Direct answer
Anti-plagiarism software is a text-matching system that compares a submitted document against indexed databases of web pages, journal articles, and previously submitted student papers, then reports the passages that match. It does not judge intent or declare misconduct — it produces a similarity report that a human reads and interprets [1]. Most university submissions in the US, UK, Canada, Australia, New Zealand, and Ireland pass through one of these systems before a grade is released, which is why understanding the report matters as much as understanding the writing.
How Does Anti-Plagiarism Software Actually Detect Copied Text?
The core mechanism is string matching at scale. When a file is uploaded, the system converts it to plain text, breaks it into overlapping phrases, and searches those phrases against its index of internet content, subscription journal databases, and its own archive of student papers [2]. Every phrase that appears in an indexed source is marked, and the marked spans are stitched together into a match list. There is no semantic "judgement" step in that pipeline — a correctly quoted sentence and a copy-pasted one look identical to the matcher.
What changes the output is the configuration around the matcher. Instructors can exclude quoted material, exclude the bibliography, exclude matches below a word threshold, or exclude a specific source, and each of those switches alters the final percentage without a single word of the essay changing [2]. This is the practical reason two reports on the same document can show different numbers: they are answering slightly different questions. When you read a report, always check the exclusion settings before you react to the headline figure.
A second layer now sits alongside text matching: AI writing detection. Modern platforms score text for statistical patterns associated with generative models, and that score is reported separately from similarity, because a fully original essay written by ChatGPT has near-zero similarity and can still carry a high AI indicator. Treating the two signals as one number is the most common misreading students make.
What Does a Similarity Score in a Plagiarism Report Really Mean?
A similarity score is the percentage of a document's text that matched at least one indexed source — nothing more and nothing less [3]. It is not a plagiarism score, it is not a misconduct finding, and it is not comparable across assignments. A literature review with heavy direct quotation can legitimately score in the twenties, while a short reflective piece with one badly paraphrased paragraph can score in single digits and still contain a genuine integrity problem.
Because the number is descriptive rather than evaluative, institutions set their own expectations. Some departments publish a review threshold, others explicitly refuse to publish one on the grounds that context decides, and many ask markers to read the match list rather than the percentage [3]. The match list is where the real information lives: it shows which source matched, how many words matched, and whether the matched text sits inside quotation marks with a citation attached.
The practical skill is reading the report the way a marker does. Work down the matches from largest to smallest, and for each one ask a single question — is this attributed? If the answer is yes, the match is expected and defensible. If the answer is no, you have found the paragraph that needs rewriting before the deadline, which is exactly the outcome a pre-submission check is for.
Can Students Run an Anti-Plagiarism Check on Their Own Draft Before Submitting It?
Inside an institutional Turnitin account, students generally cannot run a check whenever they like. Self-checking depends on the instructor enabling a draft submission or a similarity-check assignment, and the report may be hidden from the student depending on course settings [4]. That design is deliberate: the system is licensed to the institution, and the instructor controls what students can see and when.
Where self-checking is available, the important detail is the repository setting. A standard submission is added to the institutional paper database, which means your draft becomes a future match source for other students. A non-repository check runs the same comparison but does not retain the file in the student paper index, so a draft can be tested without side effects [4]. If you are checking early versions of a dissertation chapter, this distinction matters.
Independent checking services solve the access problem for students whose courses do not enable drafts. They run the draft through the same comparison process and return the report before the real deadline, so citation gaps and weak paraphrases are caught while revision is still possible rather than after a grade is returned [4]. The trade-off to check before using any service is whether the file is retained — a checker that quietly adds your draft to a shared database can create the very match it was meant to prevent.
Reading the report is only half the job; the other half is having a report to read while there is still time to act on it. turnitin0 gives students a pre-submission check that returns the same AI detection and similarity reports their institution sees, delivered in under 15 minutes in 98% of cases, without adding the file to Turnitin's student paper database. Upload a.docx,.pdf, or.txt between 300 and 30,000 words, and both PDFs arrive in one checkout.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Does anti-plagiarism software store my paper after I submit it?
Standard institutional submissions are usually added to the student paper database, while non-repository checks run the same comparison without retaining the file [4]. Always confirm the repository setting before uploading a draft you may reuse.
Is a 20% similarity score bad?
Not inherently. The percentage only measures matched text, and properly cited quotations and reference lists both count toward it [3]. Markers read the match list to see whether the matched passages are attributed.
Can anti-plagiarism software detect AI-generated writing?
Text matching and AI detection are separate signals. Similarity compares your words against indexed sources, while AI detection scores statistical patterns in the writing itself, so an original AI-written essay can show 0% similarity and still carry an AI indicator [1].
Will paraphrasing a copied sentence lower the score?
Sometimes, but not reliably. The matcher looks for overlapping phrases, so light synonym swaps often still match, whereas genuine restructuring with a citation usually does not [2].
Why do two plagiarism checkers give different scores for the same essay?
Because they index different databases and apply different exclusions. Match thresholds, quoted-text exclusion, and bibliography exclusion all shift the percentage without changing the document [2].