The tool
What the Llama thesis checker does
Paste a section of a thesis and the Llama thesis checker reads it and writes back what it sees: whether the central claim is stated plainly enough to argue with, whether the paragraphs that follow actually support that claim, where the prose repeats itself, where the register slips. The answer arrives as sentences, not as a dashboard.
Three different tools go by this name and they do different jobs. One grades a thesis statement against a rubric. One proofreads a whole manuscript and hands back a file with track changes. One prints a probability that a machine wrote it. This is none of those. It reads a passage and describes it, which is the step before you decide what to change.
The subject
What a thesis check actually looks at
Two things sit in the same passage and are easy to confuse. The first is the argument: is there a claim, is it specific, do the supporting points connect to it, does the conclusion end up somewhere the opening promised. The second is the prose: sentence variety, repeated openers, hedging, the vocabulary a section leans on.
Findings come back on both, because in practice they fail together. A section with no clear claim tends to circle, and circling reads as repetition. Naming the argument problem usually fixes more sentences than editing the sentences would.
How it works
How to run the Llama thesis 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. Between 1 and 12,000 characters, which is roughly two thousand words, so an introduction or a single chapter section rather than the whole thesis.
- Step 2
- Add an instruction of up to 300 characters if you want a particular focus, for example: check whether the argument in this section matches the claim in the first paragraph.
- Step 3
- Read the findings, edit one thing, and re-run the same section if you want to see whether it landed. You have five checks per rolling 24 hours.
Length
How much of a thesis you can check at once
12,000 characters per run, and there is no file upload. A full thesis will not fit, and nothing here pretends it will. Tools that take a 60,000 word PDF exist and this is not one of them.
In exchange you get something a whole-document scan is bad at. Sections have different jobs. An introduction that hedges is a problem, a methods section that hedges is doing its job. Checking one section at a time means the findings are about that section's job rather than an average across the document.
The honest answer
Why the Llama thesis checker gives no AI percentage
It will not tell you that a thesis is 82 percent likely to be machine written, because that number does not exist here. The instruction sent with every check says never claim certainty about authorship, plagiarism, or which AI model wrote text, and describe observable signals and limitations instead.
Detectors that do print a percentage publish false-positive rates alongside it, and a false positive lands on a real student with a real submission. What you get here instead is a description of the writing: this paragraph repeats the same construction, this section has no sentence longer than twenty words. Those are things you can check against the text and fix. A percentage is not.
Common confusion
Why a literature review reads as machine written
Academic genres are conventional on purpose. A literature review moves through sources in a fixed rhythm. A methods section names materials and steps in a fixed order. Both come out uniform, hedged and repetitive, because that is what the genre asks for, and uniform hedged repetition is exactly what a reader calls robotic.
So read findings about a review or a methods chapter with that in mind. Flatness there is often correct. Where it is worth acting on is in the introduction, the discussion and the conclusion, which are the sections meant to sound like someone arguing.
Privacy
Where your thesis text goes when you check it
It goes to our AI server. This route needs a language model to read a passage, so the text leaves the browser. The watermark and hidden character tools on this site are the local ones, and they upload nothing. This is not one of them.
No account is made and no email is collected, and the five check allowance is counted against a hashed identifier rather than a profile. If your work is under embargo or covers confidential data, that is a reason to paste a representative section rather than the sensitive one.
In practice
Who uses the Llama thesis checker, and when
Postgraduates a week from submission who want to know if the introduction still promises what the conclusion delivers. Supervisors reading a chapter quickly and wanting the obvious prose problems listed before they write comments. Writers who used a model to draft a summary section and want to know how much of it still reads as a model wrote it.
The pattern in each case is the same: something feels off in a section and a second reading would help. This gives you that reading in a few seconds, with no verdict attached and nothing changed in your file.