The job
What the Llama assignment checker online reports
You have a section of an assignment and you want to know how it reads before somebody else reads it. This returns findings: what the structure is doing, where the argument goes thin, where the prose runs even, what stands out.
It is not a plagiarism checker. Nothing is compared against a database, an index or a set of sources, because none exists on this route. No similarity percentage is calculated and none is reported. If a similarity report is what you need, this is the wrong page and it is better to say so here than to imply otherwise.
The run happens on our server: 5 free runs per rolling 24 hours, 1 to 12,000 characters per run, no account. Plan the session around five runs and spend them on the sections that matter.
The workflow
How to check an assignment in three steps
Three steps, and the third is the one that turns findings into edits.
- Step 1
- Paste one section, split at a heading, 1 to 12,000 characters. A complete argument reads better than an arbitrary slice of a chapter.
- Step 2
- Ask a specific question in the instruction box, 300 characters. Is the evidence doing work, are the claims hedged, does the structure repeat. A narrow question beats no question.
- Step 3
- Take each finding back to the draft and confirm it by eye. Anything you cannot confirm is an observation about text, not a fact about your work.
Before you run anything
The manual pass to do first
These four checks need no tool at all, and they catch what markers actually notice. Do them first and the run is more useful, because you are asking about what is left.
- Look up every citation
- Confirm the source exists, says what you claim it says, and supports the sentence it is attached to. Invented or misattributed references are the most damaging problem in the list and the easiest to find.
- Read for a change of voice
- Find the point where the writing shifts from your usual voice to something flatter and more even. A draft assembled in stages reads as assembled.
- Test the analysis
- For each paragraph, ask what it claims that is not obvious. Tidy structure over surface-level insight is a pattern a marker spots in seconds.
- Measure the paragraphs
- If they are all the same length and start the same way, the uniformity itself is the problem. Break it by hand before you run anything.
Output
What the findings look like, and what they are not
Prose, not a dashboard. The instruction our server sends asks for concise, useful findings and explicitly forbids claiming certainty about authorship, plagiarism or which model produced any text.
That means no number comes back. No AI percentage, no originality score, no risk level. There is nothing to screenshot and nothing to submit alongside the work, which is deliberate, because a number attached to a draft would be treated as evidence and it would not be evidence.
The upside is that every finding is checkable. If the output says the third paragraph asserts without support, you can look at the third paragraph. A score offers nothing to verify.
The honest answer
Can this prove an assignment was written by AI?
No. Guidance written for teachers is explicit that a detector result should support an investigation rather than serve as the sole basis for an accusation or a penalty, and it puts running any tool fifth in a seven-step process.
That process is worth following whichever side of the desk you are on. Confirm what was permitted, look for a reason to review closely, read the work in context, verify the facts and sources, then run a check, then talk to the person, then decide on several pieces of evidence together.
Drafts, notes, sources and document version history all describe how the work was made. Text analysis only describes how the finished text reads, which is a much weaker thing.
Fairness
Why careful writers get flagged
Anything reading text for signals will find them in text a person wrote carefully. The pattern is consistent and worth knowing before anyone acts on an output.
Non-native English speakers write with safer sentence shapes and a tighter working vocabulary, which produces exactly the evenness that reads as generated. So does anyone taught to write to a template, which is most people taught to write essays.
So treat evenness as a prompt to look, never as a conclusion. That is also why this route returns descriptions rather than a number: a description invites a second look, and a percentage invites agreement.
Constraints
Llama assignment checker limits to plan around
Five runs per rolling 24 hours from first use, one section per run, 1 to 12,000 characters. No file upload, so copy the text out of your document first.
No plagiarism database, no source lookup, no citation verification, no grammar correction and no rewriting. It reads and it reports, and each of those absences is a thing some tool with this name does and this one does not.
No history, no account, no API and no batch queue. And the Llama in the name refers to the work people bring, not to the engine, which does not run Llama and is not connected to Meta.
In practice
Where the Llama assignment checker fits
In a revision session, after the manual pass and before the final read. The findings are most useful on a section you already suspect is weak, because then you are testing a hypothesis rather than fishing.
For teachers it belongs where the guidance puts it: one signal, late in the process, never the deciding one, and never a substitute for reading the work or talking to the student.
For a similarity report against published sources, use a tool with a database. This page says so rather than leaving you to discover it from an output that never mentions sources.