LLM SEO: the engineering behind getting quoted.
Large language models do not rank pages. They retrieve passages, judge whether a passage answers the question, and quote it with a source. LLM SEO is the work of making your pages the passages that get retrieved. Most of it is plain HTML done properly, and we can show you exactly what changes.
// what a retriever actually keeps const passage = { question: "how long does a site migration take?", answer: "Most 500-page migrations take 4-6 weeks...", source: "denverseo.io/website-migration-seo/", signals: ["direct", "specific", "in-HTML", "consistent"] };
LLM SEO is the practice of structuring a website so large-language-model systems (ChatGPT, Gemini, Claude, Perplexity and Google AI Mode) can retrieve, trust and quote its pages. In practice it means serving complete answers in the initial HTML, keeping business facts identical everywhere, using valid structured data, answering specific questions in self-contained passages, and letting the AI search crawlers in. Google’s own guidance says no special file or markup is required; it is still SEO, done at the code level.
The phrase gets sold as something new. It is not. Every LLM-backed search product works the same way under the hood: a query goes in, a retrieval system pulls candidate passages from an index, the model composes an answer and attaches the sources it leaned on. If your page is retrievable, specific and trustworthy, it gets quoted. If it is a wall of marketing adjectives rendered by JavaScript after load, it does not.
This page is the definition, the engineering checklist we use, the things vendors sell that do nothing, and how to measure whether it worked. If you want the business-owner version with pricing, the agency side of Eye To Ad Media covers AI SEO as a service; this page is the code.
How an LLM search system actually reads a page
Three steps, and each one has a failure mode you can fix in code.
- Fetch. A crawler (OAI-SearchBot, PerplexityBot, ClaudeBot, Googlebot for AI Mode) requests the URL and receives whatever your server returns. Many of these crawlers do not execute JavaScript; Google renders it in a later, separate pass. If the answer is injected client-side, several engines never see it. Our note on what AI crawlers actually read lists what each one fetches.
- Chunk and index. The page is split into passages, roughly heading-by-heading. A passage that starts with the question and contains the answer in the first two sentences is a clean chunk. A passage that needs the three paragraphs above it to make sense is a broken one.
- Retrieve and cite. For a given query the system pulls the passages that match best and scores them on specificity, consistency with other sources, and the trust it has in the domain. Then it writes, and attaches the source. Seer Interactive’s 2026 click study found a cited page earns roughly double the clicks per impression of an uncited one on the same query, which is the whole reason to care.
What actually changes in the code
This is the checklist we run on a client site. Each item is verifiable in View Source or a validator, which is how you can tell real LLM SEO from a slide deck.
| Change | Why a retriever cares | How to verify |
|---|---|---|
| Answer in the initial HTML | Non-rendering crawlers only see server output. Prices, FAQs and service details that appear after a script runs are invisible to them. | Disable JavaScript and reload. If the answer is gone, it is gone for several engines. |
| Question-shaped headings, self-contained passages | Chunks are built around headings. A heading that states the question and a first sentence that answers it is a chunk that can be quoted alone. | Read any H2 plus its first two sentences in isolation. Does it stand up? |
| One canonical set of business facts | Retrievers cross-check name, address, phone, hours and service area across your site, your Business Profile and directories. Disagreement reads as uncertainty. | Diff the facts on your About page, your schema and your Google listing. |
| Valid, connected structured data | Organization, Service, FAQPage (where the FAQ is real and visible) and BreadcrumbList give the system typed facts. Disconnected or invalid nodes are ignored. | Run the Schema.org validator and Google’s Rich Results Test; look for one connected @graph. |
| Crawler access | If robots.txt blocks OAI-SearchBot or PerplexityBot, you are not in those indexes at all. | Our AI crawler checker reads your robots.txt and tells you who is in. |
| Specific numbers, dated | “Most migrations take four to six weeks” is quotable. “We deliver fast results” is not. A date on the page tells the model the figure is current. | Count the numbers on the page. If there are none, there is nothing to quote. |
| Stable URLs and fast responses | Crawl budgets for AI bots are small. Redirect chains, 5xx errors and five-second responses waste the visit. | Server logs; our log file analysis guide shows how to read them. |
What to skip (and why vendors sell it anyway)
- llms.txt as a ranking lever. Google’s Search Central changelog (June 15, 2026) says Google Search does not use it. Ahrefs’ June 2026 study of 137,210 domains found 28% publish one and 97% of those files received zero requests in May. We keep one on this site because it costs nothing; we do not bill for it. Full write-up: llms.txt does nothing for Google Search.
- Special “AI schema.” There is none. Google’s AI optimization guide (updated July 10, 2026) is explicit that no new markup or machine-readable file is needed and that pages do not need to be broken into fragments.
- Prompt-stuffing and hidden text aimed at models. Google’s spam policies were updated in May 2026 to cover attempts to manipulate generative AI responses. Hidden instructions in your HTML are now a spam policy issue, not a clever hack.
- Mentions bought from content farms. Google’s guide notes that seeking inauthentic mentions “isn’t as helpful as it might seem.” Retrieval systems weight the source; a citation from a site nobody trusts transfers nothing.
How to measure LLM SEO
Three instruments, in order of how much they cost.
- A prompt transcript. Twenty to forty real customer questions, run through each engine on the same day, logged with the answer, the businesses named and every cited URL. Repeat monthly. It is the only direct measure, and it is free apart from time. The agency side publishes a six-prompt self-check version as an AI visibility audit.
- Search Console’s generative AI report. Since August 31, 2026 it shows impressions from AI Overviews and AI Mode as their own line. It reports impressions only, so pair it with the transcript to see which impressions became citations.
- Referrals and branded search. SE Ranking’s June 2026 study across 101,574 sites put AI referrals at 0.32% of traffic, with ChatGPT sending about three quarters of it. Small, but Similarweb data reported by HubSpot shows those visitors convert at roughly double the rate of ordinary organic traffic. Watch the referrer line in analytics and the branded-query line in Search Console; both rise when citations work.
What we do not report: a made-up “AI visibility score” with no transcript behind it. If a vendor cannot show you the prompts, the number is decoration.
LLM SEO engineering as a service
We do this work on hand-coded, WordPress and Shopify sites, usually inside a technical SEO engagement: the HTML and rendering fixes, the structured data graph, the entity consistency pass, the passage rewrite on the pages that matter, and the monthly transcript. Engine-specific notes live on their own pages: ChatGPT, Perplexity, Gemini, Claude, Grok and Google AI Overviews. Start with a free technical review and we will show you what each crawler currently receives from your server.
Questions, answered straight
What is LLM SEO?
LLM SEO is structuring a website so large-language-model search systems such as ChatGPT, Gemini, Claude, Perplexity and Google AI Mode can retrieve, trust and quote its pages. It is done in the HTML, the structured data and the business facts, not with a special file.
Is LLM SEO different from GEO or AEO?
Same goal, different vocabulary. Generative engine optimization and answer engine optimization describe optimizing for AI answers; LLM SEO emphasizes the retrieval mechanics. Google’s guidance treats all of it as ordinary SEO.
Do I need llms.txt for LLM SEO?
No. Google says Search does not use it, and Ahrefs found 97% of published llms.txt files got no requests in May 2026. It is harmless, not a lever.
Does JavaScript rendering hurt LLM SEO?
Often, yes. Several AI search crawlers do not execute JavaScript, so any answer that only appears after a script runs is invisible to them. Server-render or pre-render the pages that matter.
How do you measure whether LLM SEO is working?
With a monthly prompt transcript across engines (who is named, which URLs are cited), Search Console’s AI Overviews and AI Mode impression line, and the AI referral and branded-search trends in analytics.
Sources we cite on this page
- Google Search Central: AI features and your website (AI optimization guide, updated July 10, 2026)
- Google Search Central changelog (llms.txt note, June 15, 2026)
- Google spam policies (generative AI manipulation added May 2026)
- Ahrefs: llms.txt study (June 2026)
- Seer Interactive: AI Overview impact on CTR, 2026 update
- SE Ranking: AI traffic research study (June 2026)
- HubSpot: AEO trends (Similarweb conversion data)
- Google: new controls and reporting for website owners (AI Overviews and AI Mode impressions)
- OpenAI: overview of OpenAI crawlers
Let's make your site the one they choose.
Fifteen minutes on the phone usually tells us whether we can help, and how much.