Similar tips into duplicate when the main, value-bearing content overlaps so heavily that two pages serve no separate purpose. The line is about the substance, not the surface, and it is a selection problem rather than a penalty. Pinning it to a percentage of overlapping words gets it wrong, because there is no such number; Google has never published a threshold, and the tools that report a similarity figure are using their own setting, not a rule from Google. What decides it is purpose, not a count.
That reframing changes what is actually at stake. When Google finds pages it considers duplicates, it does not punish them; it clusters them, picks the one it judges best to represent the group, and consolidates the signals onto that representative. The others are filtered, not penalized. So the fear of a duplicate-content penalty points at the wrong risk. The real cost is loss of control over which version gets shown and a dilution of signals across pages that should have been one.
The practical test is whether the core content of the two pages genuinely differs and each serves a distinct intent. Product variants that share a description but answer different needs can coexist with a canonical. Two articles that cover the same ground in slightly different words do not, because once you set the wording aside, they are the same page twice. Near-identical location pages that change only a city name fall on the duplicate side for the same reason: the value-bearing content is the same.
So the editor judges two pages by whether their core content differs and serves separate purposes, not by how much the surface wording happens to match, and treats the question as one of consolidation rather than of avoiding a penalty.