How to humanize a dissertation chapter

Before and after

Five sentences of near-identical shape became four that argue. The studies are now in a relationship — first, reproduced, extended — rather than in a list, and the final clause states why the chapter cares. No citation was altered.

The literature review is where this goes wrong

Almost every weak dissertation chapter fails in the same place, and it is not the analysis. It is the literature review, which turns into an annotated bibliography with the annotations run together — study, finding, study, finding, and a closing sentence observing that the area has been studied.

The reason is structural rather than stylistic. Summarising one source at a time is a task you can do while only holding one source in your head. Arguing about them requires holding several at once and deciding what their relationship is, which is harder and is the actual work of a review.

A generated draft is very good at the first and cannot do the second, because the relationship between two studies is something you know from having read both. So the tell in a dissertation is not flat sentences. It is a section where every paragraph is individually competent and the section as a whole says nothing.

Fix the joins before the sentences

Read a section and mark every point where one study ends and the next begins. Those joins are where your argument is missing. At each one, write the sentence explaining why the second study follows the first — it reproduced it, it contradicted it, it moved the question somewhere new, it used a method the first could not.

Most of those sentences will be short, and that alone fixes the rhythm problem the surface-level advice is aimed at. Variation in sentence length falls out of a paragraph that is doing something, and no amount of synonym-swapping produces it in a paragraph that is not.

Then delete the summary sentences that survive. A paragraph opening with "several studies have examined" and closing with "these findings suggest the relationship is well established" has a beginning and an end wrapped around a middle that says the same thing. Both can go.

Length makes register drift the real risk

A chapter is too long to rewrite in one pass, so it gets done in pieces over days or weeks. That is fine for the argument and dangerous for the voice, because a section rewritten on Tuesday and a section rewritten a fortnight later will not sound the same.

Inconsistent register is the thing an experienced examiner notices, and unlike a detection percentage it is a legitimate observation about the text — they can point at the paragraph where the writing changes. It is also what makes a document read as assembled from several sources, which is the impression you least want in a viva.

The remedy is a final read of the whole chapter in one sitting, aloud if you can stand it, doing nothing but flattening the register. Not improving sentences. Just making the third section sound like the first.

The citation check is not optional

Rewriting prose around fixed citations is the specific hazard of academic work at this length, because the failure is silent. The reference stays correctly formatted and points at a real paper, and the sentence it is attached to has quietly changed what it claims.

The version that causes trouble is the hedged claim tightened into an assertive one. Your draft said a study reported an association; the rewrite says the study demonstrated a cause. The citation did not move. You have now misrepresented a source in a document that will be examined by someone who may well have read it.

So the last pass is mechanical and separate: go through every citation, open what it points at, and confirm the sentence around it still says what the source supports. It is slow, it is the least interesting hour of the process, and it is the one that protects you.

Formats with the same problem

Research papers — the shorter form of the same citation hazard, with the methods-section exception.

Grant proposals — where the gap sentence does the work a literature review does here.

Questions

Can I run a whole chapter through at once?

Not usefully. Beyond a few thousand words the useful unit is a section, because the edit you are making — fixing the relationship between two studies — is local, and a whole-chapter pass produces uniformity rather than argument. Work in sections and read the joins afterwards.

Will my supervisor be able to tell?

They will notice register drift and missing argument long before they think about software, because those are the things they are trained to read for. A chapter that summarises competently and argues nothing gets the same feedback whether a tool was involved or not.

What about the methods chapter?

Leave it largely alone. Methods sections are formulaic because the convention requires it, and rewriting one for variety makes it worse and harder to follow. Predictable prose is correct there, even though predictability is exactly what a detector reads as machine-like.

Do I have to declare this?

That depends on your institution and increasingly on your department, and doctoral regulations are often stricter than the general student policy. Check the rules that govern submission, not the ones on the university's public page, and where declaration is required, describe what the tool did rather than naming it and stopping.

What to watch for

• •Never let a rewrite touch a citation, a page number or a quoted passage. Check every one afterwards.
• •A tightened sentence can make a source claim more than it did. Reread the original before accepting a stronger phrasing.
• •Chapters rewritten separately drift apart in register. Read the joins between them last, in one sitting.
• •If your institution requires disclosure of AI assistance, disclose it. A rewrite does not change that obligation.

Other use cases