LLM SEO: How to Get Your Site Cited in AI Answers

Updated October 30, 2026, by the Rillow team. We make Rillow, an SEO data server for AI agents, and every search result quoted in this post came from it on October 30. There are no affiliate links here.

LLM SEO is the work of getting your site named and linked in answers from large language models: ChatGPT, Claude, Perplexity, Gemini and Google's AI Overviews. Most of it is ordinary SEO, because these answers are built from pages the AI found in a search index. The rest is making your pages easy to fetch and easy to quote, and getting your product mentioned on the pages AI answers already cite. This guide shows how AI answers pick their sources, how to check whether yours is one of them, and the fixes worth your time, in the order we'd do them.

What LLM SEO means

LLM SEO goes by several names. Generative engine optimization (GEO), answer engine optimization (AEO) and LLM optimization (LLMO) all describe the same goal, which Google's own AI answer for "llm seo" summed up on October 30 as getting AI platforms to "cite, summarize, and recommend you". The names differ in which AI product the writer has in mind, not in what you do.

The useful split is between the two ways an AI model can know about your site:

1. From training data. The model read your pages, or pages about you, before it was trained. You can't control this much, and it changes only when a new model ships.
2. From a live search. ChatGPT search, Claude with web search, Perplexity and Google's AI Overviews search the web when you ask, read a handful of pages and cite them. This is the part you can change this week.

Almost all of LLM SEO is about the second path. If your page ranks well for a search and says something useful in plain words, it has a good chance of being one of the pages an AI answer reads.

How AI answers pick their sources

Google says it plainly in its guide to AI features and your website: there are no extra requirements to appear in AI Overviews or AI Mode. A page has to be indexed and allowed to show a snippet, and the same ranking systems decide what gets used. So for Google, LLM SEO and SEO are the same job.

ChatGPT search and Claude's web search run their own searches and fetch pages with their own crawlers. A page they can't fetch, or a page whose text only appears after JavaScript runs, gives them nothing to quote.

To see what that looks like in practice, we pulled Google's AI answer for three searches with Rillow on October 30, 2026, and listed the sites each answer cited:

• "llm seo": llmpulse.ai, dwao.in, llmrefs.com, getmorphic.com, plus YouTube, LinkedIn and Reddit.
• "ai visibility checker": ahrefs.com, frase.io, semrush.com, honeyb.ai, arobis.ai, plus Reddit and YouTube.
• "how to rank in chatgpt": powerdigitalmarketing.com, ahrefs.com, neilpatel.com, plus Reddit, LinkedIn and YouTube.

Two things stand out. Small sites like llmpulse.ai and arobis.ai sit next to Ahrefs and Semrush, so a young site can get cited. And Reddit and YouTube appear in all three answers, which tells you where else your product should be talked about.

Check whether AI answers cite you

Start by measuring, so you know whether a change worked. Pick the five to ten searches a customer would type before they find you, such as "invoice app for freelancers" rather than your brand name.

For each search, check three places:

1. Google. Search it in a private window and open the AI Overview if one shows. Note the cited sites. Our free AI Overview checker does this for you: type the search and your site, and it shows whether Google gives an AI answer, which sites it cites, whether yours is one, and whether Claude names you for the same question.
2. ChatGPT. Ask the search as a question with search turned on, and look at the sources it lists.
3. Claude. Ask the same question with web search on.

Write the results in a spreadsheet with the date. AI answers change from day to day, so one check is a sample, not a verdict. Checking the same searches every week tells you more than any single run.

If you work with an AI agent like Claude Code or Codex, it can run these checks for you. Rillow's get_serp tool returns Google's AI answer and its cited sites for any search, so a prompt like this works once Rillow is connected (setup takes a minute on the SEO MCP server page):

For each of these searches, get Google's results with Rillow and tell me
whether an AI answer shows, which sites it cites, and whether example.com
is cited or in the top 10: [your searches]

Let AI crawlers read your site

AI companies fetch pages with named crawlers, and your robots.txt decides which ones get in. The ones that matter for answers today:

• OpenAI: OAI-SearchBot for ChatGPT search results, ChatGPT-User when a user asks ChatGPT to open a page, and GPTBot for training.
• Anthropic: Claude-SearchBot for search, Claude-User for user requests, and ClaudeBot for training.
• Perplexity: PerplexityBot and Perplexity-User.
• Google: AI Overviews use the normal Googlebot. The separate Google-Extended token only controls whether Google may use your pages to train Gemini models, and Google says it doesn't affect Search.

Check that you aren't blocking them by accident, for example through a firewall rule or a robots.txt copied from a template. Open yoursite.com/robots.txt and look for Disallow: / under any of those names. If you want to be cited but not used for training, you can allow the search crawlers and block the training ones:

User-agent: OAI-SearchBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

User-agent: ClaudeBot
Disallow: /

Rillow's free site report on the home page checks this for you, along with whether your pages work without JavaScript.

Make your pages readable without JavaScript

AI crawlers often fetch the raw HTML of a page and don't run its JavaScript. If your site is a single-page React app, the HTML they get can be an empty <div id="root"> with nothing to quote.

You can test this in ten seconds. Run curl -s https://yoursite.com/your-page | grep "a sentence from the page" in a terminal. If the sentence comes back, crawlers see your text. If it doesn't, you need server-side rendering or prerendering. Next.js, Astro and Remix render HTML on the server by default, and sites built with app builders often need a setting or a rebuild. Our guide to Lovable SEO walks through that case.

Write passages an AI can quote

AI answers quote one passage at a time, so each section of your page has to make sense on its own. A few habits help:

• Answer the question in the first sentence under each heading, then explain.
• Name the thing instead of writing "it" or "this tool", so the passage still makes sense when it is lifted out.
• Put the facts in the text: prices, limits, dates, steps and comparisons as real HTML lists and tables, not inside images.
• Link the source of every number. A claim with a source is easier for an answer to repeat.
• Keep the date on the page current when you update it, and say what changed.

Use headings that match how people ask. "How much does X cost?" is a better heading than "Pricing philosophy", because it matches the question an AI answer is trying to answer.

Get mentioned where AI answers already look

The three AI answers above cited Reddit, YouTube and LinkedIn, plus lists of tools on other people's sites. AI answers lean on these because they show many people's opinions in one place. A product that no one else mentions is hard for an answer to recommend, however good its own pages are.

What works without a budget:

• Answer real questions on Reddit and in communities where your customers ask them, and mention your product only where it fits. Moderators remove plain ads.
• Ask the authors of "best X tools" lists to review your product. Many of them update their lists every few months.
• List your product in the directories for your category, such as MCP directories for an MCP server or app stores for an app.
• Record a short screen video showing the product solving one problem and post it on YouTube with a plain title.

What you can skip for now

Some LLM SEO advice costs time and has little evidence behind it:

• A separate copy of each page written "for AI". Search Engine Land compared LLM-only pages to the old keywords meta tag: anyone can add them and the systems ignore them.
• An llms.txt file. It is a proposed file that lists your important pages for language models. It takes ten minutes to add, and it won't hurt, but Google's AI Overviews come from Google's normal index, so it won't get you into them.
• Paying for an "AI visibility score" before you have the basics. A score tells you less than the list of sites an answer cites, and that list is free to check.

A weekly LLM SEO routine

Here is a weekly routine you can run yourself, or hand the lookups to an agent:

1. Check the same searches as last week in Google, ChatGPT and Claude, and note which sites each answer cites.
2. For any search where a competitor is cited and you're not, open their cited page and see what it says that yours doesn't.
3. Update one page with that missing fact, a clearer answer under the heading, or a source.
4. Find one place where your customers ask about the problem you solve, and answer one question there.

If you run your SEO through an agent, read our guides to agentic SEO and the Claude SEO skill, which show the same loop with real tool output.

Frequently asked questions