How to reduce plagiarism in a research paper: lower your similarity score the honest way
A high Turnitin or iThenticate score does not always mean plagiarism, and a low one does not always mean a clean paper. This guide explains what a similarity score measures, what level is acceptable, how to read the report, and ten practical ways to reduce plagiarism in a thesis or journal article by making the writing genuinely your own.
Almost every university and journal now checks submissions with a similarity tool. Students meet Turnitin when they submit a thesis or dissertation; authors meet iThenticate, often through Crossref Similarity Check, when a journal screens a manuscript. When the report comes back at 30% or 40%, the natural reaction is to start changing words until the number drops. That is the slowest way to fix the problem, and it often makes the paper worse.
The better approach is to understand where the matches come from and fix the habits that caused them. Most high scores come from a handful of writing patterns that are easy to change once you can see them.
What a similarity score actually measures
A similarity checker compares your text with a large database of web pages, published articles and, for Turnitin, previously submitted student work. It highlights every passage that matches a source and reports the share of your document that matched. That percentage is a similarity index, not a plagiarism verdict.
The difference matters. A properly quoted and cited sentence still matches its source. So do your reference list, the names of scales and guidelines, and standard phrases such as "a cross-sectional survey was conducted". Meanwhile, a paper can score low and still plagiarise, for example by taking another author's ideas or structure without citation. Editors and examiners therefore read the matches, not just the number, and so should you.
What similarity score is acceptable?
There is no universal threshold. Each journal and institution sets its own policy, and most look at two things: the overall percentage, and whether a large share of the text matches a single source. A paper at 18% spread across forty sources, mostly references and method terms, is usually unremarkable. A paper at 12% where most of the matches come from one article is a much bigger concern.
In India, the UGC (Promotion of Academic Integrity and Prevention of Plagiarism in Higher Educational Institutions) Regulations, 2018 set out graded levels for research work such as theses and dissertations:
| Level | Similarity | What it means under the UGC regulations |
|---|---|---|
| Level 0 | Up to 10% | Minor similarity; no penalty |
| Level 1 | Above 10% to 40% | The work must be revised and resubmitted |
| Level 2 | Above 40% to 60% | Revision with a longer debarment from resubmission |
| Level 3 | Above 60% | The most serious level, with the strictest consequences |
The same regulations exclude some material from the calculation, including properly cited quotations, the reference list, common knowledge and generic terms. Always check the exact rules of your own university or target journal, because they take precedence over any general guide.
Read the report before you rewrite anything
Open the full report and sort every highlighted passage into one of the groups below. This takes twenty minutes and tells you exactly what needs work.
| Type of match | Example | What to do |
|---|---|---|
| Reference list and citations | Journal names, article titles, DOIs | Usually nothing; ask whether the bibliography can be excluded from the check |
| Names and standard terms | "Preferred Reporting Items for Systematic Reviews and Meta-Analyses", a scale's full name | Nothing: naming a work is not copying it |
| Quotations | A definition in quotation marks with a citation | Keep only if the exact wording matters; otherwise paraphrase |
| Stock method phrases | "Data were analysed using SPSS" | Rewrite to describe your own specific procedure |
| Close paraphrase | A sentence with a few words swapped from an abstract | Rewrite from understanding, keep the citation |
| Copied passage without citation | Two sentences lifted from another paper | Must fix: rewrite and cite, or quote and cite |
| Your own earlier work | Text from your published paper or master's thesis | Cite yourself and rewrite; follow the journal's policy on text recycling |
In most reports, the last three groups account for most of the real problem. Those are what the steps below address.
Ten ways to reduce plagiarism in a research paper
1. Take notes in your own words, then close the source
Most accidental plagiarism starts in the notes. If you copy sentences from a paper into your notes, those sentences tend to reappear in the draft weeks later, and you no longer remember they were not yours. Instead, read a section, close the PDF, and write down what the study did and found in your own words, with the citation beside it. When you draft from those notes, the wording is already yours.
2. Synthesise studies instead of summarising them one by one
A literature review written as "Author A found X. Author B found Y. Author C found Z." is close to a list of abstracts, and it usually reuses their wording. Synthesis groups studies by what they show together, which forces new sentences:
Summary (high similarity risk): Smith (2021) found that remote work increased employee autonomy. Lee (2022) found that remote work increased autonomy but reduced team communication.
Synthesis (your own analysis): Studies agree that remote work gives employees more control over how they work (Lee, 2022; Smith, 2021), but Lee (2022) suggests this gain may come at the cost of weaker communication within teams.
3. Paraphrase from understanding, not by swapping words
Replacing a few words with synonyms while keeping the original sentence structure is called patchwriting. Similarity checkers still match it, and editors recognise it. A real paraphrase changes the structure, the order of ideas and the vocabulary, and still carries a citation. Here is an invented example:
Original: "Regular physical activity substantially lowers the risk of developing type 2 diabetes in adults with obesity."
Patchwriting (still a match): Frequent physical activity considerably reduces the risk of developing type 2 diabetes in obese adults.
Paraphrase: Among adults with obesity, those who exercise regularly are much less likely to go on to develop type 2 diabetes (Author, Year).
4. Quote only when the exact wording matters
Quotations are legitimate when the wording itself is the point: a definition, a policy statement, an item from a questionnaire. Put the words in quotation marks, cite the source with a page number where your style requires it, and keep quotations short and rare. A paper built from quotations has a high similarity score even when every quote is cited.
5. Cite every idea, not just every quotation
Plagiarism is not only about words. A finding, an argument, a framework or even a distinctive structure taken from another author needs a citation, even when every word is your own. Sentences such as "Research shows that..." or "Studies have found..." with no citation are a warning sign for examiners. Either cite the studies or rewrite the sentence as your own argument.
6. Describe your own methods, not stock methods
Methods sections often score high because authors reuse familiar phrases. Write about what you actually did: which databases, which search terms, which inclusion criteria, how many records at each stage, which software version and which tests, in which order. Specific methods are more original and more useful to reviewers.
7. Handle your own earlier work carefully
Reusing text from your own published paper without citation is called self-plagiarism or text recycling, and many journals treat it as misconduct. If your thesis chapter becomes a journal article, check the journal's policy, cite the earlier work, and rewrite the overlapping passages, particularly in the introduction and discussion.
8. Keep names and technical terms as they are
Do not twist the name of a theory, scale, guideline or model to avoid a match. "The Preferred Reporting Items for Systematic Reviews and Meta-Analyses" should stay exactly that. Reviewers and examiners expect these matches and ignore them.
9. Never use paraphrasing tools to disguise text
Automated spinners replace words with odd synonyms and produce what research-integrity researchers call "tortured phrases", such as "counterfeit consciousness" for "artificial intelligence". Publishers now screen for these phrases, and they are one of the clearest signals of misconduct. The same applies to tricks such as swapping letters for look-alike characters, hiding text in white font, or pasting text as images: checkers and editors detect them, and they turn a writing problem into a misconduct case.
10. Check before you submit, and fix the causes
Run a similarity check on the final draft, sort the matches as described above, and fix the copied and closely paraphrased passages first. If you rewrite a passage, rewrite the whole sentence from your understanding of the source, not just the highlighted words.
What not to do when your similarity score is high
- Don't chase a number by swapping words. It rarely lowers the score much and makes the writing worse.
- Don't remove citations to hide sources. Uncited ideas are plagiarism even when the score is low.
- Don't use "humanizer" or spinner tools. They produce tortured phrases and are treated as deliberate concealment.
- Don't panic over references and method names. Those matches are expected; explain them if asked.
How Soba Researcher helps you write original papers
Soba Researcher, an AI Research Assistant, is built to produce drafts that are original from the start. Its writing rules use each source's abstract to learn what a study found, never as wording to reuse: it may not copy five or more words in a row from a source, it compares and contrasts studies instead of summarising them one by one, it describes the paper's own methods in specific terms, and it quotes only where exact wording matters. Every reference is a real published work found in OpenAlex.
Before the PDF is built, its Originality Checker compares every sentence with the abstracts of the paper's sources. Sentences that come too close are rewritten in new words with their citations kept, and claims about earlier research without a citation are either cited from the paper's sources or written as the paper's own point. The Originality tab on each paper shows every change. It compares against the paper's own sources, not the whole web, so you should still run your institution's or publisher's check before you submit. You can read more in the FAQ.
A checklist before you submit
- Have you read the full similarity report, not just the percentage?
- Does every idea, finding and framework from another author carry a citation?
- Are quotations short, rare, in quotation marks and cited?
- Is the literature review a synthesis of studies rather than a list of summaries?
- Does the methods section describe your own specific procedure?
- Is reused text from your own earlier work cited and rewritten?
- Are names of theories, scales and guidelines kept exactly as they are?
- Have you checked your institution's or journal's similarity policy?
Frequently asked questions
Does a high Turnitin score mean I plagiarised?
Not necessarily. The score counts every match, including cited quotations, references and standard terms. What matters is whether matched text is copied without credit. Read the report to see what each match is.
What is an acceptable similarity index for a journal article?
There is no single number. Policies vary by journal, and editors also check whether many matches come from one source. Check the journal's own guidance; in India, the UGC 2018 regulations treat up to 10% similarity in research work as Level 0, with no penalty.
Can I reduce plagiarism by using synonyms?
Swapping words while keeping the sentence structure is patchwriting, which checkers still match and editors recognise. Rewrite the idea in your own structure from your understanding, and keep the citation.
Is reusing my own published text plagiarism?
It can be. Reusing your own text without citation is self-plagiarism or text recycling. Cite your earlier work and follow the journal's policy on overlap.
Key takeaways
- A similarity score measures matching text, not plagiarism; read the matches before acting.
- There is no universal acceptable percentage; follow your institution's or journal's policy.
- Most high scores come from copied notes, summary-style reviews, patchwriting and stock methods.
- Synthesise, paraphrase from understanding, cite every idea, and quote sparingly.
- Never disguise text with spinners or tricks: that turns a writing problem into misconduct.
For more on preparing a submission, see how to respond to reviewer comments and how to spot fake references in AI-written papers.
This guide reflects common practice in research integrity and the UGC (Promotion of Academic Integrity and Prevention of Plagiarism in Higher Educational Institutions) Regulations, 2018. Similarity policies differ between institutions and journals; always follow the rules that apply to your submission.