SEO for AI: how to get ChatGPT to cite you

SEO for AI, or GEO (Generative Engine Optimization), is the work of structuring your site so ChatGPT, Perplexity, Google AI Overviews, and Claude cite you when someone asks them a question. It isn't a new channel: it's SEO adapted to a search engine that, instead of handing you ten links, writes an answer and names its sources.

The practical difference is who reads your page. You used to optimize so a person would click; now, on top of that, so a language model pulls a fragment and cites it. And models don't read whole pages: they pull passages. That single idea changes how you write.

What exactly is SEO for AI?

It's the practice of preparing your content, your structure, and your presence on the web so generative engines can retrieve, understand, and reuse your information when they answer a question. The technical term is GEO (Generative Engine Optimization), and it lives alongside classic SEO: most of the work serves both. The origin of the term, the paper that measured it, and what Google recommends are in our complete guide to Generative Engine Optimization.

How is it different from SEO as we knew it?

It isn't a replacement, it's a layer on top. What changes is the unit of work and the measure of success:

That's why Search Engine Land keeps stressing that good GEO is almost always good SEO: the fast, well-structured, authoritative site that ranks on Google is also the one AI prefers to cite. If you want the point-by-point breakdown, we cover it in GEO vs SEO: how they differ.

How does ChatGPT or Perplexity decide who to cite?

Generative engines start from billions of pages and pick fragments they can reuse without ambiguity. In practice, they favor three things: direct, self-contained answers (ones you understand without the rest of the page), verifiable data with a source, and credible mentions of your brand spread across the web. According to Semrush data from January 2026, the most-cited domains in ChatGPT, Perplexity, and Google AI Mode are Reddit and LinkedIn: presence beyond your own site counts too.

7 things you can do today to get cited by AI

1. Answer before you explain. Start each section with the answer in the first 300 characters; the detail comes after. The AI pulls that first block.
2. One H2 = one question. Title by what people ask, and put an answer underneath that stands on its own, without leaning on the previous paragraph.
3. Publish data no one else has. A study, a benchmark, an original figure. Original content and research are what attract citations; ten identical pages don't.
4. Signal freshness. Engines penalize stale content. Update your key pages and show a visible "last updated" date.
5. Use structured data (schema). Markup reduces ambiguity and speeds up extraction; for GEO it matters even more than for classic SEO.
6. Mind your speed. If your site is slow or unstable, the engine can skip it and cite a faster one. Speed is a cost of entry, not an extra.
7. Exist beyond your own site. Reddit, LinkedIn, YouTube, and industry media feed AI answers. A brand visible only on its own domain starts at a disadvantage.

An example: the same paragraph, before and after

The difference between text the AI ignores and text it cites is clearest with a case. Same information, two ways of writing it:

The second block answers in the first sentence, stands on its own, and defines the term with its acronym. It's exactly the kind of passage a model can clip and attribute without ambiguity; the first forces it to read three lines before finding anything citable, and by then it has already picked another source.

Should I let AI bots crawl my site?

Yes. For ChatGPT, Perplexity, or Claude to cite you, their crawlers have to be able to read your site. The main ones are GPTBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, and CCBot (Common Crawl, the open archive that trains a good share of the models). If your server or your robots.txt blocks them, you're out of the generated answer no matter how good your content is.

The block isn't always intentional. Sometimes a security plugin, the hosting firewall, or an inherited rule introduces it, and it doesn't show up in robots.txt but at the server level, where no one sees it. The check is simple: request your own page while imitating each bot's identifier and confirm it returns 200 and the real HTML, not a block page. It's the first thing we review in a GEO audit, because it's the most expensive failure and the most invisible.

Mistakes that make AI ignore you

• Blocking AI crawlers without knowing it. The most common failure and the hardest to spot: check GPTBot, ClaudeBot, and CCBot access before anything else.
• Burying the answer. Three paragraphs of context before you answer the question. The model doesn't wait: it pulls from the top.
• Content that adds nothing new. Rewriting what ten other pages already say gives no reason to cite you.
• A slow or unstable site. If it lags or fails to load, the engine cites a more reliable source.
• Walls of text with no structure. Without clear headings or marked-up data, the extractable passage doesn't exist.

How do I know if AI is already citing me?

The metric that replaces position is citation frequency: how often your brand appears in the answers generated for questions in your sector. You measure it by systematically querying the different engines and logging when they name you and with which link. It's exactly the diagnosis we start any GEO project with: first find out whether the AI sees you, then work so it cites you.

So is classic SEO no longer worth it?

Quite the opposite: it's the foundation. There's no GEO without healthy technical SEO — crawling, indexing, speed, architecture — because those are the same signals a model uses to decide whether your page is trustworthy. SEO for AI doesn't replace what you were already doing; it adds a new demand for clarity and structure, built so that, on top of ranking, you get cited.

Frequently asked questions