How to Get Cited by ChatGPT, Perplexity, and Google AI

More and more people aren't typing a search into Google anymore. They're asking an AI directly — ChatGPT, Perplexity, Google AI Overviews — and trusting the answer it gives them, without clicking through to any site. If your business isn't structured for those AI systems to find you and cite you with the right information, you simply don't exist in that conversation, no matter how good your site is for a traditional search engine.

This is GEO — Generative Engine Optimization. It doesn't replace traditional SEO; it's a layer on top of it. Here's what actually moves the needle, without tricks or guaranteed-results promises nobody can keep.

Step one: make sure AI engines can actually get in

Before optimizing anything, confirm AI crawlers can access your site. Each platform has its own bot:

• GPTBot and ChatGPT-User (OpenAI / ChatGPT)
• PerplexityBot (Perplexity)
• ClaudeBot (Anthropic / Claude)
• Google-Extended (Google Gemini and AI Overviews)
• Bingbot (Microsoft Copilot, via Bing)

Check your robots.txt. If any of these are blocked with a Disallow rule, that engine literally cannot cite you — it's the single most common self-inflicted mistake, and the easiest one to fix. A robots.txt with Allow: / for all user-agents blocks none of them.

Add an llms.txt — a map you hand AI systems directly

An llms.txt is a file at your site's root, similar in concept to robots.txt, that gives AI engines a structured, reliable summary of who you are, what you offer, and where to find verified information — instead of letting the AI guess from scattered fragments of your site. It's not required (Google has said explicitly that no special markup is needed for AI Overviews), but non-Google engines like ChatGPT and Perplexity do make use of it when it exists, and almost no small business has one yet — a cheap edge while that's still true.

Citable content: structure matters more than volume

AI systems don't cite whole pages, they extract passages. Every key claim on your site should be able to stand on its own, out of context:

• Answer directly before you explain. If someone asks “what is GEO?”, the first sentence should answer it — not warm up for three paragraphs first.
• 40-60 word answer blocks are the sweet spot for a generative engine to extract a complete, citable answer.
• FAQs written in natural language — the exact way someone would actually ask an AI, not marketing jargon.
• Tables for comparisons, numbered lists for processes. Long paragraphs get lost; structure gets extracted.

Structured data (schema.org)

An FAQPage, Organization, or Service schema in JSON-LD gives AI systems explicit context they'd otherwise have to infer from visible text. This isn't optional for serious technical SEO, and for GEO specifically it reinforces the exact same information already on the page, in a format a model can parse without ambiguity.

The part almost nobody does: verifiable authority

This is where most businesses fall short, and where the real difference shows up. Public research on generative engine optimization (Princeton, presented at KDD 2024) found that adding citations and sources to content can boost its visibility in AI-generated answers by roughly 40%, and adding sourced statistics by roughly 37%. Keyword stuffing, by contrast, actively reduces visibility — the same kind of trick that already backfired in traditional SEO backfires worse here.

In practice, that means:

• A real name behind the business. A site with no identifiable person — just a brand — gives AI systems (and Google, via E-E-A-T) fewer real expertise signals to trust.
• Real numbers, not invented ones. If you don't have a verifiable figure yet, don't publish one. No statistics beats fake statistics — an AI that catches an inconsistency stops citing you, and a customer who catches it stops trusting you.
• Mentions outside your own domain. Brands get cited noticeably more often through third-party sources — directories, reviews, mentions on other sites — than from their own blog. Your own content is the foundation, not the whole job.

What not to do

• Don't write separate content “for AI” — Google treats this as scaled content abuse.
• Don't chunk your content into artificial fragments thinking it helps extraction — normal structure (headings, paragraphs) already works.
• Don't block AI bots if your goal is to get cited — the most common contradiction we see.
• Don't fabricate third-party mentions, reviews, or quotes to force authority you don't have yet.

The example we actually have on hand

We're not going to invent a success story that doesn't exist yet — this site (blackindex.dev) is about to launch. But you can verify today that it has a real llms.txt, a robots.txt with no AI bot blocks, and FAQPage/Organization structured data on every service page — that's the technical implementation described above, not a promise. Once there's real data on how often an AI actually cites us, we'll share it here, with a source and a date, not before.