You can run a dissertation through an AI checker, and the way to do it is one chapter at a time rather than as a single document. Each check returns a 0-100 score and one of three verdicts, takes seconds, needs at least 65 words, and costs one credit whatever the length, so a full chapter costs exactly what a single paragraph costs. The free allowance is 10 credits a day per network with no account, no card and no email address, which is ten chapter checks a day. Splitting by chapter is the right method rather than a workaround. One score for an entire thesis is an average, and an average hides the one section that reads as machine-written.
The checker returns a 0-100 reading and one of three verdicts: reads as AI-generated, borderline, or reads as human-written. That reading describes how densely certain patterns appear across whatever you handed it. Give it a whole dissertation and you get the density across a whole dissertation, dominated by the many thousands of words you wrote yourself over several years. A methods section that reads completely flat vanishes into that average.
Nobody who reviews your work will read it that way either. A supervisor or an academic integrity panel looks at the passage that bothered them, usually a page or two. If your discussion chapter is the part that reads as machine-written, that is the reading you want in front of you before the meeting, and a document-level number will never surface it.
So run one check per chapter and keep the results. Chapter name, word count, score, verdict, date. Ten runs cost ten credits, which is a single free day, and they give you a map of the document instead of one number you cannot act on.
Split at chapter boundaries first. Where a chapter runs long, split again at section headings rather than mid-argument, because the rhythm measurements need whole paragraphs to mean anything. A check needs at least 65 words, which every real section clears.
Strip the material that is not your prose before you paste. A checker asked to score a page of citations gives you a number about citations.
A dissertation is not written in one voice or one sitting. A literature review drafted in your first year and a discussion written in the final month normally differ in sentence length, in how often you hedge, and in how much you restate the question before answering it. That unevenness is what a long human document looks like from the inside.
Uniform rhythm across every chapter is itself a signal. If each one comes back with the same sentence-length variance and the same verdict, something flattened them. A rewriting pass applied end to end does that. So does one heavy copy-edit, a template followed too literally, and generation. An examiner who has read your earlier drafts notices the flattening before any checker does, because your first-year voice has disappeared from your first-year chapter.
The practical consequence is simple. Do not run a rewriting tool over the entire thesis to make it consistent. A thesis needs consistent terminology, consistent citation style and a consistent argument. Consistent sentence rhythm in every chapter is the thing to avoid.
Methods chapters, by a wide margin. Procedure written to be reproducible is written in short declarative sentences, with fixed vocabulary, no synonyms and no variation in structure. That is the same profile as highly predictable text, and predictability is what a detector measures. Writing your methods section well makes it read more machine-written, not less.
None of that is evidence of anything. Formal academic prose written by a careful person is predictable by design, so read the score on these sections against the rest of your document rather than on its own.
This falls hardest on researchers writing in a second language. The one large published figure on it comes from Liang et al. at Stanford HAI in 2023: 61% of TOEFL essays by non-native English speakers were falsely flagged as AI across seven detectors. Doctoral writing in a second language has the properties that study measured: careful textbook constructions, and the word you are certain of rather than the one you half remember.
The humanizer rewrites every sentence in one click. A run takes up to 12,000 characters, roughly 2,000 words, needs at least 25 words, and costs one credit per 50 words, so a full 2,000-word block is 40 credits. On a document this size that makes it a scalpel, and a scalpel is what you want. Point it at the passages that came back reading as machine-written, not at the thesis.
Before spending anything, run the chapter through the registry of AI writing tells. It is free, unlimited, runs entirely in your browser, and it highlights the exact phrases a detector keys on instead of handing you a verdict. Most of what it finds in academic prose is fixable by hand in a few minutes, and fixing it by hand keeps your terminology, your citations and your argument exactly where you put them. The free word counter is useful for cutting a chapter into runs of the right size.
On the policy question, briefly. Rewriting a draft you were permitted to generate is ordinary editing. Where your programme does not permit AI-generated text, that rule covers the draft however it is edited afterwards. Doctoral regulations vary between institutions and often between faculties, so read the one that binds your submission.
The drafting record captures the shape of a writing session: when you worked, how much you cut and rewrote, and whether large blocks arrived in a single paste. It signs that record and commits it to your exact text with a hash, so anyone can check the signature and the match on the verify page. It stores lengths and timings, never your content, and it is free with no daily limit.
It records forward, so start it on the chapter you are writing next rather than the one you finished last spring.
Pair it with the evidence you already hold. Version history in Word, Google Docs, Overleaf or git. Dated drafts emailed to your supervisor. Meeting notes, marked-up PDFs, raw data files, the notebook you took to the archive. Put those in date order, then add the signed record for the chapters you write from here, so you can answer a question with specifics rather than a denial.
Run it a chapter at a time. A single reading across a whole thesis averages your flattest and most distinctive sections together, so it cannot tell you which chapter to look at. A check needs at least 65 words and costs one credit whatever its length, so a 2,000-word chapter and a 200-word abstract cost the same, and ten free credits a day covers ten checks. Rewriting is capped at 12,000 characters a run, about 2,000 words, so long chapters get rewritten in sections in any case.
It means the passage you checked carries the patterns detectors score. It is not a prediction of what another system will output. Different detectors are trained on different text and set their cut-offs differently, and nobody can tell you what any of them will say about your writing without running it. Treat a high reading as a reason to look hard at that passage.
No. Leave out the reference list, appendices, tables, block quotes and transcript extracts. None of it is your prose, and including it produces a reading about somebody else's sentences. Check your own paragraphs, chapter by chapter.
Methods chapters do that more than any other part of a thesis. Reproducible procedure is uniform, unambiguous and repetitive on purpose, which is the shape of predictable text. Check the chapters on either side of it. If the methods section is the only part reading as machine-written, that pattern is itself the explanation. Predictable prose is also what drove the 61% false-flag rate Liang et al. measured on TOEFL essays by non-native speakers across seven detectors.
The allowance is 10 credits a day per network, with no account, no login, no card and nothing to verify. A check is one credit whatever the length. A rewrite is one credit per 50 words. The tell registry, proofreader, word counter, readability checker, AI vocabulary checker, originality check and drafting record cost nothing and have no daily limit.
Rewriting a draft you were permitted to generate is ordinary editing. Where AI-generated text is not permitted on your programme, that rule covers the draft however it is edited afterwards. Doctoral regulations differ by institution and often by faculty, so read the one that applies to your submission and ask your supervisor where it is ambiguous.
Start with what you already have: version history from Word, Docs, Overleaf or git, dated drafts sent to your supervisor, meeting notes and data files. Then add the signed drafting record for the chapters you write from here. No record proves authorship by itself, but a dated one carrying timings and revision ratios gives you specifics to point at instead of a denial.