# Designing for AI search: getting cited > AI search optimization for company websites: how AI answers pick sources, why server HTML matters, and how to write pages that get cited and clicked. Canonical page: https://kyvrel.com/blog/ai-search-optimization Published: 2026-07-28 · 6 min read · Kyvrel AI search optimization is mostly good SEO done properly, plus a few technical checks. AI answers in Google, ChatGPT and Perplexity pull from pages they can fetch, parse and trust. Put your content in plain server HTML. Answer questions directly and state facts with numbers. Then you are most of the way to being cited. The harder part is that AI answers reward a different writing habit. They lift short, self-contained passages, so a page full of slogans gives them nothing to quote. This guide covers what we change in structure, code and copy when we build a site that should show up in AI answers. ## What AI search optimization actually means Search used to end with a list of links. Now many queries end with a written answer, followed by a few cited sources. Google shows AI Overviews and AI Mode. ChatGPT, Perplexity and others run their own search and cite pages inline. Getting cited means three things happen in order: 1. A crawler can fetch your page and read the text. 2. A retrieval system decides your page is relevant to the question. 3. The model finds a passage that answers the question clearly enough to quote. You influence all three. The first is technical, the second is classic SEO, and the third is writing. None of it needs a separate "AI version" of your site. Google says this plainly in its guidance on [AI features and your website](https://developers.google.com/search/docs/appearance/ai-features). There are no extra requirements or special markup for AI Overviews or AI Mode. The usual fundamentals apply: crawlable pages, helpful content, and text in the HTML. ## How do AI search engines choose which pages to cite? No vendor publishes its ranking formula, so be wary of anyone who claims to know it. What we can see is what cited pages tend to have in common. They answer a specific question early. They use headings that match how people phrase questions. They give concrete facts, such as prices, timelines, steps and limits, instead of adjectives. ![The path from a user's question to a cited answer: crawl, index, retrieve relevant passages, compose the answer, cite sources](https://kyvrel.com/blog/inline/ai-search-optimization-1.png) Two practical points follow from this: - **Passages matter more than pages.** A model often quotes two or three sentences. Each section should make sense when read alone. - **Trust signals still count.** Clear authorship, a dated page, a real company behind it and consistent facts across your site all help a page look like a reliable source. ## Make every word readable without JavaScript This is the most common technical gap we find. Many sites render key content in the browser after JavaScript runs. Googlebot renders JavaScript, but many AI crawlers do not. Vercel's analysis of [AI crawler traffic](https://vercel.com/blog/the-rise-of-the-ai-crawler) found that major AI crawlers fetch JavaScript files without executing them. If your pricing table, FAQ answers or product details load client-side, those crawlers may see an empty shell. The fix is architectural: - Render content on the server or at build time, so it is in the first HTML response. - Keep FAQ answers in the DOM, even if they sit inside a collapsible element. - Avoid putting important text in images, canvas or embedded widgets. - Check the raw HTML with `curl` or "view source", not just the rendered page. This also helps speed. Server-rendered pages usually paint faster, which matters for both users and rankings. We covered the performance side in our guide to [Core Web Vitals for marketing sites](https://kyvrel.com/blog/core-web-vitals-marketing-sites). ## Write pages that answer, not pages that announce The writing change is bigger than the technical one. Marketing copy often opens with a mood: "We craft digital experiences." An AI system cannot cite that, because it answers nothing. Compare it with "A landing page takes 1-2 weeks and starts from $5,000." That sentence can be quoted, checked and trusted. ### Lead with the answer Open each page and each major section with a two or three sentence answer. Then add the detail, the exceptions and the reasoning. Journalists call this the inverted pyramid. It works for skimming readers and for machines. ### Use question-shaped headings where they fit If buyers ask "How long does a website redesign take?", a heading with that question is easy to match. Do not turn every heading into a question. Use them where people really ask that question. ### State facts in a form that survives quoting - Put numbers next to their units and conditions: price, scope, time. - Name things precisely: the framework, the standard, the regulation. - Keep facts consistent across pages. Conflicting prices on two pages make both less trustworthy. - Date time-sensitive content, and update it when the facts change. ### One page per question, one question per page A page that tries to answer everything usually answers nothing well. Clear page roles help retrieval systems pick the right page. They also stop your own pages from competing with each other. ## Structured data, sitemaps and llms.txt Structured data does not earn citations by itself. It does make your facts easier to parse. We add JSON-LD for the organization, services, articles, breadcrumbs and visible FAQ content. Every value in the markup must match the visible text. Keep the XML sitemap accurate, with real last-modified dates. It is the cheapest way to tell crawlers what changed. The [llms.txt proposal](https://llmstxt.org/) suggests a plain-text index of your key pages for language models. Adoption by the big AI search products is unclear, so treat it as a cheap extra, not a strategy. We publish one on our own site because it costs little to keep accurate. ![A grid of site assets that help AI crawlers: server HTML, headings, JSON-LD, sitemap, robots rules and a plain-text index](https://kyvrel.com/blog/inline/ai-search-optimization-2.png) ## Decide which AI crawlers get in AI crawlers do different jobs. Some fetch pages to answer a live search. Others collect data for model training. OpenAI, for example, documents a separate search crawler, OAI-SearchBot, alongside its training crawler. You can allow one and block the other in `robots.txt`. ``` # robots.txt: allow AI search, block a training crawler User-agent: OAI-SearchBot Allow: / User-agent: GPTBot Disallow: / ``` This is a business choice, not a technical one. Blocking search crawlers means you will not be cited in those products. Blocking training crawlers is a separate decision about how your content is reused. Also check your CDN and firewall settings. A bot-protection rule can block AI crawlers even when `robots.txt` allows them. If you are moving platforms, plan this with your redirects and sitemaps. Our post on [website migration without losing SEO](https://kyvrel.com/blog/website-migration-seo) covers the parts that transfer directly. ## How to measure AI search visibility Measurement is still rough, so set expectations early. A few signals are available today: - **Referral traffic** from AI assistants and answer engines shows up in your analytics, often with the product's domain as the source. - **Search Console** includes AI Overviews and AI Mode traffic in the overall Web search numbers, according to Google's guidance. - **Manual checks** on the ten questions your buyers ask most, repeated monthly, show whether you are cited and how you are described. - **Lead forms** can ask "Where did you hear about us?" with an AI assistant option. Treat these as directional. The goal is to see whether answers describe you accurately, and whether the visits that arrive are qualified. ## Build a site that AI answers can quote Most of this work happens in the architecture and the content model, before any copy is written. Server-rendered pages, clear page roles and consistent facts are standard in every site we build. Content structured for AI search also feeds a site chatbot. Our [AI chatbot for your website](https://kyvrel.com/products/ai-chatbot) answers visitors from those same pages, from $12,000. If you are weighing a site builder against a custom build for this, our [Webflow vs custom comparison](https://kyvrel.com/compare/webflow-vs-custom) sets out the trade-offs.