Academic Writing Style in a Thesis: Tense, Voice, Hedging, and Why Detectors Flag It

The short answer

Academic style in a thesis means: present tense for what holds and past tense for what you did; passive voice only where the actor genuinely does not matter; verbs rather than stacked abstract nouns; claims hedged exactly as far as the evidence requires and no further; and filler words deleted. It does not mean long sentences and rare words.

There is also a second thing worth knowing, and almost nobody writing style guides says it: the conventions above make prose predictable, and statistical predictability is precisely what AI detectors measure. Well-executed academic writing is therefore among the writing most likely to be flagged. That is a real tension, it is not your fault, and the answer to it is not to write worse. There is a section on this below.

Which tense goes where?

Two levels, and the rule follows from them:

• Present tense for anything that holds. Definitions, theory, the state of research, the meaning of your findings, references to the document. “Braun and Clarke distinguish six phases of thematic analysis.” · “Chapter 3 sets out the methodology.”
• Past tense for anything you did. Recruitment, collection, analysis, the findings themselves. “Interviews were conducted between March and May 2026.” · “Six of the eight participants named coordination as the main obstacle.”

How they combine in one sentence:

“The interviews were analysed using structuring content analysis (past: what I did), a procedure that suitsmaterial in which categories derive partly from theory and partly from the data (present: what holds).”

So switching inside a paragraph is not sloppiness — it carries information about which level you are on. The one thing to keep consistent is the reporting convention for literature: present (Schmidt argues) is the norm in most fields, present perfect (research has shown) for accumulated bodies of work, past (Schmidt found) when you are describing one specific study as an event. Pick one and hold it.

Passive or active voice?

The passive is widely treated as the badge of academic writing. It is really a tool with a narrow job: it foregrounds the action and pushes the actor out of view. In the methods chapter that is exactly right.

• Works: “The recordings were transcribed in full and then coded.” Who did the transcribing is irrelevant to whether the study can be repeated.
• Fails: “It is assumed that the results can be considered transferable to comparable settings.” Assumed by whom? The sentence hides the responsibility that an academic claim is supposed to make visible.
• Better: “These results are transferable to comparable development teams.” Shorter, checkable, and open to challenge — which is a virtue here.

The construction to watch is the empty it is opening: it is argued, it is shown, it becomes clear. Three of those on a page and the text sounds as if it does not dare to say anything. Give the sentence a subject that actually acts: “Chapter 4 examines …”, “The interview data show …”

Hedging: precision, not timidity

Hedging means matching the strength of a claim to the strength of your evidence. It is not politeness or throat-clearing — it is accuracy, and a marker will read an unhedged overstatement as a methodological error rather than a stylistic one.

A rough scale, strongest first:

• shows, demonstrates — only for direct, unambiguous measurement.
• indicates, suggests— the normal register for a bachelor's or master's thesis.
• may be attributable to, one possible explanation is — for interpretation that reaches beyond the data.

The opposite failure is stacking: “It could possibly be argued that there may perhaps be some indication of a certain tendency.” Four hedges and no claim left standing. One per claim.

Better than an extra hedge is a stated scope. “In the four firms studied, …” is simultaneously more precise and more confident than “it may possibly tend to be the case that …”, because it tells the reader exactly how far the claim travels.

Nominalisation — the main cause of heavy sentences

Nominalisation is turning verbs into abstract nouns — -tion, -ment, -ance, -ity — and joining them with empty verbs like conduct, perform, undertake, carry out.

• Before: “The implementation of the collection of the data was undertaken through the utilisation of a semi-structured interview procedure.” — Four abstract nouns, one hollow verb, nineteen words.
• After: “Data were collected in semi-structured interviews.” — Seven words, identical content.

A second pair from a theory chapter: “For the clarification of the question of the comparability of the findings, a juxtaposition of the definitions is performed.” becomes “To clarify whether the findings are comparable, the definitions are compared.”

A search strategy for revision: search your document for of the. Two or three in one sentence is a reliable symptom of stacked nouns. Not every nominalisation must go — analysis, validity and generalisation are established terms — but where three sit in one sentence, at least one can go back to being a verb.

Filler words and the deletion list

Filler is any word whose deletion leaves the sentence unchanged. It costs characters you need under a word limit, and it makes evidenced claims sound uncertain.

The usual suspects: actually, basically, essentially, obviously, clearly, of course, really, very, quite, rather, somewhat, in a sense, it should be noted that, it is important to mention that, in the context of, at this point in time.

• Before: “It is important to note at this point that the results can basically be considered to be quite meaningful in a certain sense.” — Twenty-two words.
• After: “The results are meaningful for the cases examined.” — Eight, with the qualification now placed where it belongs.

Worn transitions deserve the same treatment: Furthermore, Moreover, In addition, It is worth mentioning that. They are not wrong, but when every paragraph opens with one they carry no information. A good transition says why the next paragraph follows: “That leaves open whether …”

One caution against over-applying this. A text stripped of every modulation reads mechanically. The aim is that each remaining word does something — not that the prose becomes as uniform as possible.

Precision beats formality

• One concept, one word. Alternating between participants, respondents, interviewees and subjects for the same group looks like varied writing and is an error: in academic prose, different words signal different things. Save the synonyms for essays.
• Resolve your references. This, that and which at the start of a sentence force the reader backwards. Add the noun: “This discrepancy …” rather than “This …”
• One idea per sentence. A sentence with two subordinate clauses and a parenthesis is usually two sentences.
• Be specific with quantities. Several interviews is not repeatable; eleven interviews of 45 to 70 minutes is.

Why good academic style is what detectors flag

Now the part that belongs in a style guide in 2026 and is usually missing from one.

AI text detectors do not recognise ideas, and they cannot tell where a sentence came from. What they measure is statistical predictability: given the words so far, how unsurprising is the next one, and how little does that surprise vary from sentence to sentence? Machine-generated text tends to sit in a narrow, highly probable band, because that is what the generating model optimises for.

Now look back at every recommendation on this page:

• Standardised phrasing. “This chapter sets out …”, “The findings indicate that …”, “Limitations of this study include …” — conventional openings are conventional because everyone uses them, which is another way of saying they are highly predictable.
• Consistent terminology. We just told you never to vary your words for the same concept. Lexical repetition lowers surprise directly.
• Hedging. A small set of qualifiers — suggests, indicates, may be attributable to — is used across the entire literature, in similar positions.
• Formal register with filler removed. Idiosyncrasy is exactly what filler, asides and personal voice supply, and academic revision removes them on purpose.
• Regular structure. Uniform paragraph shapes and predictable section patterns reduce variation at the document level too.

The conventions exist for good reasons: they make studies comparable, claims checkable and arguments quick to navigate for a reader who reads hundreds of these. But the effect of following them well is prose with low surprise and low variance — the same statistical region that detectors treat as evidence of generation.

This is not a hypothesis about your writing in particular; it follows from what the tools measure. It is also why the pattern shows up most sharply in the methodology chapter, which is the most formulaic prose in a thesis, and in literature reviews, which are built from reporting formulas. Those are the sections where careful students most often see a score they did not expect.

Two groups are hit hardest by this and neither did anything wrong. Writers working in a second language often lean on standard phrasing precisely because it is safe and correct, which lowers surprise further. And meticulous writers who revise thoroughly remove exactly the irregularity that would have distinguished them. The better you follow the guidance, the more conventional your prose becomes.

What to do with that, and what not to do

What not to do: do not write worse. Deliberately roughening your prose to move a classifier trades marks you are actually assessed on — clarity, precision, consistent terminology — against a number that nobody assesses. It also does not reliably work: there is no published mapping from a specific stylistic change to a specific score change, so you would be damaging your thesis on a guess. Any advice you find that promises otherwise is selling something.

What is worth doing:

• Know before someone else does. If your university runs a detector, seeing how your own methodology chapter reads to one is simply information you would rather have early than in a meeting.
• Ask for the error rate, always. A verdict without a false-positive rate is not a measurement. This is the question to put to any tool, including ours — how often is this verdict wrong about writing that is verifiably human?
• Keep your evidence. Version history, drafts, notes, reading annotations, interview recordings. A document with a visible history of becoming is the most straightforward answer to any question about its origin.
• Be able to explain your work. Why this method, why this source, why this sentence says what it says. That conversation is one that no classifier participates in.

If you want to see what your own chapter looks like to a detector, our AI check is free for the first 1,500 words and reports, next to every verdict, how often that verdict is wrong about verified human writing — the measurements and how they were produced are here. We publish the error rate because a verdict without one cannot be acted on responsibly, by you or by an examiner. If it flags your methods chapter, the most likely explanation is the one this section describes: a fact about the genre, not about you.

Revising in four passes

Style happens in revision, not in drafting. Watching tense and filler while composing produces slow writing that is not better. Four separate passes, one job each:

1. Structure. Does each paragraph carry exactly one idea, in the right place?
2. Sentences. Unpack nominalisations, split overlong sentences, replace empty it is constructions.
3. Words. Delete filler, unify terminology, cut stacked hedges to one.
4. Consistency. Tense by chapter, citation style, spellings, abbreviations, number formats.

Two checks that find a surprising amount: read a paragraph aloud — where you run out of breath, the sentence is too long. And have the text read to you by your operating system's screen reader; repetition and broken references are audible in a way they are not on the page.