How to reduce similarity in a research paper — the right way
A high similarity score is not an accusation — it is a list of places where your text matches something else. Some of those matches are fine, some need a citation, and a few need rewriting. The goal is not a lower number at any cost; it is a paper where every borrowed word and idea is credited.
Read the matches, not the percentage
Before changing anything, sort the matches in the report:
- Quotations and references — correctly quoted and cited text, and your reference list, are expected matches.
- Standard phrasing — method descriptions, definitions and common phrases match across many papers.
- Your own earlier work — reused text from your thesis or a previous paper.
- Uncited overlap — sentences or ideas from a source with no citation. This is what must be fixed.
There is no universal "acceptable" percentage. Editors look at what matched and whether it is credited.
Matches that look bad but usually aren't
- Your own preprint. If the paper is on arXiv, SSRN or a university repository, the checker finds it and reports a large match. Tell the editor it is your preprint — most journals accept preprinted work.
- The reference list. Titles and author names in your bibliography match the papers they point to; that is the point of a reference.
- Correctly quoted text. A quotation in quotation marks with a citation is not plagiarism, even though it matches word for word.
- Standard phrases and names — the name of a method, a dataset, an instrument or a statistical test appears the same way in every paper.
- Many tiny matches. Dozens of 1% matches of common phrasing are background noise; one long match from a single source deserves a look.
Fix uncited overlap
- Add the citation if the idea came from a source.
- Quote if the exact wording matters — with quotation marks and a citation.
- Paraphrase properly: read the source, close it, explain the idea in your own words and structure, then cite it. Changing a few words is not paraphrasing.
- Summarize long borrowed passages into a sentence or two, cited.
The paraphrasing tool helps you find a different way to phrase a sentence you have already understood — keep the citation.
A worked example
Suppose the similarity report flags this sentence in your introduction, copied from a 2019 paper without a citation:
Deep learning models require large labelled datasets, which are expensive to produce in medical imaging.Option 1 — quote it if the wording matters: put it in quotation marks and cite the 2019 paper.
Option 2 — paraphrase and cite: explain the point in your own structure, then cite the source:
Because annotating medical images takes expert time, the labelled datasets that deep learning depends on are costly to build [12].Option 3 — make it yours: if the point is common knowledge in your field, state it briefly and cite a review, then move on to what your paper adds.
What does not work: swapping a few words for synonyms and leaving the citation out. The idea is still borrowed, and a paraphrase-aware checker will still match it.
Handle overlap with your own work
Reusing your own published text without saying so is self-plagiarism. Cite the earlier paper, rewrite the reused parts, and disclose substantial overlap to the editor — especially for a thesis turned into articles.
Method sections and standard phrasing
Methods are often described in similar words. Cite the protocol or paper you followed and describe what you did differently; editors understand method overlap when it is credited.
Re-check before you submit
Run the revised manuscript through the plagiarism checker and read the remaining matches again. If your venue has an AI-use policy, check that too — the AI content detector shows which sentences read machine-written.