The tool
What the Llama Turnitin checker does
First, what it is not. This is not Turnitin, it is not connected to Turnitin, and it cannot run a Turnitin check or see a Turnitin score. Nobody outside an institution's licence can. Any site that says it will tell you your number is guessing.
What it does is read the text you paste and describe it: where the sentences all land at the same length, where every paragraph opens the same way, where transitions are generic, where a claim sits without an example under it. Those are the features of writing that attract attention from a human marker and from a detector, and they are things you can actually change.
Context
What Turnitin's AI writing indicator actually is
It is a pattern detector, not a database search. Its similarity report compares against sources; the AI indicator does something different and looks at statistical regularities in the writing itself. As one explainer puts it, machine-written prose tends to be statistically predictable, and sentence lengths stay consistent.
The number it produces is reported in bands, and the low band is often not shown at all. According to that same explainer, scores below 20 percent are typically hidden from instructors to reduce misinterpretation, 20 to 50 percent reads as mixed, and above 50 percent tends to prompt a closer look. Whether a student can see any of it depends entirely on how their institution has configured the system, because the tool is built for instructors rather than for student self-checking.
How it works
How to run the Llama Turnitin checker in three steps
This is a server tool. Your text is sent to our AI server, read there and returned as written findings.
- Step 1
- Paste one section of the draft, between 1 and 12,000 characters, which is roughly two thousand words.
- Step 2
- Add an instruction of up to 300 characters if you want a focus, for example: this is a discussion chapter and it should sound like an argument.
- Step 3
- Read the findings and fix what is actually weak. You get five checks per rolling 24 hours.
The uncomfortable part
Why honest writing gets flagged anyway
Detectors make two kinds of mistake and both are documented. A university guide for its own faculty cites a 1 percent false positive rate on documents containing 20 percent or more machine-written text, and says the tool may miss about 15 percent of machine-written text in a document. One number means innocent papers get flagged. The other means flagged papers are not the only ones that used a model.
The causes of a false flag are ordinary virtues. Formal tone, technical vocabulary and heavy editing all push writing toward the uniformity a detector reacts to, which is why strong, polished, carefully revised work can look machine-generated. Multilingual writers are also worth a specific mention: the vendor says its tool shows no bias against them, and the same university guide notes that other evaluations have identified issues and more peer-reviewed study is needed.
The honest answer
Why the Llama Turnitin checker gives no percentage
Because any percentage here would be invented. We have no access to Turnitin's model, no calibration against it and no way to know what it will do with your file next week. A number would look authoritative and mean nothing, and a student who relaxed because our number was low would have been misled.
The check is also instructed never to claim certainty about authorship or about which model wrote a text. So what comes back is a description: this paragraph repeats the same construction four times, this section has no concrete example in it. You can check every one of those against your own page, which is more than a percentage ever lets you do.
If it has already happened
What to do if your work has been flagged
Take the framing from the institutions themselves. One university's guidance tells its faculty that the AI detector should be seen as a starting point for a conversation rather than a strict enforcement tool, that they should use the result to start a conversation with the student, and that they should stay open to the possibility that the result was incorrect. That is your ground.
Bring evidence of process rather than argument about the score. Draft history, version dates, notes, search history, the messy earlier version with the wrong structure in it. A detector reads the final text only, and the final text is the one piece of evidence that cannot show how it got there. The same university disabled the tool for a term in 2023 before turning it back on in January 2024, which is a reasonable measure of how settled any of this is.
Privacy
Where your draft goes when you check it
To our AI server. Reading a passage needs a language model, so the text leaves your browser. The watermark and hidden character tools on this site run locally and upload nothing; this route is not one of them.
Nothing is submitted to any academic database, nothing is stored as a paper and no account is created. That matters here specifically, because unlike a real submission, checking something here does not put a copy of your work into a repository that a later check could match against.
In practice
Who uses the Llama Turnitin checker, and when
Students a few days from a deadline who want to know whether their own writing has gone flat. Students who used a model to draft an outline and rewrote it, wanting to know how much of the original shape survived. Writers whose work has been questioned once and who are now anxious about every formal sentence they produce.
The useful outcome in all three cases is the same, and it is not a score. It is a list of specific places where the writing is generic, which happens to be the list of places worth editing whether or not anything ever gets flagged.