Paraphrasing is the skill of restating another author’s idea, finding, or argument entirely in your own words and your own sentence structure, while crediting the source. It sounds simple and is routinely done badly: the most common failure is not omitting the citation but rewording too lightly — swapping synonyms into the source’s original sentence shape. That practice, called patchwriting, is still a form of plagiarism even when a citation is attached. This guide works through what separates an acceptable paraphrase from patchwriting, with worked before/after examples across three registers of academic writing, and covers the related questions of self-plagiarism, AI paraphrasing tools, and what similarity-checking software can and can’t tell you.
Paraphrase vs. Summary vs. Quotation
Academic writing draws on other people’s work in three distinct ways, and conflating them is a common source of both weak prose and integrity problems:
- Quotation reproduces the source’s exact wording, set off in quotation marks (or block-quoted) and cited. Use it when the original phrasing itself is the point — a precise legal or regulatory definition, a striking or contested turn of phrase you need to analyze, or a technical term of art that has no equivalent substitute (see below).
- Paraphrase restates a specific passage — typically a sentence or a few sentences — in your own words and structure, at roughly the same level of detail as the original, and cites the source. Paraphrase is the default choice for reporting someone else’s finding, method, or argument in the flow of your own writing.
- Summary condenses a larger unit — a section, a whole paper, a body of literature — into a much shorter statement of its overall point, again in your own words and cited. Summary trades detail for compression; paraphrase does not.
All three require attribution. The only difference between them is how much of the source’s original wording and detail survives — quotation keeps all of it (marked as such), paraphrase keeps none of the wording but similar detail, and summary keeps neither the wording nor the full detail.
Why Over-Quoting Weakens Scientific Writing
Reviewers and readers of scientific and scholarly writing generally treat a manuscript that is heavily built from strung-together quotations as a weakness, not a strength, for reasons distinct from plagiarism itself:
- It signals the author has not fully synthesized the material — paraphrasing forces you to demonstrate understanding, because you cannot reword what you have not understood.
- It disrupts your own voice and argumentative thread; a literature review or discussion section built from quotations reads as a scrapbook rather than an analysis.
- Quotation marks are appropriate for wording that matters as wording. Most research findings, methods, and background claims do not — the finding matters, not the sentence reporting it — so quotation is usually the wrong tool for reporting them.
The practical guideline used across most academic style guides and writing centers: quote sparingly and only when the original wording is doing real work; paraphrase or summarize everything else.
Patchwriting: Why Changing a Few Words Isn’t Paraphrasing
Patchwriting — sometimes called mosaic or patchwork plagiarism — is what happens when a writer starts from the source sentence and edits it in place: swapping in synonyms, reordering a clause or two, but keeping the source’s underlying sentence structure and logical sequence intact. The result reads as “new” text, and a citation is often attached to it in good faith, but the sentence is still substantially the source author’s construction, not the writer’s own.
This matters because most institutional and standards-body definitions of plagiarism turn on appropriation of the source’s expression, not merely its exact wording. The US Office of Research Integrity (ORI) — the federal body that oversees research-misconduct findings in PHS-funded research under 42 CFR Part 93 — divides plagiarism into two categories: plagiarism of text (copying a portion of text from another source without giving appropriate credit, which explicitly includes verbatim copying, mosaic plagiarism, and patchwriting as variants of the same underlying problem) and plagiarism of ideas (thoroughly paraphrased material presented without credit at all). Patchwriting sits in the first category, alongside verbatim copying, because the wording is still substantially borrowed even though no single unbroken string matches. See CASRAI’s Textual Plagiarism entry and the fuller types of plagiarism guide for how patchwriting relates to the other recognized categories.
Two consequences follow directly from that classification. First, adding a citation to a patchwritten sentence does not fix it — attribution and adequate rewording are separate requirements, and both must be satisfied (see below). Second, patchwriting is often harder for similarity-checking software to flag cleanly than verbatim copying, because the word-level changes break up long exact-match strings into shorter fragments — which is precisely why a passing similarity score is not proof that a passage was paraphrased properly (see the similarity-software section below).
The Technique That Actually Works
Patchwriting happens almost automatically when a writer paraphrases with the source sentence open in front of them and edits it live. The reliable alternative used across academic writing-center guidance is a four-step sequence that deliberately breaks that habit:
- Read the passage fully, more than once if needed, until you understand the claim it is making — not just the words it uses to make it.
- Put the source out of view. Close the tab, turn the book face-down, minimize the PDF. This step is the one patchwriting skips, and it is the one that matters most.
- Write the idea from your own understanding, as if explaining it to someone else, without looking back at the original wording. Because you are constructing the sentence from your own mental model of the claim rather than editing existing text, the resulting structure is naturally your own.
- Check back against the source — not to compare wording, but to verify you have represented the claim accurately, preserved any qualifiers or limitations the original author included (e.g., “in this sample” or “under these conditions”), and haven’t introduced a distortion. Then add the citation.
The order matters: writing happens with the source closed, and verification happens afterward. Doing it in the reverse order — writing with the source open — is what produces patchwriting even when the writer has no intention to plagiarize.
Three Worked Examples
Each example below is an illustrative, invented passage written for this guide — not a real publication or a real dataset — to avoid reproducing anyone’s actual copyrighted text. Each shows the same structure: the original, an unacceptable patchwritten version, and an acceptable paraphrase, with commentary on exactly what changed.
Example 1 — Quantitative / Methods Register
Original (illustrative source passage): “Participants who reported fewer than six hours of sleep per night showed a statistically significant increase in reaction time on the working-memory task, relative to participants who reported seven to nine hours of sleep.”
Unacceptable (patchwriting): “Subjects who reported less than six hours of sleep nightly displayed a statistically significant rise in reaction time on the working-memory task, compared to subjects who reported seven to nine hours of sleep.”
Acceptable paraphrase: “Sleep duration under six hours a night was linked to slower working-memory reaction times; participants who slept seven to nine hours performed faster on the same task (Author, Year).”
What changed and why it matters: The unacceptable version swaps individual words — “participants”/”subjects,” “showed”/”displayed,” “increase”/”rise” — but keeps the exact clause order and sentence architecture of the original. It is a synonym exercise, not a paraphrase, and a citation attached to it would still be patchwriting. The acceptable version restructures around the relationship itself (shorter sleep, slower reaction time) rather than around the original’s subject-then-comparison sentence shape, and compresses the two comparison groups into a single clause instead of mirroring the original’s two-clause structure.
Example 2 — Biomedical / Clinical Register
Original (illustrative source passage): “The intervention group demonstrated a 12% reduction in 30-day readmission rates compared with the control group, an effect the authors attributed to structured discharge counselling rather than to differences in baseline comorbidity.”
Unacceptable (patchwriting): “The treatment group showed a 12% decrease in 30-day readmission rates versus the control group, an effect the researchers ascribed to structured discharge counselling rather than to differences in baseline comorbidity.”
Acceptable paraphrase: “Structured discharge counselling, rather than baseline differences in comorbidity, appears to explain the 12-percentage-point drop in 30-day readmissions seen in the intervention arm relative to controls (Author, Year).”
What changed and why it matters: Again, the unacceptable version is a near word-for-word substitution (“intervention”/”treatment,” “demonstrated”/”showed,” “attributed”/”ascribed”) on an unchanged frame. The acceptable version leads with the explanation (discharge counselling) rather than the result, which both avoids mirroring the source’s sentence order and, as a side effect, reads more naturally as analysis rather than restatement.
Example 3 — Qualitative / Interview-Based Register
Original (illustrative source passage): “Interview participants frequently described their transition to remote work as isolating, even when they reported high satisfaction with their formal productivity metrics, suggesting a disconnect between perceived output and subjective wellbeing.”
Unacceptable (patchwriting): “Interviewees often characterized their shift to remote work as isolating, even when they reported strong satisfaction with their official productivity measures, implying a gap between perceived output and subjective wellbeing.”
Acceptable paraphrase: “Even interviewees satisfied with their own measured output frequently felt isolated after their move to remote work — a recurring split between productivity metrics and subjective wellbeing (Author, Year).”
What changed and why it matters: The patchwritten version substitutes near-synonyms (“interview participants”/”interviewees,” “described”/”characterized,” “suggesting”/”implying”) while preserving the original’s exact concessive structure (“even when… suggesting…”). The acceptable version keeps the concessive relationship — because that logical relationship is the actual finding being reported, and inventing a different one would misrepresent the source — but restates it with different sentence mechanics and a different lead.
Citation Is Still Required — Both Conditions Must Hold
Two separate conditions determine whether a paraphrase is acceptable, and both must be true at once:
- Adequate rewording. The passage must be genuinely restated in the writer’s own words and structure, not patchwritten (see above).
- Attribution. The source of the idea must be cited, even though no quotation marks are needed for a properly reworded paraphrase.
A passage that satisfies only one of these is still plagiarism. A perfectly reworded paraphrase with no citation is plagiarism — specifically ORI’s “plagiarism of ideas” category, since the wording no longer matches anything a similarity checker could flag, but the idea is still uncredited. A patchwritten passage with a citation attached is also plagiarism — the citation signals engagement with a source but does not license reproducing its expression essentially unchanged. See CASRAI’s guide to quoting and citing sources correctly and citation styles and referencing formats for the mechanics of attaching a correct, complete citation once the paraphrase itself is sound.
When the Original Phrasing Can’t Be Improved: Technical Terms and Terms of Art
Not every phrase in a source passage should be reworded. Established technical terminology, defined legal or regulatory language, standardized instrument or scale names, and precise statistical or methodological terms of art carry a specific, agreed meaning that a synonym would blur or change outright — “statistically significant,” “double-blind,” “intention-to-treat,” a named questionnaire, or a defined regulatory term like “adverse event” are not interchangeable with looser phrasing. Forcing a rewording of language like this for its own sake risks introducing inaccuracy, which is a worse outcome than repeating an unavoidable technical phrase.
The correct handling in this situation is not to patchwrite around the term but to quote the specific unavoidable phrase — even a two- or three-word quotation, set off in quotation marks and cited — while paraphrasing the surrounding sentence normally. This is a legitimate, common pattern in scientific writing: a paraphrased sentence with one small quoted technical phrase embedded in it, rather than either quoting the whole passage or forcing an inaccurate rewording of precise terminology. See CASRAI’s guide to evaluating source credibility and primary vs. secondary sources for related judgment calls about how closely to track an original source’s own language.
Paraphrasing Your Own Prior Work
The same patchwriting problem applies to an author’s own earlier writing, with an added wrinkle: reusing your own previously published words, findings, or structure without disclosure is self-plagiarism (also called text recycling), a publication-ethics concern even though it does not involve appropriating anyone else’s work. Two things distinguish the self-plagiarism case from the standard paraphrasing case above:
- Under the US federal research-misconduct regulation (42 CFR Part 93), self-plagiarism does not meet the definition of research misconduct, because that definition is scoped to “another person’s” ideas or words. It is nonetheless routinely treated as a publication-ethics violation by journals and institutions, because undisclosed reuse misrepresents a paper’s novelty to editors, reviewers, and readers.
- Methods sections are the recognized exception. Many journals and editorial-ethics bodies accept — and some explicitly permit — an author reusing substantially similar methods-section language across closely related papers from the same study or research programme, on the reasoning that a precise, previously validated description of a procedure has limited room for novel wording without sacrificing accuracy, and that methods text is not where a paper’s originality claim rests. The safer practice where reuse across papers is more than trivial is still to disclose it (e.g., “as previously described in [citation]”) rather than assume it is automatically acceptable.
See CASRAI’s Self-plagiarism entry and types of plagiarism guide for the fuller distinction between the federal misconduct definition and journal-level publication-ethics policy on text recycling.
AI Paraphrasing Tools Don’t Solve the Attribution Problem
“Paraphrasing” or “spinning” tools that mechanically substitute synonyms and lightly reorder a pasted sentence produce, by construction, exactly the patchwriting pattern described above — the underlying sentence structure and logical sequence of the source is preserved because that is how the tool operates, only the individual words change. Running a passage through such a tool does not turn it into an acceptable paraphrase, and it does nothing at all to satisfy the separate attribution requirement: a citation still has to be added by the writer, and the tool has no way to know what source the passage came from or verify the resulting sentence still accurately represents the original claim.
General-purpose AI writing assistants used to paraphrase source material raise the same two issues plus a third: accuracy drift, where a model’s rewording subtly changes a claim’s meaning, qualifiers, or scope in ways that are easy to miss without checking back against the source. A growing number of journals and institutions now require authors to disclose AI-assisted writing tool use in a methods or acknowledgments statement, separate from and in addition to the underlying citation requirement. See CASRAI’s dedicated guide on AI paraphrasing tools and plagiarism risk for a fuller treatment of the disclosure landscape and detection risk, and whether AI-generated text counts as plagiarism for the broader question beyond paraphrasing specifically.
What a Similarity Score Does and Doesn’t Tell You
Similarity-checking software (Turnitin, iThenticate, Crossref Similarity Check, and comparable tools) works by string-matching a submitted document against a large reference corpus and reporting what percentage of the text matches existing sources closely enough to flag. That is a narrower thing than “is this plagiarized,” in both directions:
- A high similarity score is not automatically plagiarism. Properly quoted material, a reference list, standard methodological boilerplate, and common technical phrasing all legitimately match other documents and inflate the raw percentage without being improper.
- A low similarity score does not certify proper paraphrasing. This is the more consequential gap for this guide’s topic: because similarity checkers work by matching strings of text, and patchwriting deliberately breaks up long exact-match strings through word-level substitution, a patchwritten passage frequently returns a low or unflagged similarity score even though it fails the adequate-rewording test above. A passing score reflects the absence of a long verbatim match — it does not reflect whether an idea was properly credited or a sentence was genuinely restructured.
Vendor guidance is explicit on this point: Turnitin’s own documentation states that its similarity tool does not itself determine whether plagiarism occurred — it supports a human reviewer’s judgment. No standards body (ORI, COPE) or major vendor publishes a universal numeric threshold above which a score constitutes misconduct; any specific cutoff in use is a local institutional or journal policy. See CASRAI’s anti-plagiarism software guide and the Acceptable Similarity Percentage and Originality Report dictionary entries for more on how these tools and their thresholds actually work.
Frequently Asked Questions
What is the difference between paraphrasing and patchwriting?
A genuine paraphrase restates a source’s idea in the writer’s own words and sentence structure. Patchwriting starts from the source sentence and edits it in place — swapping synonyms, minor reordering — while keeping its underlying structure intact. Patchwriting is still classified as plagiarism (specifically ORI’s “plagiarism of text” category) even when a citation is attached, because the expression is still substantially borrowed.
Do I still need to cite a paraphrase, even if I’ve reworded it completely?
Yes. Adequate rewording and attribution are two separate requirements, and both must be satisfied. A thoroughly reworded passage with no citation is still plagiarism — ORI’s “plagiarism of ideas” category — even though no similarity checker would flag it.
Is using an AI paraphrasing tool considered plagiarism?
Running a passage through a synonym-substitution or AI paraphrasing tool does not, by itself, satisfy either requirement above: the tool’s output typically preserves the source’s sentence structure (patchwriting), and it cannot add the citation for you. See CASRAI’s guide to AI paraphrasing tools and plagiarism risk for disclosure norms.
Can I paraphrase my own previously published work?
Reusing your own prior wording or findings without disclosure is self-plagiarism (text recycling) — a publication-ethics issue at most journals, though not research misconduct under the US federal definition, which is scoped to appropriating another person’s work. Methods-section language is the commonly accepted exception, since precise procedural descriptions have limited room for novel wording.
Does a low similarity score mean my paraphrasing was done correctly?
No. Similarity software detects string matches; patchwriting is specifically designed, even unintentionally, to avoid long string matches through word-level substitution, so a patchwritten passage often scores low or unflagged despite failing the adequate-rewording test. A low score reflects an absence of verbatim copying, not confirmation of a properly reworded, properly cited paraphrase.
Should I ever quote instead of paraphrase in a research paper?
Yes, when the original wording itself matters — a precise legal, regulatory, or technical term of art, a defined instrument name, or language you are directly analyzing or contesting. Quote sparingly and only where a paraphrase would blur necessary precision; paraphrase or summarize everything else.
Related CASRAI Resources
- Types of Plagiarism: The Full Taxonomy — where patchwriting and paraphrasing plagiarism sit within the wider taxonomy
- Quoting and Citing Sources Correctly — the mechanics of attribution once a paraphrase is sound
- Citation Styles and Referencing Formats
- Primary vs. Secondary Sources in Research
- CRAAP Test: Evaluating Source Credibility
- AI Paraphrasing Tools and Plagiarism Risk
- Anti-Plagiarism Software
- Self-plagiarism
- Textual Plagiarism
- Scholarly Writing & Publishing Craft — the full cluster hub
