The tool
What the Llama assignment checker does
Paste an assignment and it reads the text and tells you what it notices: how the writing is put together, where it reads evenly, where the argument thins out, what stands out about the structure. It comes back as findings in plain prose.
It does not compare your work against a database, because there is no database. Nothing is searched, nothing is matched against sources, and no similarity percentage exists to report. Half the tools with this name are plagiarism checkers. This is not one of them, and that is worth knowing before you paste anything.
The text goes to our server for the run: 5 free runs every rolling 24 hours, no account, and an input between 1 and 12,000 characters.
Being straight
No score, no verdict, no accusation
The instruction our server sends with your text tells the model to describe observable signals and limitations, and never to claim certainty about authorship, plagiarism or which model wrote something. That constraint is built into the run rather than added as a disclaimer.
So there is no percentage. No AI probability, no originality score, no confidence figure. Nobody can read a piece of text and know who wrote it, and a number printed next to a paragraph does not change that.
What you get instead is a description you can argue with. If it says the paragraphs are uniform in length and the claims are never qualified, you can look and see whether that is true.
How to
How to check an assignment in three steps
One section per run, because five runs a day does not stretch to a whole dissertation.
- Step 1
- Paste one section, between 1 and 12,000 characters. Split at a heading so the model sees a complete piece of argument rather than half of one.
- Step 2
- Say what you want looked at in the instruction box, up to 300 characters. Structure, evidence, repetition, whether the claims are hedged. A narrow question gets a more useful answer than none.
- Step 3
- Read the findings against your draft and check them yourself. They are observations about the text, not facts about how it was made.
The problem
What people actually look for in an assignment
Markers notice these before any software is involved, and you can check every one of them by hand right now.
- Sources that do not hold up
- Citations that are invented, misattributed or attached to a claim they do not support. This is the most damaging one and the easiest to verify: look each one up.
- A sudden change in voice
- A draft that goes from a distinct voice with rough edges to uniformly flawless prose partway through reads as two different pieces of work joined together.
- Thin analysis under tidy structure
- Well-organised paragraphs that restate the obvious. Surface-level insight arranged neatly is a pattern a marker recognises quickly.
- Everything the same size
- Matching paragraph lengths, repeated sentence openings, sections that all resolve the same way. Uniformity is what reads as machine-made more than any individual word choice.
The honest answer
Can a checker prove an assignment was written by AI?
No, and nothing can. Guidance written for teachers on this subject is clear about it: a detector result should support an investigation rather than serve as the basis for an accusation on its own.
The sequence those guides recommend puts a tool well down the list. Confirm what was allowed. Read the work in context. Verify the facts and sources. Then, maybe, run a checker. Then talk to the person. Then decide on several kinds of evidence together.
Process evidence beats text analysis every time. Drafts, notes, a reading list, version history in a document: those say something about how a piece was made. Text analysis only describes how the finished thing reads.
Who gets hurt
Why clean writing gets flagged
Any tool that describes signals in text will find signals in text that a person wrote. The people this lands on hardest are consistent: non-native English speakers, and anyone whose style is careful and even.
Writing to a formula because you were taught a formula produces the same evenness that a model produces. So does writing in a second language with a smaller working vocabulary and safer sentence shapes. Neither is evidence of anything.
That is one of the reasons this route describes rather than scores. A paragraph of findings invites you to check. A number invites you to believe it.
Limits
What the Llama assignment checker does not do
No plagiarism check, no similarity score, no source matching, no database lookup. It cannot tell you whether a source exists, because it does not look anything up.
It does not correct your grammar or rewrite anything. It reads and reports. No file upload, so a document has to be copied out first. No batch mode, no API, no history and no account.
The Llama in the name describes the kind of work people bring here, not the engine. This route does not run Llama and is not connected to Meta.
In practice
Who uses the Llama assignment checker, and when
Students reading their own draft before handing it in, looking for the flat spots rather than a verdict. The useful question is where the argument thins, not whether a checker approves.
Teachers use it the way the careful guidance suggests: as one signal among several, after reading the work and before any conversation, never as the thing that decides.
If what you actually need is a similarity report against published sources, that is a different kind of tool with a database behind it, and this page will not pretend to be one.