Which sources does ChatGPT cite when it mentions a brand?

What a citation is, in an AI answer

When an answer engine searches the web before replying, it reads a handful of pages and writes from them. The URLs it shows beside the answer are those pages. They are not a bibliography of everything it knows; they are the specific pages it trusted for this question, today.

Not every engine shows them, and not every answer has them. Perplexity always searches and always cites. ChatGPT, Gemini and Claude cite when web search is on and say nothing about sources when they answer from memory, because there are none. Prism asks all four with search on wherever the engine supports it, and records which answers searched.

How to see them

1. 01

   Ask with search on, from a clean context

   The free brand check does this across four engines at once. Under each answer sits the list of URLs that engine cited, with the domain and the page.
2. 02

   Sort them into three piles

   Your own pages. Your competitors’ pages. Everything else: editorial lists, review sites, forums, documentation. Prism’s Citations report does this split for every URL across every run, and charts how the three piles move.
3. 03

   Find the prompts where you lost the citation

   A question where an engine cited a competitor and not you is a citation gap. Prism lists every one, with the page that won, and exports the list. It is an outreach list and a content brief in one file.
4. 04

   Look at what the winning pages have

   Prism’s Page Intel crawls the pages that keep getting cited and scores them, so you can see what they carry that yours does not: a comparison table, a price, a date, a named author.
5. 05

   Read the fan-out

   Before answering, an engine often runs its own follow-up searches. Prism records them. They tell you which question the engine actually answered, which is frequently not the one you thought you asked.

What the list tells you

Read the cited domains for a question and a pattern shows quickly. Some questions are answered from two or three editorial sites the engines return to again and again. Some are answered from the brands’ own pages. Some are answered from a forum thread that happened to rank. Each pattern is a different job.

Editorial sites win

The engines trust a third party for this question. Being named by that third party is the route in, and the tool has just given you the site and the page.

Brand pages win

The engines are reading vendors directly. The competitor’s cited page is a spec for the page you need: what it states, how plainly, what it structures.

Forums and threads win

Nobody authoritative has answered this question well. The first clear page usually takes it.

Your pages win, and you are not named

The engine read you and still did not say your name. Usually the page does not say it either, plainly, near the top.

Why the sources beat the mention count

A share of voice figure tells you that you are behind. It does not tell you what to do on Monday. The cited pages do: they are the finite list of places an engine looks when it answers this question, and the work is to be on that list, either by being the page or by being named on it. Everything else in AI visibility is a way of measuring whether that work is landing.

The same list joins to your search data. Prism’s AI to SEO Bridge matches your prompts to the Search Console queries you already rank for, by meaning rather than wording, and shows the queries you win in Google and never get cited on. Those are usually the cheapest wins on the list, because the page exists.