Can Turnitin check a whole thesis?

Turnitin does not generate an AI writing report for a submission over 30,000 words, or for one with fewer than 300 words of prose. Two of its own help pages state the same limits. Outside that band there is no percentage: the indicator shows a grey double dash, which means the submission could not be processed, not that it came back clean.

A full thesis is normally well past the ceiling. So if you are reasoning about "my thesis AI score", the first question is which file produced it.

The first is a short page called File requirements for an AI Writing Report. The whole page is about 700 characters. It opens with "In order for a submission to generate an AI writing report and percentage, the submission needs to meet the following requirements", then lists them. Three of those decide whether a long document gets scored at all:

• File size must be less than 100 MB
• File must have at least 300 words of prose text in a long-form writing format
• File must not exceed 30,000 words

The rest of the list covers supported languages and accepted file types.

The second page is the AI writing detection capabilities FAQs, which repeats the same requirements inside its explanation of what the report indicators mean. Same numbers, same order.

Having it twice on the vendor's own site matters more than it sounds, because most of what circulates about Turnitin's limits is one blog copying another blog.

The FAQ describes what the indicator can show. A blue indicator with a number between 0 and 100 means the submission processed. An asterisk means AI was detected in the 1% to 19% range, which Turnitin displays without a number, and without highlights, to avoid false positives. And then:

"Gray with no percentage displayed (- -): The AI writing detection indicator is unable to process this submission."

The page gives the reasons. One is that the submission predates the release of AI writing detection, which can only be fixed by resubmitting. The other is that the file does not meet the requirements above.

A separate error state, shown as an exclamation mark, means Turnitin failed to process the submission and asks you to try again or contact support.

So the grey dash on a 60,000-word thesis is the expected outcome, and it carries no information about your writing. It is worth knowing which of the three states you are looking at before drawing any conclusion, especially if what reached you was a screenshot.

In practice, long work is submitted in pieces: a chapter to a supervisor, a section to a progression panel, the final file to the repository. Each of those is a separate prediction, and this is where the arithmetic stops being intuitive.

Turnitin describes the mechanism as cutting a submission into overlapping segments, scoring each segment between 0 and 1, letting each qualifying sentence inherit the score of the segments containing it, then pooling and aggregating those into the document number. Chapter scores are not slices of a thesis score, because the thesis score does not exist. They are separate runs over different amounts of text.

The short end of the range behaves differently again. Turnitin says that in documents of only a few hundred words the prediction is mostly "all or nothing", because there is a single segment and no overlap to average across, and that this means mixed content in a short document can be flagged as entirely AI-generated. A 400-word introduction submitted on its own sits right in that zone. An abstract on its own does not clear the 300-word prose floor at all.

At the other end, the FAQ is candid about long documents: in a longer document mixing authentic and AI-generated writing, it says, it can be difficult to determine exactly where one ends and the other begins, and describes the output as a guide for starting a conversation with the student.

Even inside the band, the number covers less than you think. Turnitin analyses what it calls qualifying text, which it defines as prose sentences, blocks written in standard grammatical sentences. Lists, bullet points and other non-sentence structures are outside it. The FAQ then states plainly that the percentage is not necessarily the percentage of the entire submission, and that text not considered long-form prose is not included.

Its definition of long-form prose names the format directly: individual sentences contained in paragraphs that make up a longer piece of written work, such as an essay, a dissertation, or an article.

Now picture a methods chapter. Numbered procedure steps, a parameter table, equations, code, figure captions. Very little of it is paragraphs of sentences. The denominator for the percentage shrinks to the discussion-shaped parts of your writing, which is also where the score comes from. How to read a Turnitin AI writing report works through the gap between the number and the highlighting.

Turnitin does not publish which counter it uses, and that is not a nitpick, because two tools counting the same file routinely disagree.

Dashes are the usual culprit. Word treats a dash between two words as a word boundary in some forms and not in others, which is demonstrated neatly on a UT Austin law-writing blog using a page range, and we have watched our own counters disagree with Word on documents full of them.

The margin is small. It only matters in one place, and this is it: if Word says 29,800 and whatever counts on the other end says something slightly larger, the difference between a report and a grey dash is a handful of dashes. Near the ceiling, split the file rather than find out afterwards.

1. Ask which file was submitted before interpreting any score. A chapter, a compiled thesis, a portal-generated PDF and an appendix-stripped version are four different submissions.
2. Treat a grey dash as "not processed". Do not report it upward as a clean result.
3. Do not add chapter percentages together, or average them. They are separate predictions over different denominators.
4. Expect short sections to behave badly. Turnitin's own explanation of all-or-nothing scoring on short text is the reason, not your writing.
5. Remember what the vendor says the number is for. Turnitin states across several of its own pages that the model can be wrong and should not be the sole basis for adverse action against a student, and elsewhere calls the score a single data point rather than a definitive response. That is a fair sentence to bring into a supervision meeting.

If you are already in that conversation and you wrote the thing yourself, flagged, but you wrote it yourself covers what evidence is worth assembling.

Two things get expensive at chapter length: what you have to reread afterwards, and what you paid to have rewritten. Both are decided by the same setting, which is how much of the file was in scope. Three flagged paragraphs and a whole chapter are very different amounts of proofreading.

HumanPen works the other way round. Import the Turnitin or iThenticate report for that chapter and its flagged passages become the scope: only that text is rewritten, everything else is left alone, and the billing counts the words actually rewritten rather than the document you uploaded. A paragraph is the smallest unit the engine touches, and a selection covering part of one is expanded to the whole paragraph and shown to you before the job starts. There is no per-request word cap on the upload side, so a chapter goes in as one file. Eligible results can continue lowering AI for free.

None of that is a prediction about your next report. It is a statement about what gets touched and what gets billed.

Can Turnitin check a document longer than 30,000 words? Not for AI writing. Turnitin's file requirements page and its FAQ both state that a submission must not exceed 30,000 words for an AI writing report and percentage to be generated. Similarity checking is a separate system with its own behaviour.

My thesis came back with no AI percentage. Is that good? It is most likely that no report was produced. A grey double dash means the indicator was unable to process the submission, and exceeding the file requirements is one of the reasons Turnitin lists for it.

Why does my chapter score differently from the file my supervisor submitted? They are different runs. Scores are computed over overlapping segments of whatever was submitted, and the percentage covers only qualifying prose, so two files containing the same words can produce different denominators and different numbers.

Is there a minimum length? Yes. The file needs at least 300 words of prose text in a long-form writing format. Turnitin also says predictions on documents of only a few hundred words tend toward all-or-nothing, because there is a single segment with no overlap to average across.