Estimate originality score from quoted text length.
A similarity score is not a plagiarism score. Checkers count every matching string, including correctly quoted material and the reference list, which is why a scrupulously cited essay can report 20 percent similarity. Excluding the bibliography and separating attributed quotation from unattributed matches produces the figure that actually matters: how much text overlaps a source without acknowledgement. Even a small unattributed share is a problem, which is why the risk band is deliberately strict.
Raw similarity
Raw = (quoted + bibliography + unattributed) / total words x 100
Body-text figures
Body similarity = (quoted + unattributed) / (total - bibliography) x 100; originality = 100 - body similarity
No threshold makes work acceptable. A 30 percent score built entirely from cited quotations and references is fine; a 3 percent score from one uncited paragraph is misconduct. Read the matches, not the number.
Only if it is genuine restatement and still cited. Substituting synonyms while keeping the structure is patch-writing and is treated as plagiarism by most institutions even though a checker may not flag it.
Because references are supposed to match — they are the same author names, titles and journals everyone else cites. Most checkers offer this exclusion for exactly that reason.