Why it matters
Before building content, a bearings distributor analyzes who ChatGPT cites for "SKF to NTN interchange." It finds two PDF spec mills and one forum thread doing the work. Those three domains are the incumbents. The distributor then ships a crawlable interchange table that fills the gaps the PDFs leave, and aims to either replace them or get cited alongside them.
Keyword rank tells you who Google lists. Cited-domain analysis tells you who the model actually quotes, which is the surface your buyer now reads. For an MRO catalog, that gap matters: a part-number answer rarely cites the manufacturer. It cites whoever structured the cross-reference cleanly.
How to do it
Run a stable set of buyer prompts across each engine, log the domains cited in every answer, and tally them. Sample repeatedly, since answers vary per run. Sort by frequency to see the incumbents. Then read the cited pages to find what makes them citable and where they are thin.
In practice
A hydraulics distributor pulls the cited domains for 25 cross-reference prompts like "Parker to Eaton fitting equivalent." One aggregator and a single OEM PDF show up in most answers. That tells the team to publish a structured equivalence table aimed at the prompts those two sources answer poorly, instead of guessing at topics.