citation observatory
Who the engines draw on.
Counted over ChatGPT web app · en-US · wave luce-live-w30.
When an AI recommends a brand, it cites its sources. This is which publishers it drew on, per category, counted from the real answers Luce captured. In running shoes, runnersworld.com takes 16 of 88 citations, and 12 effective sources feed the whole shelf.
Running shoes ChatGPT web app · 30 answers captured · 29 cite sources · 88 citations · 12 effective sources
Bar width is the share of this section's most cited publisher, which is 16 citations. Each section is scaled to its own widest bar, so widths compare inside a section and never across two.
Wireless earbuds ChatGPT web app · 30 answers captured · 28 cite sources · 82 citations · 7.4 effective sources
Bar width is the share of this section's most cited publisher, which is 19 citations. Each section is scaled to its own widest bar, so widths compare inside a section and never across two.
Air fryers ChatGPT web app · 6 answers captured · 6 cite sources · 14 citations · 3.2 effective sources
Bar width is the share of this section's most cited publisher, which is 7 citations. Each section is scaled to its own widest bar, so widths compare inside a section and never across two.
Coffee machines ChatGPT web app · 6 answers captured · 6 cite sources · 19 citations · 11.6 effective sources
Bar width is the share of this section's most cited publisher, which is 3 citations. Each section is scaled to its own widest bar, so widths compare inside a section and never across two.
Robot vacuums ChatGPT web app · 6 answers captured · 6 cite sources · 11 citations · 3.5 effective sources
Bar width is the share of this section's most cited publisher, which is 5 citations. Each section is scaled to its own widest bar, so widths compare inside a section and never across two.
Counting rules: a publisher is counted once per citation at its registered domain, so every subdomain of one site lands on one row; the image hosts an answer cites alongside its pages are left out; each count reads a single surface at a single tier. Each section reports its effective sources, the diversity figure behind the counts.