Repeated passages and ambiguous answer ownership

Search engines collapse duplicates: Google chooses one representative (canonical) URL from a set of duplicate pages and shows only that version, and its choice can differ from the site's preference. Fröbe et al. (ECIR 2020) note that web search engines typically keep only the 'best' of a set of near-duplicates by some selection criteria, and show that not discounting near-duplicates inflates evaluation scores, costing systems that filter them between 8 and 53 ranks in the worst case across TREC tracks. That near-duplicate passages across a site also split citations in answer engines, or that the non-canonical copy gets picked, is an inference from these mechanics, not a measured result.