How to rank on LLMs: getting cited by ChatGPT, Claude, Gemini & Perplexity

Why "ranking on LLMs" is now its own discipline

For twenty years the game was simple: rank on the first page of Google, get the click. That game is shrinking. A growing share of your customers now type their question into ChatGPT, Claude, Gemini, or Perplexity and read the answer the model gives them, without ever visiting a website. They don't see ten blue links. They see one synthesized answer, sometimes with a handful of cited sources.

That changes the goal. You're no longer trying to be result number three on a page. You're trying to be the source the model pulls from when it writes the answer, and ideally the business it names. This is answer-engine optimization (AEO), and while it shares DNA with SEO, the winning moves are different enough that treating them as identical will cost you.

Here's the encouraging part: very few small businesses are optimizing for this yet. The pages that get cited are usually whichever ones happen to be structured well, not the ones with the biggest budgets. That's a window.

How do LLMs actually decide what to cite?

An answer engine isn't running your page through a keyword density checker. It's reading your content the way a fast, skeptical research assistant would, and asking three questions: Does this directly answer what was asked? Can I trust it? Can I lift a clean, self-contained statement from it without dragging in fluff?

Pages that win on all three tend to share a shape. They open with a direct answer. They use headings that match how people actually phrase questions. They make specific, checkable claims instead of hedging. And they carry trust signals, reviews, citations, mentions on sites the model already trusts, that tell the engine this source is credible. Miss any one of those and you become the source that almost got cited.

Answer the question first, in the first three sentences

The single biggest mistake we see: burying the answer. A page titled "How much does an AI chatbot cost?" that spends four paragraphs on the history of chatbots before mentioning a number will not get cited. The model wants the answer, and if you don't give it cleanly near the top, it'll find a competitor who does.

Open every page with a two-to-three sentence direct answer to the question in the title. Don't tease, don't set up, don't "in this article we'll explore." State the answer, then spend the rest of the page supporting and expanding it. This is good for human readers too, but for LLMs it's close to mandatory: that opening block is the chunk most likely to be lifted verbatim into an answer.

Structure your pages the way models read them

Answer engines lean heavily on structure to understand and extract content. The concrete moves:

• Phrase your H2 headings as the literal question someone would type or speak, "How much does an AI voice agent cost?" beats "Pricing."
• Put a real FAQ section at the bottom with direct question-and-answer pairs. This is the single most-cited block type across answer engines.
• Add Article and FAQPage schema (structured data) to your HTML. Both Google and the answer engines read it to understand what your page is and which parts are Q&A.
• Keep paragraphs tight and self-contained, so any one of them can be lifted without needing the paragraph before it for context.
• Use lists and tables for anything comparative, models extract structured data far more reliably than prose buried in a wall of text.

Be specific, dated, and concrete, vagueness gets filtered out

LLMs are trained to prefer confident, specific, verifiable claims over vague hedging. "Many businesses see significant improvements" is the kind of sentence a model skips. "Service businesses that answer inbound leads within five minutes book roughly 3x more appointments than those that reply the next day" is the kind it cites, because it's specific, it's checkable, and it stands on its own.

Use real numbers. Name real tools and real examples. Date your claims ("as of 2026") so the model knows the information is current. Counterintuitively, being willing to state a precise, falsifiable claim makes you more citable, not less, because it signals the kind of authority models are tuned to surface.

Off-page signals: why being mentioned elsewhere matters more than ever

On-page structure gets you eligible to be cited. Trust signals get you actually cited. Answer engines weigh third-party validation heavily: reviews, mentions, and citations on sites the model already trusts. If three industry sites and a stack of Google reviews all describe you as "the AI automation agency for home-services businesses," the model absorbs that association and is far more likely to name you for that query.

Practically, that means the old reputation work, getting reviews, earning mentions, being listed and described consistently across the web, is now also AEO work. The model is reading the whole web's opinion of you, not just your homepage.

Don't abandon SEO, AEO is built on top of it

None of this means classic SEO is dead. Answer engines still crawl, index, and trust the same signals search engines do, fast pages, clean HTML, real backlinks, genuine authority. AEO is a layer on top: the same content, restructured to be answer-first and machine-extractable, with schema and FAQs added. Do AEO well and you typically improve your Google ranking too, because answer-first structure and FAQ schema help there as well. You're not choosing between the two. You're future-proofing.

A practical starting checklist

If you want to start ranking on LLMs this quarter, do these in order. Each is a few hours of work and compounds.

1. Pick your three highest-intent buying questions, the ones people ask right before they'd pay you.
2. For each, write or rewrite a page that answers it directly in the first three sentences.
3. Rephrase the page's H2s as real questions and add a 4-6 item FAQ section at the bottom.
4. Add Article + FAQPage schema to the page's HTML.
5. Replace every vague claim with a specific, dated, checkable one, real numbers, named examples.
6. Run the 10-minute citation test, then re-run it monthly and watch your name start to appear.

That's the work. It's not a secret algorithm hack, it's making your best answers the clearest, most trustworthy, most extractable versions of themselves on the internet. That's exactly what answer engines are built to reward.