How Perplexity ranks sources in real time

The short answer

Reverse-engineering Perplexity's retrieval and ranking pipeline. The pattern repeats across ChatGPT, Perplexity and Gemini: the model isn't choosing the highest-ranked page — it's choosing the page that makes its answer easiest to write.

What the data shows

For two decades, search ranking was a list. Generative search broke that format. There is no list anymore — there is a paragraph, written by a model, that mentions some brands and ignores the rest.

The model isn't choosing the best page. It's choosing the page that makes its answer easier to write.

The four signals that actually move the needle

• Entity clarity — the page resolves to a single, named subject.
• Definitional language — sentences shaped like answers, not ads.
• Comparative scaffolding — explicit "vs.", "best for", "alternative to" structures.
• Recency proof — visible last-updated dates and current data points.

How to test this in your own stack

Pick five questions a high-intent buyer would ask about your category. Run each one through ChatGPT, Perplexity and Gemini. Record which brands are cited, in what order, and from which page. Repeat in seven days. The delta is your visibility surface.

Frequently asked questions