Turnitin vs AI detector: what schools actually need to know

Turnitin is the default answer to 'academic integrity software' in most schools and universities. It's been around since 1998, has an enormous reference corpus, and is deeply embedded in many institutions' assessment workflows. But as AI generation has become the dominant academic integrity challenge, it's worth asking honestly: what does Turnitin actually detect, and where are the gaps?

What Turnitin does

Turnitin is primarily a plagiarism detection system. It compares submitted text against a vast database of web content, academic publications, and previously submitted student work, then flags sections showing substantial similarity to existing content. The similarity score it produces is one of the most widely recognised outputs in academic assessment, and for traditional plagiarism - copying from sources - it remains a robust tool.

In 2023, Turnitin added an AI writing detection feature. This works differently from the similarity-matching core: it uses statistical text analysis to estimate the likelihood that content was AI-generated, producing a percentage score. The combination means a single Turnitin submission now generates both a similarity score (for plagiarism) and an AI writing indicator (for generation).

Where Turnitin's AI detection struggles

The AI detection feature Turnitin added uses the same category of method as other text-based detectors - statistical analysis of text patterns. This means it inherits the same fundamental limitations. The ESL false positive problem applies: non-native writers produce text with statistical characteristics that the tool associates with AI generation, at higher rates than native speakers. Turnitin's own published guidance acknowledges this and cautions against using the AI indicator as the sole basis for action.

Paraphrasing circumvents it. A student who takes AI-generated text and rephrases it - using one of several widely available paraphrasing tools - will typically produce output that scores low on AI detection whilst still having done no meaningful intellectual work. This isn't a Turnitin-specific problem; it applies to all text-based AI detection methods.

What Turnitin doesn't capture

The most significant gap in Turnitin's coverage is process data. Turnitin receives a finished document; it has no visibility into how that document was produced. If a student typed their work over two hours, revised it extensively, and submitted a genuine draft, Turnitin sees exactly the same thing as if they pasted AI-generated text into a document and submitted it unchanged. The document contains no information about its own production history.

This isn't a criticism specific to Turnitin - it's an inherent limitation of any tool that works on submitted documents rather than on the submission process. But it means that a school relying entirely on Turnitin for AI detection is working with a tool that cannot access the most reliable evidence available.

Cost and procurement considerations

Turnitin operates on institutional licensing models, with pricing typically negotiated on a per-submission or per-seat basis. For large universities, the value proposition is clear: the scale justifies the cost and the integration is deep. For smaller schools, independent sixth forms, or individual teachers, the pricing structure can be a significant barrier - and Turnitin is not designed for small-scale or individual use.

Turnitin's AI detection feature continues to evolve; the company has been transparent that accuracy on different types of content and student populations is still being refined. Schools making procurement decisions now should expect the product to look somewhat different within a couple of years.

What to look for in an alternative

If you're evaluating alternatives - whether to Turnitin specifically or to document-submission-plus-text-analysis more generally - the most valuable capability to look for is process data. A tool that records writing behaviour during submission gives you qualitatively different evidence from one that only receives a finished document. This evidence is fairer (language-neutral), harder to circumvent, and more defensible in formal proceedings.

Other factors worth weighing: the teacher workflow and how much additional overhead the tool creates; student-facing transparency about what is being recorded; data protection provisions, including where data is stored and under what terms; and the transparency of the AI scoring methodology. Tools that can't explain clearly how they produce their scores are difficult to defend in a formal challenge.

Questions worth asking any vendor

Before committing to any academic integrity tool, these questions are worth asking directly. What is the false positive rate for non-native English writers, and can they provide data? What happens when a student challenges a finding - is there a clear appeals process? Where is student data stored and who can access it?

Does the tool capture process data as well as submitted text? Can teachers adjust settings or set per-assignment policies? What training and support is provided for teachers new to the tool? A vendor that answers these questions clearly and with evidence is usually a more reliable partner than one who deflects or responds primarily with marketing language.