AI-era SEO: how to get cited by ChatGPT, Perplexity, and Google AI Overviews

The funnel changed shape: crawl → cite → click

For twenty years the search funnel was crawl → rank → click: a crawler read your page, an algorithm ranked it, and a human clicked a blue link. A growing share of queries now runs crawl → cite → click: a model reads your page, decides whether to cite it inside a generated answer, and the human clicks through from that citation — if they click at all. The blue links still exist, but an AI answer increasingly sits in front of them.

That single change reorders your priorities. Ranking tenth on a page nobody scrolls to is worth little; being the source a model quotes in the answer everyone reads is worth a great deal. To be cited you have to clear two independent bars: the model has to be able to parse you (legibility), and it has to have a reason to quote you over the alternatives (info-gain). Miss either and you’re invisible in the fastest-growing discovery channel. This is exactly what our AI-SEO service is built to deliver.

Legibility: build a clean entity graph

A model decides whether to trust and cite you partly on whether it can cleanly answer three questions: Who is this? What do they offer? How do their pages relate? An entity graph answers all three in machine-readable form. The discipline is to have one canonical Organization node — a single @id that every other node references — rather than a scatter of unlinked schema blocks.

On this site, for example, there is exactly one Organization at /#organization, one WebSite at /#website, and every service, location, industry, and guide page hangs a Service, WebPage, or Article node off those two roots via provider, publisher, isPartOf, and about references. The result is a connected graph a machine can walk, not a pile of disconnected snippets.

• One root entity. A single Organization @id, reused everywhere — never redefined per page.
• Cascading nodes. Service/WebPage/Article nodes reference the root by @id, so relationships are explicit.
• Cross-references, not duplication. Link nodes by @id instead of restating the same entity in five places.
• Consistent facts. The same name, URL, and contact info in schema, on-page, and in your llms.txt — contradictions erode trust.

llms.txt: a machine-readable identity file

Just as robots.txt tells crawlers where they may go, an llms.txt file gives language models a clean, curated summary of who you are and what your key URLs are — in plain markdown, without the navigation, scripts, and clutter of a rendered page. A companion llms-full.txt carries the extended version.

It is not a magic ranking lever, and no model is obligated to read it. What it does is remove ambiguity: when a model or its tooling does consult it, you’ve handed over a distilled, contradiction-free statement of your entity and your most important pages. That’s cheap insurance for legibility, and it pairs naturally with the entity graph — same facts, two formats. We ship one on every site we build, including this one, and keep it in sync with the sitemap.

Worth quoting: the info-gain gate

Legibility gets you readable. Info-gain gets you cited. Info-gain is the discipline of publishing something the rest of the web isn’t already saying — a specific number, a real framework, a real trade-off, a worked example. A model has no reason to quote the tenth page repeating the same generic advice; it quotes the page that adds a fact it couldn’t get from the other nine.

• Add a number. A worked example, a real range, a concrete threshold — even when clearly framed as illustrative — beats a vague adjective.
• Name the trade-off. Say when your own approach is the wrong choice. Being clear about limits reads as expertise and earns trust.
• Answer the actual question. Lead with the direct answer in the first two sentences, then support it — models lift direct answers.
• Structure for extraction. Clear headings, short answer blocks, and tables give a model clean, quotable units.

Speakable and FAQPage schema

Two schema types directly help assistants quote you accurately. FAQPage marks up question/answer pairs, giving models pre-packaged, self-contained answers to lift — which is why every hub and guide on this site carries a real FAQ block. Speakable (a SpeakableSpecification with CSS selectors) flags the passages best suited to be read aloud by a voice assistant, typically your H1, lead, and FAQ answers.

Neither forces a citation. What they do is lower the friction: you hand the model a clean, bounded chunk of text that already answers a question, so quoting you is the path of least resistance. Combined with the entity graph, this is the structured half of the work — the content is the other half. Both matter to industries that live on being found, from professional services to home & trade services to health & wellness.

A practical checklist

1. Define one canonical Organization entity and reference its @id from every page node.
2. Add WebPage + Article/Service nodes that cross-reference the root by @id.
3. Publish llms.txt and llms-full.txt, and keep their facts identical to your schema.
4. Run every page through the info-gain gate — cut or rewrite anything that adds nothing.
5. Mark up FAQs with FAQPage and flag key passages with Speakable.
6. Lead each page with a direct-answer intro a model can lift verbatim.
7. Keep names, URLs, and contact details identical across schema, page, and llms.txt.
8. Measure citation presence (do assistants quote you?) alongside classic rankings.

That checklist is, essentially, the OptimoWeb method — and it’s what runs under this very site. If you want it applied to yours, our AI-SEO service builds the graph, the identity files, and the info-gain content together, and our web builds bake the structure in from the first commit. Start a project to put your business in front of the assistants people now ask first.