The percentage answers a question nobody asked
The similarity index is the proportion of the document that matched something, anywhere, in the comparison set. It aggregates every match of every kind — a cited quotation, a reference list entry, a standard sentence about ethical approval, a phrase that two papers happened to share.
Two manuscripts can return the same figure and be in entirely different situations. A paper at a given percentage made up of dozens of short, scattered, properly cited matches is usually fine. A paper at the same percentage where most of it comes from one source, in continuous prose, is a problem regardless of what the number says.
This is why no responsible source publishes an acceptable threshold, and why you should be wary of anyone who does. Journals set their own practice, editors apply judgement, and the figure alone determines nothing.
One 8% match from a single paper matters more than 25% spread across forty correctly cited sources.
Read the filters before you read the score
Reports can be generated with or without exclusions, and the same manuscript returns very different figures depending on the settings. The common exclusions are quoted material, the bibliography, and matches below a minimum word count.
A raw, unfiltered score on a well-referenced paper will look alarming and mean little — the reference list alone can account for a large share of it, since every citation you share with another paper is a genuine string match.
So establish what you are looking at first. If the report excludes quotes and the bibliography, the remaining figure is closer to something meaningful. If it does not, the number tells you almost nothing until you have opened the matches.
Which matches actually need action
Work through the source list rather than the summary. Continuous prose from a single source, particularly in the introduction or discussion, is what draws editorial attention: it looks like a passage that was pasted and lightly altered, and it usually is.
Matches to your own earlier work need handling rather than hiding. Reusing your own text — text recycling — is acceptable in some sections and some fields and not in others, and the correct response is to cite the earlier paper and, where the overlap is substantial, tell the editor in the cover letter.
Methods sections match heavily and that is normal. There are a limited number of ways to describe a standard assay or a validated instrument, and rewriting an accurate description into an awkward paraphrase to move a number is the wrong trade. Where a method genuinely comes from another paper, cite it.
Reducing similarity honestly
The legitimate fixes are all about attribution and expression. Quote and cite properly where you are using someone’s words; paraphrase genuinely, in your own structure, where you are using their idea; and cite the source either way.
What does not work — and is straightforwardly detectable — is synonym-swapping, inserting characters, or running text through a paraphrasing tool. These leave the sentence structure intact, they degrade the writing, and editors reading a flagged passage recognise the pattern immediately.
The underlying test is simple and worth applying yourself: would this paragraph be acceptable if the report did not exist? If yes, leave it. If no, the report has done its job and the paragraph needs rewriting — not the number lowering.