Turnitin for students: what it checks, what it flags, and what it misses

Turnitin is probably the most widely used plagiarism checking tool in secondary schools and universities globally. Most students hand in work through it without a detailed understanding of what it's doing. A better understanding helps you interpret your similarity report correctly, avoid inadvertent flags, and use any self-check access you have effectively.

Turnitin compares your submission against its database, which includes websites and internet content, published academic journals and books, previously submitted student work from institutions using Turnitin, and certain news sources and repositories. The comparison is primarily text-matching: it looks for sequences of words that appear in both your submission and something in the database.

The output is a similarity percentage: the proportion of your submitted text that matches something in the database. It does not score the quality of your writing, assess whether AI was used, or make a determination about whether plagiarism occurred. That determination is always made by your teacher or institution - Turnitin produces evidence for that decision, not the decision itself.

When you (or your teacher) view a similarity report, you see highlighted sections of your text alongside the matched source and the percentage of the total document that source accounts for. A 25% overall similarity score might be entirely harmless if it's composed of properly quoted and attributed material, standard academic phrases, and your bibliography.

The threshold that matters varies between institutions and subjects. A scientific report might legitimately have higher similarity due to standard methodology descriptions and specific technical terms that must be used as-is. A creative writing essay would normally have very low similarity. There is no single 'red flag' percentage that applies universally - context determines whether any given score is a concern.

Turnitin added an AI writing detection feature in 2023. It's available on institutional licences, though not all universities and schools have enabled it. The AI detection module uses a separate analysis from the similarity check and produces its own score alongside the similarity percentage.

The AI detection capability is subject to the same limitations as other text-based AI detection tools: false positives for non-native English writers, reduced accuracy on mixed-origin content, and the fundamental issue that it measures textual properties rather than authorship. Turnitin publishes claims about its false positive rate, but independent validation of those figures in real classroom conditions is more limited than vendor literature suggests.

Many universities provide students with self-check access to Turnitin, either through the library website, the VLE (Canvas, Moodle, Blackboard), or a dedicated plagiarism checking portal. This lets you see a similarity report before your final submission and address any issues. If your institution offers this, it's worth using - you see the same database comparison your marker will see.

If you're not sure whether you have self-check access, ask your library, your department administrator, or your supervisor. Many institutions have it configured but don't actively promote it. For dissertations in particular, running a self-check a few weeks before submission is standard good practice.

If your similarity score is higher than expected, work through the report section by section. Legitimate causes are very common:

Turnitin is a powerful similarity-checking tool with known blind spots. It can't check sources that aren't in its database - paywalled content from publishers without agreements, recent publications not yet indexed, and privately circulated materials. It doesn't check translated text effectively: an essay plagiarised from a French source and translated to English will score low similarity.

Most significantly, Turnitin sees the submitted text, not the process that produced it. A submission pasted in from an AI tool looks the same as one typed and revised over two weeks - unless the source material happens to be in the database. For detecting shortcuts that don't leave a textual trace, process-based tools are more appropriate.