How do I paraphrase a source without plagiarizing?

The close-the-source method works because it forces comprehension. If you can restate a passage without looking at it, you understood it, and understanding is what produces genuinely different phrasing. If you cannot, that is useful information: reread it, because you were about to paraphrase something you had not grasped.

Synonym substitution is the failure to avoid, and it is what most people do when they paraphrase with the original visible. Changing "significant increase" to "notable rise" while keeping the sentence structure produces text that is not the author's words and is still the author's sentence. Institutions call it patchwriting and treat it as an integrity offence.

There is a subtler point about attribution. A paraphrase still needs a citation, because the idea is the author's even when the wording is yours. The citation is not payment for borrowed words, it is a statement about where the knowledge came from.

The reviewed literature defines plagiarism as using another person's thoughts, words, phrases, sentences, facts, or intellectual work without authorization while presenting it as your own (Sozon et al., 2024, pp. 4-5). That definition covers the idea, not only the sentence, which is why the citation survives even when none of the original wording does. Where you place it matters as much as whether you include it. Penders argues that a source deserves credit for the exact contribution it makes rather than for a vague contribution to the topic, so attach the citation to the clause that depends on it, and cite the same source again wherever it supports a different part of your argument (Penders, 2018, p. 2).

The last thing to check is accuracy in both directions. A paraphrase can be too close, which is a plagiarism problem, and it can also drift, which is a truth problem. Reopening the source and comparing is one minute of work that catches both.

That check is the step writers skip and the step that fails most often. Jergas and Baethge's systematic review of quotation accuracy in medical journal articles pooled 27 studies for a major error rate of 11.9% and a minor error rate of 11.5%, and 28 studies for a total error rate of 25.4% (Jergas and Baethge, 2015, p. 11). Those figures describe quotations in published medical papers, not paraphrases in graduate work, so do not read them as a measure of your own risk. They do show how routinely a citation ends up attached to a passage that does not say what the citing author claimed.

How different does the wording have to be?

Different in structure, not just in vocabulary. Swapping words while keeping the sentence shape is patchwriting and is treated as plagiarism at most institutions. A genuine paraphrase usually reorders the ideas and changes the sentence boundaries, because you understood it rather than rewrote it.

Technical terms are the exception and should not be paraphrased. If the field calls it a random effects model, call it a random effects model. Inventing a synonym for a standard term is not originality, it is confusion.

A rough test: if you and the original share more than a few consecutive words outside technical terms, the paraphrase is too close and either needs rewriting or should become a quotation.

No validated threshold exists for how many words you have to change, and a similarity score is not one. Sozon and colleagues reviewed 45 articles published between 2018 and 2022 and traced plagiarism to pressure, weak awareness of academic integrity, outdated honour codes, and unethical use of AI tools, recommending ethical-development tutorials, updated honour codes that address AI, and teaching that strengthens critical thinking, research, and writing (Sozon et al., 2024, pp. 1-2). The same review reports that improper citation, honest mistakes, and culturally different understandings of academic integrity also produce plagiarism (Sozon et al., 2024, p. 5). So the question to answer is whether you represented the source accurately and credited its contribution, not whether a detection tool cleared you.

When should I quote instead of paraphrase?

When the exact words are the evidence, when a definition is contested and the author's phrasing matters, or when the original is more precise than any paraphrase could be. Quote sparingly in the sciences, more freely in the humanities, and always with a locator.

The strongest case for quoting is when you intend to disagree. Paraphrasing a position you are about to attack invites the reply that you misrepresented it, and quoting removes that objection entirely.

The weakest case is when a quotation is doing the work your own sentence should do. A paragraph built from three quotations stitched together has no argument in it, only assembly, and examiners read that as an absence of analysis.

One habit prevents most of this trouble. When you take a note from a source, mark whether it is a quotation, a paraphrase, or your own thought. Notes that lose that distinction are how an author's sentence ends up in a draft six months later with nobody remembering where it came from.