Same first brand, different source lists

ChatGPT and Perplexity named the same brand first when we asked, “what's the best facial cleanser for sensitive skin.” Both started with Vanicream in every saved answer: 32 from ChatGPT and 28 from Perplexity.

Their attached source lists told a different story. ChatGPT’s saved citations covered five hosts, including aad.org, cerave.com and vanicream.com. Perplexity’s covered eight, including nytimes.com, ulta.com and fortune.com. None of the cited hosts overlapped between those two sets.

That does not mean the engines drew on entirely separate knowledge. It means the source links preserved alongside their answers differed, even when the first named product brand was the same. The export does not retain the original inline citation anchors, so it cannot tell us which link supports each sentence.

The accounting question offered a similar example. Asked “best accounting software for small business,” both engines named QuickBooks first in all 12 of their saved answers. Their citation-host sets shared only techradar.com.

These are findings about two exact questions in one batch. Each Perplexity run reused the same saved ten-URL source set across its answers, so those repeated links should not be treated as independent discoveries or proof of source influence. A cited host is also not necessarily a publisher: platform and shopping links require separate treatment.

For marketers, this gives source analysis a specific job. It can show which links accompany sampled answers and where the engines overlap. It cannot, on its own, explain why a brand was recommended or prove that getting onto one of those sites would change the answer.