# The optimal page for AI search: Structure, checklist and research findings for 2026 > URL: https://4eck-media.de/en/blog/the-optimal-page-for-ai-search-structure-checklist-and-research-findings-for-2026/ > Language: en > Description: Optimising for AI search means structuring a website so that language models such as ChatGPT, Claude, Perplexity or Google AI Overviews can find, understand and cite its content a… --- **Optimising for AI search means** structuring a website so that language models such as ChatGPT, Claude, Perplexity or Google AI Overviews can find, understand and cite its content as a source. Specialists refer to this as AEO (Answer Engine Optimization) or GEO (Generative Engine Optimization). The difference from traditional search engine optimisation: on Google, a person decides which result to click. In AI search, a language model decides which passage to use. In September 2026, the US agency Nectiv published a framework called “The Perfectly Optimized Page for AEO”, modelled on the well-known Moz SEO classic. We reviewed the underlying studies. This article shows which twelve building blocks a page for AI search needs, what the data actually proves and where to start with a WordPress website. Key takeaways - 44.2% of ChatGPT citations come from the first third of a text (Kevin Indig, 1.2 million AI answers, 2026) - According to AirOps, pages without a quarterly update lose their AI citations three times as often - Pages cited by ChatGPT contain a table 2.3 times as often as top Google results (Nectiv, 2025) - Most AI crawlers do not execute JavaScript; content loaded afterwards remains invisible - There is no perfect page, but there are twelve building blocks where the effort pays off ## The 12 building blocks of a page for AI search at a glance | # | Building block | What it is about | | --- | --- | --- | | 1 | One intent per page | Title, H1 and URL describe the same topic | | 2 | Allow AI crawlers | Allow them in robots.txt; make content available without JavaScript | | 3 | Clear headings | one H1, logical H2s and H3s | | 4 | Summary at the top | 4 to 6 sentences that answer the question directly | | 5 | Definition | an “X is …” statement at the beginning | | 6 | Self-contained sections | each paragraph also works on its own | | 7 | Tables and lists | structured formats rather than continuous text | | 8 | Fan-out questions | answer sub-questions specifically | | 9 | Your own figures | citable data with a source and date | | 10 | Exclusive knowledge | insights that only you have | | 11 | FAQ | real customer questions, direct answers | | 12 | Author and freshness | name, qualification, visible date | The framework comes from Nectiv; the evidence for the individual building blocks comes from studies by Kevin Indig, AirOps, Cyrus Shepard, Chris Green and Vercel. You can find the sources at the end of this article. ## How do you allow AI crawlers onto your website? AI crawlers must be allowed to retrieve your page and find its content in the HTML source code. Otherwise, no language model can cite your content, no matter how good it is. Two things are important: 1. **Check robots.txt:** The search bots of AI providers should be allowed: `OAI-SearchBot` and `ChatGPT-User` (OpenAI), `Claude-SearchBot` and `Claude-User` (Anthropic), `PerplexityBot` and `Googlebot`. Whether you also allow training bots such as `GPTBot`, `ClaudeBot` or `Google-Extended` is a deliberate decision. You help determine what models know about your brand in the long term. 2. **Deliver content server-side:** According to an analysis by Vercel, the major AI crawlers do not render JavaScript. Text in sliders, accordions or tabs that is only loaded by a script remains invisible to them. For [WordPress websites](https://4eck-media.de/en/competencies/wordpress-agency-websites-that-work-for-your-business/), this is good news: by default, WordPress delivers content as finished HTML. Some page builders, headless setups and cookie banners that block content are critical. A quick test: open the page with JavaScript disabled. What is missing then is also missing for AI. ## Why does the answer belong at the beginning of the page? AI systems cite the upper part of a text disproportionately often. For his Growth Memo newsletter, Kevin Indig analysed 1.2 million AI answers and measured which part of a page ChatGPT draws its citations from: | Position in the text | Share of ChatGPT citations | | --- | --- | | First third (0 to 30%) | 44.2% | | Middle (30 to 70%) | 31.1% | | Final third (70 to 100%) | 24.7% | That is why a four- to six-sentence summary belongs directly below the heading. Each sentence must answer a sub-question completely and be understandable without context. The rest of the page is also cited. The point is weighting, not neglecting the rest. Indig also found that passages with clear definitions (“X is …”) are cited almost twice as often. Cited texts contain around 20% proper nouns such as brands, tools or places; ordinary texts contain only 5 to 8%. ## Which formats do AI systems prefer to cite? AI systems prefer content that is already structured: tables, bullet points, numbered steps and question-and-answer pairs. - **Tables:** 30% of pages cited by ChatGPT contain a table, compared with only 13% of top Google results (Nectiv, B2B SaaS queries, November 2025). This is a correlation, not a proven cause. - **Lists:** Nearly 80% of pages cited by ChatGPT contain lists (AirOps, 2025). - **Headings:** 87% of cited pages have exactly one H1 (AirOps, 2025). - **Question-and-answer format:** In a test by SEO consultant Chris Green, the Q&A format matched search queries best, while dense continuous text performed worst. Tables must be created as real HTML tables. A crawler cannot read an image of a table. ## What are fan-out queries? Fan-out queries are sub-questions into which an AI search breaks down a single prompt. If someone asks “Which agency for a WordPress relaunch?”, the system searches in the background for costs, duration, references and reviews, for example. A page that answers these sub-questions has a better chance of being cited. In a survey of more than 130 SEO professionals by Cyrus Shepard (September 2026), covering fan-out queries was one of the most important factors. Here is how to proceed: 1. Define the page’s main question. 2. Collect follow-up questions: from ChatGPT and Perplexity, from Google’s “People also ask”, and from sales conversations and customer emails. 3. Answer the most important questions under their own subheading or in the FAQ. 4. Move peripheral topics to their own pages and link to them internally. ## How important is freshness for AI search? Very important. According to AirOps, pages without a quarterly update lose their citations in AI answers three times as often. More than 70% of pages cited by ChatGPT have been updated within the last twelve months. For commercial search queries, 60% of citations come from content from the last six months (AirOps, more than 4,000 pages, 900 search queries). - a visible “Last updated” date that matches `dateModified` in the schema - figures with an as-of date, for example “Prices checked on …” - genuine content updates instead of merely moving dates forward ## Your own data and exclusive knowledge: what only you can provide Language models already know the standard definition of your topic from dozens of other articles. They cite what is new and verifiable: your own figures, project experience and surprising findings. In Shepard’s survey, “citable, concrete facts” and “first-hand information of your own” are among the strongest factors. A single original metric with a sample and time period is often the most valuable element of a page. ## What we see at 4eck Media in projects On 4eck-media.de, we deliver both: an llms.txt file and Markdown versions of all pages via content negotiation. Between May and September 2026, llms.txt was retrieved 65 times. With around 221,000 bot accesses in total, that is about 0.03%. Fourteen different bots looked for the file, most regularly OpenAI’s GPTBot and OAI-SearchBot, with 24 requests between them. More than half of the requests came to nothing, however: bots also searched for the file under language paths such as /en/llms.txt or /fr/llms.txt, which did not exist on our site. Markdown, by contrast, is used intensively. Over the last 30 days, 75% of all successful page views by bots were delivered as Markdown; excluding Meta’s crawler, the figure was 46%. ClaudeBot, ChatGPT-User, PerplexityBot and OAI-SearchBot each received more than 95% of their pages as Markdown. Our conclusion: llms.txt is found, but plays a secondary role. Clean Markdown delivery reaches the AI crawlers that matter. ![Bot-Hits](https://4eck-media.de/wp-content/uploads/2026/09/bot-hits.avif "Bot-Hits") ## Checklist: Is your page ready for AI search? | Checkpoint | Met when … | | --- | --- | | Title, H1, URL | one topic, exactly one H1, a short, descriptive URL | | Crawler access | search bots allowed in robots.txt; content visible without JavaScript | | Headings | logical H1, H2, H3 order; subheadings framed as questions | | Summary | 4 to 6 self-contained sentences with figures right at the top | | Definition | a clear “X is …” sentence in the first section | | Sections | each begins with the answer; no references such as “as mentioned above” | | Formats | at least one HTML table; lists for characteristics and steps | | Fan-out | most important sub-questions answered | | Your own data | at least one metric with a source and time period | | FAQ | real customer questions, FAQ schema | | Author | name, qualification, author page | | Freshness | visible date; updated within the last three months | | Schema | Article, FAQPage, Organization, Person | ## Sources - [Nectiv: The Perfectly Optimized Page for AEO](https://nectivdigital.com/blog/the-perfectly-optimized-page-for-aeo-ai-search), September 2026 - [Kevin Indig via Search Engine Land: 44% of ChatGPT citations from the first third](https://searchengineland.com/chatgpt-citations-content-study-469483), 2026 - [Nectiv: ChatGPT Citations 2.3x More Likely To Include A Table](https://nectivdigital.com/blog/new-data-study-chatgpt-citations-are-2-3x-more-likely-to-include-a-table-than-google-search/), November 2025 - [AirOps: The 2026 State of AI Search](https://www.airops.com/report/the-2026-state-of-ai-search), December 2025 - [AirOps: The Impact of Stale Content on AI Visibility](https://www.airops.com/report/the-impact-of-stale-content-on-ai-visibility), August 2025 - [Cyrus Shepard: Google AI Ranking Factors](https://signal.zyppy.com/p/google-ai-ranking-factors), September 2026 - [Chris Green: Content Structure for AI Search](https://www.chris-green.net/post/content-structure-for-ai-search), June 2025 - [Vercel: The Rise of the AI Crawler](https://vercel.com/blog/the-rise-of-the-ai-crawler) ## More about SEO, AI visibility and WordPress More articles and technical insights from our agency practice. [![Forest to Table: game-meat shop on tablet and smartphone next to a paper deer and two score circles showing 90 and 92.](https://4eck-media.de/wp-content/uploads/2026/09/wald-zu-tisch-responsive-webdesign-blog-02.avif "Forest to Table after the relaunch: the game-meat shop on tablet and smartphone.")](https://4eck-media.de/en/blog/forest-to-table-shop-relaunch-from-prestashop-to-woocommerce/ "Forest to Table: Shop Relaunch from PrestaShop to WooCommerce") [Forest to Table: Shop Relaunch from PrestaShop to WooCommerce](https://4eck-media.de/en/blog/forest-to-table-shop-relaunch-from-prestashop-to-woocommerce/) Matthias Petri [![Custom ERP for WooCommerce: Case Study of Precious Metals Dealer Goldmichi.de](https://4eck-media.de/wp-content/uploads/2026/09/goldmichi-erp-blog.avif "Custom ERP for WooCommerce: Case Study of Precious Metals Dealer Goldmichi.de")](https://4eck-media.de/en/blog/custom-erp-for-woocommerce-case-study-of-precious-metals-dealer-goldmichi-de/ "Custom ERP for WooCommerce: Case Study of Precious Metals Dealer Goldmichi.de") [Custom ERP for WooCommerce: Case Study of Precious Metals Dealer Goldmichi.de](https://4eck-media.de/en/blog/custom-erp-for-woocommerce-case-study-of-precious-metals-dealer-goldmichi-de/) Matthias Petri [![AI search page structure](https://4eck-media.de/wp-content/uploads/2026/09/ki-suche-checkliste-seitenaufbau.avif "AI search page structure")](https://4eck-media.de/en/blog/the-optimal-page-for-ai-search-structure-checklist-and-research-findings-for-2026/ "The optimal page for AI search: Structure, checklist and research findings for 2026") [The optimal page for AI search: Structure, checklist and research findings for 2026](https://4eck-media.de/en/blog/the-optimal-page-for-ai-search-structure-checklist-and-research-findings-for-2026/) Matthias Petri [![Markdown Negotiation beats LLMS.TXT](https://4eck-media.de/wp-content/uploads/2026/09/llms-txt-vs-markdown.avif "Markdown Negotiation beats LLMS.TXT")](https://4eck-media.de/en/blog/markdown-negotiation-beats-llms-txt/ "Markdown Negotiation beats llms.txt") [Markdown Negotiation beats llms.txt](https://4eck-media.de/en/blog/markdown-negotiation-beats-llms-txt/) Matthias Petri ## FAQ: Optimising websites for AI search Is optimising for AI search different from SEO? No, it builds on it. According to Cyrus Shepard’s survey, strong Google rankings also increase the chance of AI citations. What is new is the focus on individual, citable passages rather than the page as a whole. Which bots do I need to allow in robots.txt for AI search? At minimum, the search bots OAI-SearchBot and ChatGPT-User from OpenAI, Claude-SearchBot and Claude-User from Anthropic, PerplexityBot and Googlebot. Training bots such as GPTBot, ClaudeBot and Google-Extended are a deliberate decision because they influence what models know about your brand in the long term. Do I need an llms.txt file? Not as a priority at present. In Cyrus Shepard’s survey, more than 130 SEO professionals rated the influence of llms.txt as almost neutral (+0.05 on a scale from -3 to +3). Does structured schema markup help with AI search? Yes, but moderately. According to AirOps, pages with at least three schema types have a 13% higher likelihood of being cited. Schema does not replace good content. Do I need to rebuild my WordPress website for AI search? Usually not. WordPress already delivers content as HTML. Adjustments to robots.txt, structure, schema and content are often sufficient. A rebuild is more worthwhile when a page builder loads content only through JavaScript. How often should I update content for AI search? Important pages at least once per quarter. The AirOps data shows that pages without a quarterly update disappear from AI answers three times as often. ## Is your website already built for AI search? We assess your pages for crawler access, structure, schema and content and tell you where effort will pay off. 4eck Media from Waren (Müritz) has been developing [WordPress](https://4eck-media.de/en/competencies/wordpress-agency-websites-that-work-for-your-business/) and [WooCommerce projects](https://4eck-media.de/en/competencies/build-a-woocommerce-online-shop-with-a-specialized-woocommerce-agency/) since 2010, with more than 350 completed projects, and is ranked No. 1 among Germany’s WordPress agencies in the Agenturtipp.de live ranking (as of: DATE TO BE INSERTED). [Talk to us about your AI visibility.](https://4eck-media.de/en/contact/)