LLM SEO: What Actually Gets You Cited in AI Answers

Ask ten people how to get your business mentioned in ChatGPT or Google's AI Overviews and you will get ten answers, most of them invented. A whole vocabulary has grown up around it, LLM SEO, GEO, answer engine optimisation, and a lot of it is being sold before anyone has checked what the platforms themselves actually say.

This guide does something simpler. It goes through what Google, Microsoft and OpenAI have published in their own documentation, what the largest public studies have measured, and then gives you four things worth checking on your own site. Every claim below has a source and a date against it. If you are still getting to grips with the basics first, our beginner's guide to SEO covers that ground.

The short version: there is no separate AI ranking system to optimise for, there is no special file you need to publish, and the one technical difference that genuinely matters is whether your page's text is in the HTML your server sends or only appears once JavaScript has run.

What Google, Microsoft and OpenAI actually say

Google publishes a guide to optimising for generative AI features on Search, last updated on 10 July 2026. Three statements in it settle most of the arguments people are still having.

Eligibility is ordinary Search eligibility. In Google's words, "to be eligible to be shown in generative AI features on Google Search, a page must be indexed and eligible to be shown in Google Search with a snippet". Nothing more.

There is no new file or markup to create. Google states that "you don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them". Its companion page, AI Features and Your Website, goes further: "there's also no special schema.org structured data that you need to add".

There is no second ranking system. Google says its generative AI features "are rooted in our core Search ranking and quality systems". Your controls are the controls you already have: robots.txt for Googlebot, plus nosnippet, data-nosnippet and max-snippet if you want to limit what can be displayed.

Microsoft's published guidance, alongside the AI Performance report that entered public preview in Bing Webmaster Tools in February 2026, points in the same direction. It asks for depth and expertise on the page, structure and clarity using headings, tables and FAQs, claims supported by evidence, content kept accurate and current, and IndexNow to notify search engines of changes quickly.

So both of the companies with the most to say about it are describing good SEO, clearly written. That is not a disappointing answer. It is a useful one, because it means the budget you were about to spend on a novelty tactic is better spent on the page.

Ranking still matters, but less than it did

Here is where the picture has genuinely moved, and it is worth being precise about what has been measured.

In July 2025, Ahrefs analysed 1.9 million citations across 1 million AI Overviews and found that 76.1% of cited pages ranked in Google's organic top 10 for the same query. In an update published on 2 March 2026, covering 863,000 keyword SERPs and 4 million AI Overview URLs, that figure had fallen to about 38%. The rest split roughly evenly between positions 11 to 100 and pages that do not rank in the top 100 at all.

Separate the measurement from the explanation, because they are two different claims. The measurement is the fact. Ahrefs offers two causes and is open that both are in play: it improved its own parsing and started catching citations it had previously missed, and Google upgraded AI Overviews to Gemini 3 in January 2026, which appears to expand queries more aggressively. Nobody can currently tell you how the drop divides between the two.

What is not in doubt is the mechanism behind the wider spread. Google describes it in its own documentation as query fan-out: it issues multiple related searches behind a single question so it can "display a wider and more diverse set of helpful links".

For a smaller business this is better news than it first reads. A top 10 position is still the strongest single lever you have, and it is no longer anything like a guarantee. But pages that rank nowhere for the headline term are being cited inside answers to it, which means you do not have to outrank the biggest site in your market to be the source quoted on a narrow, specific question within it.

The one technical difference that is real

This is where LLM SEO genuinely departs from SEO, and it is the one item on this page that can be failing silently.

Google renders JavaScript. Its guide confirms that Google "is able to process content within JavaScript as long as it isn't blocked". The crawlers feeding ChatGPT, Claude, Perplexity and the rest are a different matter. Vercel measured crawler behaviour across its network and published the results on 17 December 2024: none of the major AI crawlers rendered JavaScript, including OAI-SearchBot, ChatGPT-User, GPTBot, ClaudeBot, Meta-ExternalAgent, Bytespider and PerplexityBot. They fetch JavaScript files without executing them. 11.50% of ChatGPT's requests and 23.84% of Claude's were for JavaScript files that were never run.

One honest caveat. That measurement is from December 2024, and none of the crawler operators publish a rendering policy, so treat it as the best public evidence available rather than a permanent law.

Why it matters: if your site assembles its pages in the browser, the HTML served to a crawler that does not run JavaScript can contain your navigation, your footer and very little else. Google will still see the full page. An AI crawler sees an empty shell, finds nothing to quote, and quotes somebody else. The fix is not a plugin. It is server rendering or pre-rendering, so the words exist in the HTML before any JavaScript runs, and we have written about that underlying choice in SSR vs CSR and what it does to your SEO.

Which AI crawlers to allow, and what each one gets you

OpenAI documents its crawlers separately, and choosing between them is a business decision rather than a technical one.

OAI-SearchBot surfaces sites in ChatGPT's search features. OpenAI states that sites which disallow it "will not be shown in ChatGPT search answers, though can still appear as navigational links".

GPTBot crawls content that may be used to train OpenAI's foundation models. Disallowing it signals that your content should not be used for training.

ChatGPT-User handles fetches a person has asked for, when someone tells ChatGPT to go and look at a page. OpenAI notes that "because these actions are initiated by a user, robots.txt rules may not apply".

The distinction people get backwards: blocking GPTBot does not remove you from ChatGPT's answers, and blocking OAI-SearchBot does. If you want the visibility but not the training use, those are two different lines in your robots.txt file. If you are blocking the wrong one, you have made yourself invisible for no benefit at all.

Google works differently. AI Overviews and AI Mode are served by Googlebot, so, as Google puts it, "robots.txt directives for Googlebot is the control for site owners". There is no setting that keeps you in Google Search but out of its AI features. Google-Extended is a separate crawler covering training and grounding in Google's other products, and it does not govern your appearance in Search.

A four step check you can run on your own site today

None of this needs a tool or a subscription, and only the first needs anyone else.

1. Find out whether your pages are rendered on the server

This is the one that needs a conversation rather than a setting. Put a single question to whoever built or maintains your site: does the main copy on our service pages arrive in the HTML from the server, or is it assembled in the browser after the page loads? If the answer is the second one, or nobody is sure, that is the first thing to put right, because a crawler that does not run JavaScript never sees content that only exists after it runs.

Ask about service pages and blog pages separately. They are frequently built in different ways on the same site and the answer is not always the same for both.

2. Check what your robots.txt file allows

Open yourdomain.co.uk/robots.txt in a browser. Look for Disallow rules aimed at GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot and Google-Extended, and check what your blanket User-agent rule does. If you have not opened that file since these crawlers first appeared, the rules in it may be plugin defaults you inherited rather than choices you made. Every line in there should be deliberate.

3. Check you are indexed and snippet eligible

Google's stated requirement is that the page is indexed and can be shown with a snippet, so confirm both. Run your key pages through the URL Inspection tool in Search Console. Then make sure nothing is restricting snippets: nosnippet, a low max-snippet value and data-nosnippet all limit what Google is permitted to display, and any of the three can be sitting in an SEO plugin setting or a page template that nobody has revisited. Your developer or SEO plugin will tell you in a minute which are in use. If inspection turns up problems you cannot explain, that is usually a sign you need a proper technical audit rather than a content fix.

4. Turn on the two free reports

Search Console has a generative AI performance report covering AI Overviews and AI Mode, available to all sites worldwide since 31 August 2026. It reports impressions, meaning how many times links to your site were shown inside a generative AI feature. Bing Webmaster Tools has an AI Performance report, in public preview since February 2026, which reports citations in Microsoft Copilot and Bing's AI answers along with the grounding queries behind them. Both are free, and both take minutes to set up. Neither of them tells you anything about ChatGPT.

What actually improves your chances

Nothing on this list is exotic. All of it follows from what Google and Microsoft publish.

Answer the question in the first two sentences under the heading that asks it. A model pulling an answer out of a page takes the passage that contains it, not the paragraph you were building towards.

One question per heading, phrased the way a customer would say it. Headings are the map.

Structure the facts. Tables, short lists and genuine FAQ sections are what Microsoft names explicitly, and they are far easier to lift cleanly than a wall of prose.

Put evidence next to claims. A figure with a source and a date attached is quotable. A superlative is not.

Keep the page current and say when it changed. Freshness is in Microsoft's guidance and it is visible to a reader as well.

Be specific about who you are and where you work. A large share of these answers are resolving a local or sector question, not a general one.

What to ignore

llms.txt. Google says plainly that Search does not use machine readable AI text files, and no major assistant currently documents support for one. Publishing it costs almost nothing, so publish it if you like, but do not record it as work done.

"AI schema". There is no special structured data for AI features, on Google's own word. Ordinary structured data is still worth having for all the ordinary reasons.

Guaranteed placements. There is no ranking system to buy into and no reliable public position data for AI answers, so a guaranteed position in one cannot be a real offer.

More AI written filler. Depth, evidence and expertise are what both Google and Microsoft describe. Volume on its own is not what either of them is asking for.

Frequently asked questions

What is LLM SEO? It is the practice of making your pages usable as a source by large language model tools such as ChatGPT, Google's AI Overviews and AI Mode, Microsoft Copilot and Perplexity. In practice it is technical accessibility for crawlers that do not run JavaScript, plus content clear enough to be quoted accurately.

Is LLM SEO different from SEO? Mostly no. Google states that its generative AI features run on its core Search ranking systems and need no special optimisation. The one substantive difference is rendering: Google executes JavaScript and the AI crawlers, on the best public evidence, do not.

Do I need an llms.txt file? No. Google confirms Search does not use one, and no major assistant currently documents support for it. It is optional and unproven, not a requirement.

How do I get into Google's AI Overviews? Be indexed, be eligible for a snippet, and be a good answer to the question. Ranking in the top 10 helps considerably, though Ahrefs measured that only about 38% of cited pages did so in March 2026, down from 76.1% in July 2025.

Does AI Overview optimisation need different content from normal SEO content? It needs the same content, organised better. Clear question led headings, the answer stated immediately underneath, facts in tables or lists, and sources against figures. That is what makes a passage extractable.

Will blocking AI crawlers protect my content? It changes what you get, so decide deliberately. Blocking GPTBot signals no training use but does not remove you from ChatGPT's answers. Blocking OAI-SearchBot does remove you from ChatGPT search answers. With Google there is no split at all: robots.txt for Googlebot governs both Search and its AI features together.

How do I measure it? The generative AI performance report in Google Search Console for AI Overviews and AI Mode, and the AI Performance report in Bing Webmaster Tools for Copilot and Bing's AI answers. Both are free. There is no equivalent first party report for ChatGPT.

Where to start

Start with step one, because it is the only item here that can be failing without showing up anywhere else. Search Console will not flag it, your rankings will not obviously suffer, and Google will carry on seeing the page perfectly well. Settle that question before you spend a penny on anything labelled AI visibility, because no amount of better writing helps a crawler that never receives the words.

After that, this is ordinary SEO done carefully: be indexed, be clear, be specific, and put evidence behind what you claim. If you want to know which of those your own site is actually failing on, take a look at our SEO services and get in touch, and we will tell you if you do not need us yet.