Someone Read ChatGPT's Network Traffic to See How It Picks Sources. The Findings Are Useful.

Someone read the network traffic instead of the answers, and it tells you more about getting cited in ChatGPT than most of the advice floating around. The study, from SEO researcher Suganthan, watched what ChatGPT actually fetches when it builds an answer, rather than guessing from the output.

The findings are practical. ChatGPT only triggers a live web search for current, specific or fast-changing questions, pulling and citing pages mostly through the Bing index. The opening question in a chat is 2.5 times more likely to trigger citations than a tenth follow-up, so first impressions matter. When it does search, it fires direct probes at vendor pages, including pricing pages. Of the pages it retrieves, only 15% end up cited in the final answer. The other 85% are found, weighed and thrown away.

There is a split worth understanding. Your own clean page gets your facts read. The recommendation gets cited from somewhere else.

The catch is that your facts have to be in crawlable text. Prices and specs loaded by JavaScript or baked into an image do not get read. The model reads the page and gives up when it cannot.

AI answers are becoming the first place people meet a brand, and the rules for showing up there are not the old SEO rules. You can be technically perfect and invisible if your facts hide behind JavaScript or you have no third-party validation. For Australian businesses this is a window. The behaviour is documented and most competitors are not acting on it yet.

Ranking was the old goal. Getting fetched, read and cited is the new one, and the mechanics are knowable. Build for them.