Direct answer
AI detection tools are software systems that analyse written text and estimate how likely it was produced by a generative model such as ChatGPT, Claude, or Gemini. The tool that matters most in higher education is Turnitin's AI writing detection, which is built into the Similarity Report that instructors already use and analyses only long-form prose, excluding reference lists, quotations, and short snippets [1]. These tools output a probability-style signal rather than proof, so a score is a prompt for review, not a verdict [1].
How Do AI Detection Tools Actually Work, And How Accurate Are They?
Most detectors do not "recognise" AI writing the way a plagiarism checker matches a source. They score statistical properties of the text, most commonly perplexity — how predictable each word choice is given the surrounding words — and burstiness, the variation in sentence length and rhythm across a passage [2]. Human academic prose tends to be less predictable and more unevenly paced, while unedited model output often sits in a narrower band on both measures, which is why detectors can separate the two at all [2].
That statistical approach also explains the error profile. Accuracy falls on heavily edited AI text, on writing produced by non-native English speakers, and on short passages that give the model too little signal to score confidently [2]. False positives and false negatives both occur, which is precisely why Turnitin publishes performance data for its model and warns institutions against treating a single number as evidence of misconduct [2].
The practical takeaway is that no detector should be read as a yes/no answer. A high score means a segment of prose resembles model-generated text; it does not establish who wrote it, when, or with what level of assistance [2]. Instructors are expected to open the flagged segments, read them in context, and weigh them alongside drafts, supervision notes, and the student's own account.
Which AI Detection Tool Do Universities Use, And What Does The Turnitin AI Score Mean?
In practice, the detector most students encounter is Turnitin, because it is embedded directly in the institutional learning management system and in the assignment workflow instructors already run [3]. That matters for a simple reason: the AI percentage a student sees in a Turnitin report is the same figure their marker sees, so there is no translation layer between the two [3]. Third-party detectors may disagree with it, but they are not the tool the grade depends on.
The AI score itself is narrower than most people assume. It represents the share of qualifying prose in the document that the model flagged as likely AI-written — not the share of the whole submission, and not a measure of intent [3]. Reference lists, block quotations, and very short fragments are excluded from the analysis, so a document can carry a high percentage while most of its words were never scored at all [3].
Turnitin also displays an asterisk percentage, shown as *%, when the detected signal falls below its 20% confidence threshold [3]. Students frequently misread that symbol as "zero AI" or as a technical glitch. It actually means the model found some indication of AI writing but not enough confidence to publish an exact figure — a low-confidence signal that still deserves attention before submission [3].
Can You Check Your Own Draft With Turnitin AI Detection Before You Submit?
For most students, the honest answer is no — not through the institution. Whether a student can run a Turnitin check on their own draft depends entirely on whether their university has enabled that access, and many have not, because the standard workflow routes submissions through the assignment inbox [4]. That leaves a real gap: you can finish a draft, feel uncertain about a paragraph you rewrote with a model, and have no way to see the same signal your marker will see until after the deadline has passed [4].
Independent pre-submission checkers exist to close that gap. They return the same style of AI and similarity report a student would receive institutionally, without adding the file to the Turnitin student paper database, so the check itself does not create a match against the student's own future submission [4]. The value is not the number alone — it is the ability to open the flagged segments, revise the passages that genuinely read as machine-generated, and re-check before anything is final [4].
This is also the more defensible way to use detection tools. Reviewing your own draft and editing flagged prose is revision, and revision is what the tools are supposed to encourage; discovering the score only after submission turns a fixable writing problem into an academic integrity conversation [4]. Students who check early tend to arrive at the same conclusion: the score is useful as feedback, and useless as a surprise.
If you want to see that report before your marker does, turnitin0 gives you the same Turnitin AI and similarity reports instructors see, delivered in minutes, so you can revise the flagged passages while you still have time.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Are AI detection tools accurate enough to be used as evidence?
No single score should be treated as proof. Detectors score statistical signals such as perplexity and burstiness, and accuracy drops on edited AI text, non-native English writing, and short passages, so both false positives and false negatives occur [2]. Turnitin presents its percentage as one signal for instructors to review in context [1].
What does an asterisk percentage (*%) mean in a Turnitin AI report?
It means the detected signal fell below Turnitin's 20% confidence threshold, so the exact figure is withheld [3]. It is a low-confidence indicator, not a statement that no AI writing was found, and it is worth reviewing the highlighted segments before you submit [3].
Do universities use the same AI detection tool as third-party checkers?
Not necessarily. Turnitin is the detector most commonly embedded in institutional workflows, which is why its score is the one markers actually see [3]. Independent checkers can return a comparable report for self-review, but they are not the tool the grade depends on [4].
Can I run a Turnitin AI check on my own draft?
Usually not through your university, since self-check access depends on whether your institution has enabled it [4]. Independent pre-submission services exist that produce the same style of AI and similarity report without adding your file to the student paper database [4].
Should I revise text that a detector flags?
Yes, if the flagged passage genuinely reads as machine-generated. Opening the highlighted segments, rewriting them in your own voice, and re-checking is revision rather than evasion, and it is far easier to do before submission than after [4]. Tools are most useful as early feedback, not as a post-deadline surprise [1].