How ChatGPT and Perplexity pick who to quote.

When someone asks ChatGPT or Perplexity a question your business could answer, the model isn't picking a source at random or favoring whoever has the biggest brand. It's running through a short list of concrete, checkable factors — and most sites fail at least one of them without ever knowing it.

1. Can the model actually reach the page?

Every major AI system uses its own named crawler — GPTBot for OpenAI, ClaudeBot for Anthropic, PerplexityBot for Perplexity, Google-Extended for Gemini and AI Overviews. Each of these can be individually allowed or blocked in robots.txt, separately from whether Googlebot or Bingbot are allowed. It's common for a site to have generic SEO crawlers fully open while silently blocking every AI-specific one — often from an old robots.txt template that predates any of these crawlers existing. A blocked crawler means zero citations from that engine, permanently, regardless of how good the content is.

2. Can the model extract a clean answer?

Even an unblocked page can be functionally invisible if the actual answer is hard to lift cleanly. Models favor pages where the direct answer appears early and plainly — a clear sentence stating the fact, not three paragraphs of scene-setting before the point. A page that opens with a story instead of an answer usually loses the citation to a competitor's page that states the same fact in its first sentence.

3. Does structured data back up the claim?

Schema.org markup — Organization, FAQPage, Article, LocalBusiness — gives a model a machine-readable version of a claim it would otherwise have to infer from prose. A page with FAQPage schema answering "how much does X cost" is a much safer, more confidently quotable source than a page where that same answer is implied somewhere in a paragraph. Structured data doesn't replace the prose; it's the model's confidence check on it.

What this means in practice

None of these three factors are about writing "better" content in the usual sense — a site can have excellent, accurate information and still lose every citation on all three counts. They're technical, binary, and checkable in minutes rather than something to intuit.