LLM SEO: how to get cited by LLMs

How LLMs pick which sources to cite

The key insight behind LLM SEO is that most answer engines don't answer from the model's memory alone — they retrieve. When someone asks a question, the engine expands it into many related sub-queries (query fan-out), searches each against a live index, reads across the returned pages, and composes a single answer that cites the passages it can quote confidently. That pipeline tells you exactly what to optimize:

• Be retrievable. If you're not crawlable and in the index, you can't be cited — full stop. Retrieval is the gate everything else passes through.
• Be relevant to the fan-out. You're not competing for one keyword but for a cloud of sub-queries. Content that covers the whole neighbourhood of a topic gets pulled into more of them.
• Be quotable. Engines prefer clear, factual, self-contained passages they can lift without hedging. Being citable is a content-design problem, not just an authority one.
• Be talked about. Off-site signals feed the model's trust — Ahrefs found brand mentions correlate with AI visibility far more than backlinks (0.664 vs 0.218).

And being cited is not the same as ranking. Across 863,000 SERPs, only about 37.9% of the URLs AI engines cite also rank in Google's top 10 (Ahrefs) — so the retrieval-and-citation game has its own rules, distinct from classic ranking.

The no-JS crawler problem

Here is the technical failure that quietly sinks otherwise good sites. Many AI crawlers — GPTBot, OAI-SearchBot, PerplexityBot and others — fetch your raw HTML and don't execute JavaScript. If your site is a client-rendered single-page app that paints its content after the browser runs your JS, the crawler can see an empty shell: no headings, no answers, nothing to quote. You can have the best content in your category and still be structurally invisible to LLMs.

The fix is to put your important content in the initial HTML response — via server-side rendering, static generation or prerendering — so a crawler that never runs JS still gets the full text. Pair that with a robots.txt that doesn't block the AI crawlers you want (many sites block them by accident), and fast, stable responses so crawls succeed. These are unglamorous fixes, but they're the highest-leverage moves in LLM SEO because they gate everything downstream.

Structured data and entity clarity

Once a model can read your page, structured data helps it parse what the page is and which passage answers which question. Schema.org markup that mirrors your visible content is the tool: FAQ and HowTo schema map questions to answers, Article and Organization schema clarify authorship and identity, and Product schema clarifies what you sell. The non-negotiable rule is honesty — only mark up content that's actually on the page. Injecting answers, prices or ratings you don't display violates Google's structured-data policy and erodes the exact trust you're trying to earn.

Structured data works alongside entity clarity: the model needs to know, without ambiguity, who you are. Consistent facts about your brand — across your own pages, your schema and third-party sources — let a model treat you as one unambiguous entity worth citing, rather than a fuzzy match it routes around. Contradictory names, descriptions or claims are a reason to be skipped. This is where the technical LLM SEO layer meets the broader AI search optimization playbook.

Where LLM SEO fits

LLM SEO is the technical slice of a small family of overlapping practices. GEO (generative engine optimization) is about being mentioned and recommended inside generated answers; AEO (answer engine optimization) is about being the extracted answer to a specific question; and AI search optimization is the end-to-end umbrella across every surface. LLM SEO focuses on the how — the retrieval, rendering, schema and entity mechanics that decide whether a model can find and cite you at all. Get the technical layer right and the content and off-site work above it actually pays off.

Measuring LLM visibility

LLM visibility is how often, and how favourably, large language models mention and cite your brand when someone asks a buying question — the LLM SEO equivalent of a rank tracker. Classic SEO tools don't watch it, so you need a purpose-built tool that prompts the engines the way your buyers would and tracks your citation and recommendation share across ChatGPT, Perplexity, Gemini and Google AI Overviews over time.

The catch is volatility. Independent tracking finds 40–90% of the domains an engine cites can change when the same question is re-asked (Profound), so a tool that reports a single number can send you chasing noise. steek tracks LLM visibility across every major engine and puts a 95% confidence interval on every metric, so you can tell a real movement from randomness before you act. It shows the real monthly search volume behind each prompt, offers per-SKU AI Shopping visibility tied to revenue in GA4 for stores, is EU-hosted, and ships a free live checker with no signup (10-day trial, no card). For where it sits among alternatives, see the best AI visibility tools roundup or compare steek against specific rivals. For an engine-specific walkthrough, read how to rank on ChatGPT.