Density is the wrong unit, and no percentage marks the line where keyword usage becomes stuffing. Google doesn’t count how often a keyword appears against your total word count and flip a switch at some number; its systems read for meaning, not ratios, and Google has made clear that keyword density is not a ranking factor and there is no ideal figure to hit. So any “stuffing begins at X percent” answer invents a threshold that doesn’t exist, and chasing a percentage is chasing a metric the search engine isn’t using.
What stuffing actually is, is a reading experience, not a ratio. Google defines it as filling a page with keywords to manipulate ranking, and the tell is that the text reads as unnatural, repetitive, or forced to a person. A keyword can appear many times and be fine if every use is natural, and it can appear only a handful of times and already feel stuffed if the repetition is clumsy. The line gets crossed when a reader notices the keyword being worked in, rather than reading a sentence that happens to contain it. There are diminishing returns to repeating a term anyway: the first mention or two helps Google understand the page, and once the repetition grows obvious, more of it does nothing but risk sounding like spam.
So the test that replaces the density counter is your ear. Read the passage aloud. If the keyword starts to clang, if the phrasing turns stiff or you hear the same term landing over and over, that is the stuffing, and you fix it by rewriting for natural flow, not by tuning a number toward some target percentage.