Turnitin scores ยท updated 2026-11-09

Why the same paper gets different Turnitin scores

Five reasons two Turnitin checks of the same paper show different numbers: file type, stored student papers, account settings, exclusions and model updates. With figures from checks we ran.

Two students check the same paper and get 43% and 52% AI. A paper that scored 62% similarity in March scores 2% in November. Students see this and conclude that one report must be fake. Usually neither is. Here are the five reasons, with numbers from checks we have run.

1. File type

Turnitin reads a Word file and a PDF differently. In a PDF it reads only the visible text layer and may take tables and columns in a different order. One paper, checked the same evening: the Word file read as 6,829 words and scored 52% AI; the PDF of the same paper read as 7,200 words and scored 43%. Same words, different extraction, different score.

Rule: check the exact file you will submit.

2. Stored student papers

The similarity score depends on which sources the account searches. Every Turnitin account searches the web and publications. Only some search the database of papers students have submitted before. One student's old paper, handed in at his college in 2023, scored 62% similarity on an account that searched stored papers (a 60% match to his own college's copy) and 2% on an account that did not. Both reports were genuine Turnitin reports. They were looking at different libraries.

Rule: a low similarity score on a pre-check does not clear a paper you, or anyone else, has handed in before. Your school's Turnitin searches its own stored papers.

3. Account settings and exclusions

Instructors can exclude quotes, the bibliography and small matches from the similarity score. A report with "exclude bibliography" on will show a lower number than one without. The AI score is not affected by these settings, the similarity score is.

4. Detector updates

Turnitin updates its AI model. Reports made before an update are not recalculated. The same paper scored 74% AI in March 2026 and 29% on one account in November; two other checks the same week scored it 77%. Dates and accounts matter, and a single odd result is a reason to check again, not a verdict.

5. What counts as prose

The AI percentage covers only long-form prose. A paper with many tables, lists or references has less qualifying text, so a few flagged paragraphs weigh more. Two versions of a paper with different formatting can score differently for this reason alone.

What is the same everywhere

The AI detector itself. Whatever account runs the check, the model that scores English prose is Turnitin's AI writing model. Differences in the AI score come from the text it was given (file type, formatting) and from updates over time, not from which school owns the account.

How to use this

  • Check the file you will submit, in the format you will submit it.
  • Treat the AI score as a reading of your text; treat the similarity score as a reading of one library.
  • If you reuse an old paper, assume your school will match it even when a pre-check shows 2%.
  • If two reports disagree a lot, look at the submission IDs, the word counts and the dates on page 1 before deciding anyone is lying.

We state on every order page what our similarity check covers. Check a paper.

See your own score before you submit

Paste or upload your paper and get the Turnitin AI report back, the similarity report with it; the live turnaround is shown before you pay. If the score is high, you can fix it here.