# Documentation SEO: How to Make Your Docs the Source AI Cites (/guides/documentation-seo) Documentation SEO is the work of making your product docs findable, readable, and quotable, by Google and by AI answers. Done well, your docs become the page ChatGPT, Perplexity, and Google AI Overviews cite when someone asks how your product does something. The catch is in that last clause. In our own citation data, AI cites documentation heavily when the question is **how something works**, and almost never when it is **which product to buy**. On the 30 prompts we track for RankZero, two pages of Google's own Search documentation are the most-cited sources after YouTube. Each one is cited in roughly 7% of all answers, across 18 of the 30 prompts. Most documentation SEO guides are written by docs platforms and stop at titles, headings, and sitemaps. This one covers what decides whether AI can use your docs at all: rendering, login walls, page scope, versions, and where llms.txt fits. > **Key Takeaways** > > * AI cites docs for "how does X do Y" questions. In our tracking, Google's two AI search docs pages are each cited in about 7% of answers on our 30 prompts, while docs barely appear in "best X" categories. > * Most AI crawlers do not run JavaScript. If your docs render client-side, ChatGPT and Claude see an empty page ([Vercel and MERJ, December 2024](https://vercel.com/blog/the-rise-of-the-ai-crawler)). > * A docs page behind a login cannot be cited. Put the answers in public pages and keep only account-specific content gated. > * One question per page, one stable URL per answer. Versioned duplicates split the signal. > * An llms.txt helps agents and coding assistants find your docs. Google says no special AI file is needed, so it is not a ranking lever. ## What AI Cites Documentation For, and What It Doesn't AI answers cite official documentation when the question asks how a product, standard, or platform works. Across the brands we track, docs show up in the most-cited sources only in technical categories. For buyer questions like "best X for Y", AI cites listicles, review roundups, and YouTube instead. We pulled the 50 most-cited URLs per brand over 30 days to October 22, 2026, from our own tracking across ChatGPT, Perplexity, Google AI Overviews, and Gemini. **On RankZero's own prompts**, which are about AI SEO, official docs and help pages hold 4 of the top 50 cited URLs: * Google's guide to optimizing for generative AI features, cited in 7.2% of answers on 18 prompts * Google's "AI features and your website" page, cited in 6.6% of answers on 18 prompts * Google's 2025 "Succeeding in AI search" post, cited in 2.2% of answers on 9 prompts * A Bing Webmaster Tools help page, cited in 2.2% of answers on 10 prompts Those prompts include "How can I improve my website's SEO to appear in AI answer snippets?", which is a how question with an official answer. Google wrote that answer, so Google gets cited. **Across the 12 other brands we track**, official documentation reached the top 50 for exactly one: a smart contract security firm. There, ethereum.org's developer docs on smart contract security are cited on 12 prompts, including questions about scanners and when a protocol needs a new audit. The food, travel, events, ecommerce, and B2B services brands had none. Their buyers ask which product to pick, and AI answers that with lists. **What this means for you:** documentation SEO wins the how prompts in your category. Integration questions, setup questions, "can X do Y" questions. It will not put you in a "best X" answer on its own. For that, see [how LLMs choose their sources](https://www.rankzero.io/guides/how-llms-choose-sources). ## Make Every Docs Page Readable Without JavaScript The first documentation SEO fix is rendering, because an AI crawler that cannot read the page cannot cite it. A [2024 study by Vercel and MERJ](https://vercel.com/blog/the-rise-of-the-ai-crawler) found that "none of the major AI crawlers currently render JavaScript", including OpenAI's GPTBot and ChatGPT-User, Anthropic's ClaudeBot, and PerplexityBot. The two exceptions they named are Gemini, which runs on Googlebot's infrastructure, and AppleBot. Both render JavaScript like a browser. So a docs site built as a single-page app can rank in Google and still be invisible to ChatGPT. The study's own recommendation: "Prioritize server-side rendering for critical content." To check your own docs: 1. Open a docs page and view the page source, not the inspector. 2. Search the source for a sentence from the page body. 3. If it is missing, the text arrives through JavaScript, and most AI crawlers will not see it. Most modern docs frameworks (Docusaurus, Mintlify, GitBook, Fumadocs) serve static or server-rendered HTML by default. The risk sits in custom builds and in widgets: tabs, accordions, and code samples that load their content on click. Then check robots.txt. Blocking GPTBot, ClaudeBot, or PerplexityBot on `/docs` removes you from those engines no matter how good the page is. Our guide to [AI crawlers](https://www.rankzero.io/guides/ai-crawlers) lists the user agents, and our [AI crawler study](https://www.rankzero.io/blog/ai-crawler-study) shows which ones visit. ## Take the Answers Out From Behind the Login A documentation page behind a login cannot be crawled, so it cannot be cited. [Google's guide to generative AI search](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) says its features draw on "publicly accessible, crawlable content". A crawler that lands on a sign-in form has nothing to quote. Many SaaS help centers gate everything, often because nobody decided which parts needed gating. The result is that AI answers "how do I connect X to Y" from a forum thread, a competitor's docs, or a two-year-old blog post. Split the content by what it needs to protect: * **Public:** how-to guides, integration setup, API reference, limits, error messages, and pricing logic explained in words * **Gated:** account-specific data, internal runbooks, customer-only security documents, and anything under NDA If your support team writes the same answer into tickets every week, that answer belongs on a public docs page. It is the question your buyers ask AI too. > **Pro Tip**: search your own brand plus "how to" in ChatGPT and Perplexity. Every answer that cites a third party instead of your docs is a page you are missing or a page AI cannot reach. If you want this checked across your whole site, [book a strategy call](https://www.rankzero.io/services) and we will show you which of your pages AI can read and which it cites. ## Write One Answer per Docs Page AI answers lift a passage, not a whole site. A docs page that answers one question in its first two sentences gives the engine a clean passage to quote. A page that covers six tasks forces it to pick, and it often picks someone else's page. Google's guidance puts it plainly: "Create the content yourself based on what you know about the topic." For docs, that means the facts only you hold: exact limits, exact settings, exact error text. Here is the shape that works: * **A title that is the question or task:** "Connect Google Search Console", not "Integrations (Part 2)" * **A first sentence that answers it:** what it does, where to click, what you need * **Steps as a numbered list,** with the exact UI labels * **One H2 per sub-question,** so each section still makes sense if quoted alone * **Specific numbers:** limits, timings, and plan quotas written out, not "some" or "a few" Our own metric pages follow this. The visibility page opens with "Your visibility score shows how often your brand appears in AI responses", then defines the percentage in the next sentence. An engine asked "what is a visibility score in RankZero" has its answer in two lines. Write for the reader first. The same page that gets cited is the page that saves a support ticket. ## Keep Docs URLs Stable and Versions Under Control A cited URL is only useful if it still works next month. Docs teams rename pages, restructure sidebars, and ship a new version folder for every release, and each move can break the link an AI answer, a forum post, or a customer bookmark points to. Three rules hold up: 1. **Never change a docs URL without a 301 redirect.** Redirect to the page that answers the same question, not to the docs home. 2. **Give the current version one permanent URL,** such as `/docs/api/authentication`, and put old versions under a version path. 3. **Set a self-referencing canonical on every page,** and point near-duplicate version pages at the current one where the content is the same. Google's documentation on [consolidating duplicate URLs](https://developers.google.com/search/docs/crawling-indexing/consolidate-duplicate-urls) ranks the signals by strength: redirects first, then `rel="canonical"`, then the sitemap. Google still decides for itself which page is the duplicate, based on how similar the content is. There is a trade-off. If you canonicalize every old version to the latest, the old versions drop out of search. That is right for a SaaS product everyone runs on the latest release. It is wrong for software where customers run old versions for years, which is why projects like PostgreSQL keep versioned pages indexable. Pick on purpose. Leaving it to chance is how three versions of the same page end up splitting every signal. ## Add an llms.txt, and Know What It Does Not Do An llms.txt is a markdown file at your site root that lists your most useful pages for language models. It was [proposed by Jeremy Howard in September 2024](https://llmstxt.org/), and docs platforms were among the first to pick it up. Docs platforms now generate it for you. [Mintlify, for example, hosts `/llms.txt` and `/llms-full.txt`](https://www.mintlify.com/docs/ai/llmstxt) automatically for every docs site. Be clear about what it does. Google's guide to generative AI search says you "don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search", so for Google an llms.txt is optional. No AI assistant vendor has committed to reading third-party llms.txt files when answering. Where it does help is with agents and coding assistants that go looking for it. A developer who points an assistant at your docs gets a curated map instead of a crawl of your navigation. That is a real use for a SaaS product with an API. So add one, keep it short, and do not expect it to fix a rendering or login problem. Our [llms.txt guide](https://www.rankzero.io/guides/llms-txt) covers the format and how to check it works, and the [free llms.txt generator](https://www.rankzero.io/tools/llms-txt-generator) writes a first version in a minute. ## What We Found Running This on Our Own Docs We checked [RankZero's own docs](https://www.rankzero.io/docs) against this list before publishing it. Here is the honest result. **What holds up:** * 18 public pages in five groups, with nothing behind the login except the app itself * Our robots.txt allows GPTBot, ClaudeBot, PerplexityBot, and the other answer-engine crawlers on `/docs`, and blocks only the app, account, and API routes * Our root llms.txt lists all 18 docs pages, plus the API spec and the MCP server * Metric pages open with a one-sentence definition **What we still need to fix:** * No docs page shows a last-updated date, so a reader cannot tell whether a setting is current * Pages have a "Copy Link" action but no plain markdown copy for people pasting into an assistant **And the finding that matters most:** none of our docs pages is among the 50 URLs AI cites for the prompts we track. That is expected. Our tracked prompts ask which agency or tool to use, and docs do not win those. Docs win the how questions about our own product, which is where they belong. That is the lesson for your team too. Track the how prompts separately from the buyer prompts, or your documentation SEO will look like it does nothing. ## A Documentation SEO Checklist for SaaS Teams Run this documentation SEO checklist on your 10 most-visited docs pages first, then roll it out. Each item is a pass or fail check. 1. The page body is in the HTML source, not loaded by JavaScript. 2. robots.txt allows the AI crawlers you want on `/docs`. 3. The page is public, with no login or email wall. 4. The title names one task or question. 5. The first two sentences answer it. 6. Each H2 answers one sub-question and makes sense on its own. 7. Limits, settings, and error messages are written out exactly. 8. The URL is stable, with 301s from every old path. 9. The page has a self-referencing canonical, and old versions follow a policy you chose. 10. A visible last-updated date sits on the page. 11. Your llms.txt lists the page if it is one of your key answers. 12. You [track the how prompts](https://www.rankzero.io/guides/ai-search-tracking) for your product, so you can see when docs start getting cited. Item 12 is the one teams skip. Without it you are guessing. ## FAQ ### What is documentation SEO? Documentation SEO is optimizing your product docs so search engines and AI answers can find, read, and cite them. It covers rendering, crawl access, page structure, URLs, and versions. For a SaaS team, the goal is that your docs answer how questions about your product, not a forum thread or a competitor. ### Should my docs live on a subdomain or under /docs? Either can rank and get cited if the pages are public and render as HTML. A `/docs` subfolder keeps everything in one site and is simpler to manage in one Search Console property. A subdomain like `docs.example.com` works too, as Google's own `developers.google.com` shows. Pick the one your docs platform supports cleanly and keep it stable. ### Does llms.txt help my docs rank in Google? No. Google says no special AI file is needed for its AI features, and it does not list llms.txt among the things it uses. An llms.txt helps agents and coding assistants that look for it, which is useful for products with an API. Treat it as a map for tools, not a ranking signal. ### Will better docs get my product into "best X" AI answers? Rarely on their own. In our tracking, AI answers buyer questions with listicles, review roundups, and video. Docs win the how questions about your product. To appear in "best X" answers you need to be on the lists AI cites, which is a different job. ### Should I noindex old or internal docs? Noindex internal pages that should never be public, and gate them properly too. For old versions, decide by your customers. If everyone runs the latest release, canonicalize old versions to the current page. If customers run old versions for years, keep those pages indexable and label the version clearly. ### How long does documentation SEO take to show up in AI answers? Fixes like rendering and unblocking crawlers can show up within weeks, once engines re-crawl the pages. Earning citations takes longer, because the page has to be the best answer to a question people are asking. Track your how prompts from day one so you can see the change. ## Start With the 10 Pages Your Support Team Links Most Pull the 10 docs pages your support team sends most often. Run them through the checklist above, fix the rendering and login problems first, then rewrite each one so its first two sentences answer its title. That is documentation SEO in its most useful form: the page that saves a ticket is the page AI cites. If you would rather have it done, we run this as part of our [AI SEO work for SaaS teams](https://www.rankzero.io/guides/saas-ai-seo), together with tracking the how prompts that show whether it is working. [Get Your AI Search Audit](https://www.rankzero.io/services), a 15-minute call with no commitment, and we will show you which of your docs pages AI can read today.