Meta, ByteDance and Perplexity do not execute JavaScript. Text that only loads in the browser never reaches them.
Check your firewall and CDN as well as your robots.txt. Since July 1, 2025, Cloudflare blocks AI crawlers that access content without permission by default and asks every new domain at setup whether it wants to allow them. An allow rule in robots.txt helps little if the firewall turns the bot away first.
Level 2: Content with first-hand value
- Answer in the first paragraph: Every page answers its core question in the first sentence, and that sentence makes sense on its own. SISTRIX measures a factor of 1.3 for a short summary at the top.
- Your own data and cases: Numbers from your projects, measurements and examples give a model a reason to cite your page. Google asks in its guidance on helpful content whether content shows first-hand expertise.
- Author with photo and bio: Factor 1.4. A named author with verifiable experience supports the signals Google groups under E-E-A-T.
- Visible last-updated date: Factor 1.3, and 1.9 on editorial sites. Show the date only on pages you actually revise.
- Evidence beyond your site: At least ten external links come with a factor of 1.3, a scientific source with 1.2.
How to build passages that AI systems quote word for word is covered in our playbook how to get cited by AI.
The taismo GEO audit reviews entity, structured data and machine readability across your whole website and delivers the fixes in the right order.
Level 3: Structure that machines can read
- Table of contents with jump links: Factor 1.7, the strongest structural feature in the study.
- At least five H2 headings: Factor 1.6. In AI Mode, the table of contents and the H2 outline reach 1.8 each.
- Tables with a header row: Factor 1.5, and 1.8 in ChatGPT.
- Article and FAQPage schema: Factors 1.6 and 1.5. The questions in the markup appear visibly on the page. Which types fit which page is explained in our guide to schema markup in practice.
- Semantic HTML: Headings as
h2andh3, lists asulandol, tables astable.
Level 4: Entity and proof beyond your website
The fourth level sits partly outside your website. AI systems recognize a brand as an entity when its name, location, services and people read the same everywhere: On the website, in the Google Business Profile, in directories and in the schema graph. Brand mentions in trade media and comparisons back up what your website says about you. Google explicitly warns that seeking inauthentic mentions helps less than it seems.
The table sums up the evidence for the main tasks:
| Task | Google’s guide | SISTRIX factor |
|---|---|---|
| Indexing and snippet eligibility | Prerequisite | not measured |
| Table of contents with jump links | not mentioned | 1.7 |
| Article schema | not needed for AI, recommended for rich results | 1.6 |
| FAQPage schema | not needed for AI | 1.5 |
| Author box with photo | first-hand experience counts | 1.4 |
| Visible last-updated date | not mentioned | 1.3 |
| llms.txt | not used by Google Search | no measurable effect |
| Chasing mentions | helps less than it seems | not measured |
Self-test: This article against the checklist
We ran the checklist on this article. The newer posts in the taismo SEO Magazine follow the same layout, so the result stands for the whole format. The counts come from the source code of this page.
The result shows the limit of any page-level test. Structure and content can be read from a single URL, and this article meets the features because they are built into the magazine layout. Whether an AI system names a brand also depends on the entity, on mentions and on how the source pages rank in regular search. No test of a single page can see those signals.
How to measure AI search visibility
For Google, Search Console is the first place to look. The guide points to the generative AI performance report in Search Console. Clicks from AI Overviews and AI Mode also count toward the Web search type, so they already sit in your regular performance report. What the AI report shows and where it stops is covered in our article on the Search Console AI report.
A finding from our own data: From July to September 2026, almost half of the AI impressions for taismo.de came from a single glossary page that appeared for a navigational query. A spike like that shows demand for a term and says little about whether your service gets named. Break the report down by page before you read a rise as success.
ChatGPT, Perplexity and Gemini offer no comparable report. There you track a fixed set of questions that you ask the systems on a regular schedule, and you count how often your brand is named or linked as a source. The selection decides the result: The questions should carry a buying decision, such as “Which agency for …” or “X or Y.” How to pick and track those questions is explained in our guide to prompt tracking. Google also advises caution with third-party tools that promise ranking success or claim to use internal Google metrics.
In-house or with an agency: What a team can handle alone
A company with its own editorial and web team can check levels 1 to 3 largely on its own. Search Console, a look at robots.txt and the firewall, the source code of a few pages and the checklist above are enough. The work gets demanding in four places:
- Page templates: A table of contents, an author box and Article schema live in the template of a page type. Adding them means changing the theme and testing every page type afterward.
- The schema graph: Organization, people, services and articles connect through their
@idand carry the same facts everywhere. - The entity outside the website: Aligning your business profile, directories, social profiles and trade media takes time and persistence.
- Tracking: Following a fixed set of questions across several systems for months and breaking spikes down by page.
These four tasks form the core of our work in AI search optimization. We build machine-readable structure, make the brand recognizable as an entity, write citable content, and track mentions in AI answers together with rankings. At taismo, AEO runs as part of ongoing SEO support, since Google itself treats optimizing for AI search as SEO. The starting point is the GEO audit: It maps where your website stands on all four levels and delivers the fixes in order of impact.
We review and build access, content, structure and entity signals for your website and track whether AI systems name you. Tell us about your project.
FAQ about optimizing your website for AI search
How do you optimize your website for AI search?
Make sure search engines and AI crawlers can access and index your pages, publish content with first-hand value, and structure it with a table of contents, clear H2 sections and tables. Google says that optimizing for its AI features is still SEO.
Do I need an llms.txt file to show up in AI search?
Not for Google: Google says it does not use llms.txt, and a SISTRIX study of 96,418 URLs found no measurable effect. Other AI systems may read the file, so a maintained llms.txt does no harm.
Does schema markup help with AI search?
Google requires no special schema.org markup for its AI features. Pages cited by AI systems carry Article schema (factor 1.6) and FAQPage schema (factor 1.5) more often, according to SISTRIX.
How do I show up in ChatGPT search results?
Allow OAI-SearchBot in your robots.txt and your firewall, because OpenAI does not show sites that block it in ChatGPT search answers. Keep your main text in the server HTML, since OpenAI’s crawlers do not run JavaScript.
How do you measure AI search visibility?
For Google, use the generative AI performance report in Search Console; clicks from AI Overviews and AI Mode count as Web search traffic. For ChatGPT, Perplexity and Gemini, track a fixed set of buyer questions over time and count how often your brand is named or cited.
Is AEO different from SEO?
Google treats AEO and GEO as part of SEO, because its AI features run on the same core search systems. AEO adds work on entity signals, citable content and measurement inside AI answers.
Sources
- Google Search Central: “Optimizing your website for generative AI features on Google Search“, Google, last updated July 10, 2026.
- Google Search Central: “AI features and your website“, Google, retrieved October 29, 2026.
- Johannes Beus, Julia Weißbach and the SISTRIX Content Team: “GEO-Mythen im Datencheck: Was 96.418 URLs über AI-Citations verraten (GEO myths fact-checked: What 96,418 URLs reveal about AI citations, in German)“, SISTRIX, October 22, 2026, updated October 28, 2026.
- Google Search Central: “Spam policies for Google web search“, Google, retrieved October 29, 2026.
- Google Search Central: “Creating helpful, reliable, people-first content“, Google, retrieved October 29, 2026.
- Google Search Central: “Robots meta tag, data-nosnippet, and X-Robots-Tag specifications“, Google, retrieved October 29, 2026.
- Google Search Console Help: “URL Inspection tool“, Google, retrieved October 29, 2026.
- OpenAI: “Overview of OpenAI Crawlers“, OpenAI, retrieved October 29, 2026.
- Giacomo Zecchini, Alice Alexandra Moore, Malte Ubl, Ryan Siddle: “The rise of the AI crawler“, Vercel, December 17, 2024.
- Cloudflare: “Cloudflare Just Changed How AI Crawlers Scrape the Internet-at-Large; Permission-Based Approach Makes Way for A New Business Model“, Cloudflare, July 1, 2025.
