Writing a Thesis in Your Second Language: AI Workflow for ESL Postgrads (2026)

Writing a thesis in your second language is not just harder — it is a categorically different experience. You are being asked to demonstrate original thinking, rigorous methodology, and scholarly authority in a language whose academic conventions you are still internalising, while simultaneously managing the same deadline pressure, supervisor expectations, and research complexity that your native-speaking peers face. The cognitive load is real, and the gap between what you can think and what you can write in English can feel paralysing.

The good news is that 2026’s AI tools, used correctly, close that gap in ways that were genuinely not possible five years ago. Not by writing your thesis for you — that would undermine the whole point — but by acting as a tireless language coach that knows academic English inside out. This guide gives you a concrete, step-by-step AI workflow tailored specifically to ESL postgrads: the challenges that matter most, the tools that address them, and exactly how to use them without compromising your intellectual ownership or falling foul of AI-detection systems.

The Real Challenges ESL Postgrads Face

Before laying out a workflow, it is worth being precise about what the obstacles actually are — because “my English isn’t good enough” is too vague to be actionable.

Academic register is a dialect you were never taught

Academic English is not just formal English. It is a specific dialect with its own vocabulary choices, syntactic preferences, and rhetorical conventions that even fluent conversational speakers of English may never have been taught explicitly. Phrases like “this thesis contends that,” “the evidence is consistent with,” or “the findings suggest a positive correlation” follow conventions that feel natural to a British or American doctoral student who has spent years reading journal articles — but are opaque patterns for someone who learned English in a classroom context aimed at everyday communication.

Research published in Cogent Education in 2025 confirmed that postgraduate students from Chinese-speaking backgrounds — one of the largest groups of international postgrads at UK, Australian, and Canadian universities — identify academic register as their single greatest writing difficulty, ahead of grammar, vocabulary range, or time management. The challenge is not that these students cannot write; it is that they are writing in a genre whose rules were never explicitly taught to them.

L1 interference in syntax and phrasing

When you write in your second language, your first language (L1) constantly shapes your sentence structure, phrasing, and even your argument-building logic. A sentence that makes perfect sense translated from Mandarin, Spanish, Arabic, or Portuguese can read as ambiguous or awkward in English — not because the idea is wrong, but because the grammatical structure was imported from a different linguistic system. Supervisors often describe this as a “logic gap” when it is actually a syntax gap.

Hedging under-use and over-use

Many ESL writers either hedge too aggressively — qualifying every statement to the point of meaninglessness — or not enough, making claims that strike examiners as overconfident. Getting the hedging calibration right requires absorbing hundreds of hours of academic reading, and even then it can feel arbitrary. This is one of the areas where AI can provide immediate, reliable improvement.

Idiom blindspots

Academic English is full of idiomatic collocations that fluent speakers use automatically but that do not translate: “this remains an open question,” “the literature is silent on,” “there is growing consensus.” Getting these wrong does not just sound odd — it can make examiners doubt your familiarity with the field.

The AI-Detection False Positive Problem

This issue deserves its own section because it represents a new and genuinely unfair burden placed disproportionately on ESL students.

A landmark 2023 study by Weixin Liang, Mert Yuksekgonul, and colleagues at Stanford University — published on arXiv and widely cited since — found that GPT detectors consistently misclassify non-native English writing as AI-generated, while accurately identifying native-speaker writing. The researchers tested TOEFL essays written by human non-native English speakers and found that multiple commercial detection tools flagged the majority of these genuine human essays as AI-produced. Their paper, titled “GPT detectors are biased against non-native English writers”, demonstrated that this is a systemic flaw in how these tools were trained, not an edge case.

Why does this happen? Most AI detection models were trained predominantly on texts written by native English speakers. Non-native writing — which tends to use a more formulaic sentence structure, a narrower range of vocabulary, and phrasing that reflects L1 patterns — can statistically resemble the output of language models that were themselves trained to produce fluid, “averaged” English. The irony is brutal: the more cautiously and correctly an ESL student writes, keeping to safe, well-attested academic phrasing, the more their writing may resemble AI output to a detector.

Research aggregated by Copyleaks in 2024 found that AI detection accuracy varies substantially for non-native English writers, with false positive rates high enough to cause real academic harm. An academic study reported through Hastewire found that over 60% of TOEFL essays from genuine human writers were flagged by some detectors as AI-generated.

For ESL postgrads, this creates a genuine injustice: your authentic hard work can be treated as cheating simply because of how your English sounds. The practical response is twofold:

• Use AI as an editor, not a ghostwriter — so your writing retains your idiosyncratic sentence construction, your specific examples, and your argument structure, all of which detectors struggle to flag.
• Run your work through a targeted false-positive review before submission so you can identify and address passages that might trigger a flag, and document your process if challenged.

Understanding this problem should reduce anxiety. It is not your writing that is the problem — it is a flawed detection system. And the right AI workflow actually lowers your false-positive risk by helping your writing be more distinctively yours.

The 5-Stage AI Workflow

This workflow is designed for a thesis chapter or section (3,000–8,000 words). Adapt the timings for your word count.

Stage 1: Draft in your own words — any register

The single biggest mistake ESL students make is trying to write in perfect academic English from the first draft. This produces paralysis. The first draft is about getting your argument, your evidence, and your reasoning onto the page. Write in whatever level of English comes naturally — including fragments, imprecise phrasing, or even a sentence in your L1 that you will translate later. The intellectual work is yours; the language refinement comes next.

Aim to write your argument section-by-section, matching your outline. Don’t stop to look up vocabulary or fix phrasing. A rough draft with good ideas is more useful than a blank page with perfect sentence structure.

Stage 2: AI structural review

Once you have a draft, use an AI editor — Tesify’s AI Editor is built specifically for academic thesis content — to review structure and coherence. Ask it: Does this paragraph have a clear topic sentence? Does the argument flow logically from evidence to interpretation? Is the transition between paragraphs clear? If you find yourself stuck in repeated structural edits, the guide on how AI fixes thesis revision loops has a concrete three-phase workflow for breaking that cycle.

At this stage, do not ask the AI to rewrite your text. Ask it to identify weaknesses. Read those suggestions, then rewrite the passage yourself. This keeps your authorial voice intact.

Stage 3: Register and hedging lift

This is the stage where AI delivers the most value for ESL writers. Paste a paragraph into Tesify’s AI Editor and ask specifically for:

• Academic register: Replace colloquial phrases with their scholarly equivalents.
• Hedging: Flag any claims that read as overconfident and suggest hedged alternatives.
• Collocation: Identify unnatural word combinations that suggest L1 interference.

Review every suggestion critically. The AI is offering options, not commands. If its suggested phrasing changes your meaning — even subtly — reject it and find your own reformulation. Protecting your intended meaning is more important than any stylistic improvement.

Stage 4: Plagiarism check

Once your chapter is in good shape, run it through a plagiarism checker before sending to your supervisor. ESL writers sometimes accidentally paraphrase too closely from sources they have been reading intensively — not from intent to plagiarise, but because the source’s phrasing has become internalised. A Tesify plagiarism check catches this before it becomes a problem. It also gives you documented evidence of originality, which is useful if you are ever challenged on AI-detection grounds.

Stage 5: Bibliography and final proofread

Generate your reference list automatically using Tesify’s Auto Bibliography. Bibliography formatting is an area where non-native English anxiety is often disproportionate — the rules are fiddly, the style guides dense, and the errors easy to make. Automation removes this entirely as a stress point.

Then do a final proofreading pass for surface errors. At this point, you are looking for typos, missing articles (a perennial challenge for speakers of article-free languages like Mandarin, Russian, or Arabic), and punctuation. Tesify’s AI proofreading tool can catch the systematic patterns that native English proofreaders might not even notice as L1-influenced errors.

Mastering Academic Register with AI

Academic register is not a single standard — it varies by discipline, by country, and to some degree by institution. A literature review in a humanities thesis sounds different from one in a quantitative social science thesis, which sounds different again from a life sciences methods section. ESL students are navigating this variation at the same time they are navigating the language itself.

The most reliable free resource for academic phrasing remains the University of Manchester Academic Phrasebank, created by Dr John Morley. It was designed explicitly for non-native academic writers and provides field-tested phrases organised by function: introducing a study, referring to sources, describing methods, discussing results, hedging claims. Use it alongside your AI editor — the Phrasebank gives you human-validated academic templates; the AI editor helps you adapt them naturally to your specific argument.

Common register problems AI can help fix include:

An AI editor does not just find these patterns — it identifies the specific instances in your text and offers contextually appropriate alternatives. This is far more efficient than learning register rules abstractly and far more targeted than asking a human proofreader who may not know which errors are systematic to your L1.

For broader strategies on lifting your writing quality across all chapters, the guide on how to improve thesis writing quality with AI covers this in depth.

Hedging Language: The Skill Detectors Misread

Hedging deserves special attention because it sits at the intersection of two problems ESL writers face: it is both a marker of academic sophistication that ESL writers often under-use, and a linguistic feature that AI detectors sometimes associate with AI-generated text.

In academic writing, hedging is not timidity — it is precision. When you write “this study suggests” rather than “this study proves,” you are making an epistemically accurate claim about what your evidence can actually support. Examiners mark hedging correctly as evidence of critical awareness. Hedging incorrectly — or not hedging at all — signals a lack of understanding of evidence and argument.

The key hedging categories, with examples:

Modal verbs

may, might, could, would, should — used to indicate probability or possibility rather than certainty. “These results may indicate a causal relationship, though longitudinal data would be required to confirm this.”

Epistemic adverbs

possibly, perhaps, apparently, seemingly, presumably, arguably — used to indicate the writer’s degree of confidence. “This pattern is arguably attributable to the sampling methodology.”

Lexical verbs of stance

suggest, indicate, appear, seem, tend, appear to — used instead of assertive verbs like “prove” or “show.” “The data suggest a moderate positive correlation.”

Approximators

approximately, around, roughly, in the region of, to some extent — used to signal that a number or claim is not exact. “The effect was present in approximately two-thirds of cases.”

When using Tesify’s AI Editor for hedging review, ask it to flag any unhedged causal or evidential claims — sentences that assert certainty without modal or epistemic qualification. Then review each suggestion against your actual evidence. If your data really do prove something unambiguously, a strong assertion may be appropriate. But for most empirical thesis claims, hedging is the mark of intellectual honesty.

Integrity, Originality, and What Your University Actually Allows

A genuine anxiety for ESL postgrads is: am I allowed to use these tools? The answer, for almost all UK, Australian, Canadian, and Irish universities in 2026, is: it depends on how you use them.

The near-universal distinction institutions draw is between AI as an editing tool (broadly permitted, just like human proofreading or a language editing service) and AI as a content generation tool (broadly prohibited, equivalent to ghostwriting). Using Tesify’s AI Editor to improve the clarity, register, and hedging of text that you wrote is precisely equivalent to asking a highly trained academic English editor to review your draft — a service that universities have always permitted and that many postgraduate programmes actively encourage international students to use.

What is not permitted, universally, is submitting text that an AI generated in place of your own thinking. This guide has been explicit throughout: the workflow starts with your draft, your argument, your evidence. AI edits; it does not author.

Practical steps to document your process:

1. Keep drafts showing the progression of your text before and after AI editing. Version control in a folder is sufficient.
2. If your institution asks for an AI use declaration, describe your use accurately: “AI-assisted language editing to improve academic register and clarity.”
3. Run a plagiarism check and keep the results. This gives you a paper trail showing originality.
4. Be prepared to discuss your arguments verbally with your supervisor. If you understand and can defend everything in your thesis, your intellectual ownership is clear.

For ESL writers who are also worried about tools that review non-native writing differently, there is a specific reassurance: the best AI tools for non-native English thesis writers are now purpose-built to understand that non-native English is not inferior English — it is different English, with different patterns that do not warrant a plagiarism or AI flag.

Which Tesify Features ESL Writers Use Most

Tesify was built as a thesis-specific platform, which matters because generic writing AI tools — designed for marketing copy, emails, or creative fiction — do not know what a literature review is supposed to sound like, do not know the difference between a Harvard and APA citation, and have no sense of the genre conventions that examiners evaluate. Here is how each feature maps onto the ESL workflow above:

AI Editor — the core tool

The Tesify AI Editor is trained on academic content and understands the conventions of thesis writing. For ESL postgrads, its key advantages are that it can distinguish between a register problem (wrong level of formality), a hedging problem (missing epistemic qualification), and a grammar problem (L1 interference in syntax) — and address each differently. It does not just “improve” your English; it explains the academic rationale for its suggestions, which helps you internalise the conventions for future writing.

Tesify Write — your thesis scaffold

For ESL students who struggle most with beginning a section, Tesify’s writing assistant helps you build an outline and a section framework that matches your research design. Working from a scaffold that uses correct academic genre conventions reduces the blank-page anxiety that is particularly acute when you are writing in your second language. You still populate the scaffold with your own analysis — but the structural template ensures you are not missing expected components (research gap, methodology justification, limitation acknowledgment) that examiners look for.

Plagiarism Checker — your safety net

As noted above, inadvertent over-paraphrasing from sources you have immersed yourself in is a real risk for ESL writers. The Tesify Plagiarism Checker catches this before submission. It is also the cleanest answer to a supervisor or examiner who queries whether AI was used to generate a passage: a clean plagiarism report combined with a documented editing workflow is strong evidence of originality.

Auto Bibliography — eliminate one stress entirely

For ESL postgrads already managing the cognitive load of writing in a second language, bibliography formatting should not occupy a single additional hour of mental energy. Tesify Auto Bibliography handles APA, Harvard, Vancouver, Chicago, and other common styles automatically. Enter your sources, select your style, export. It is a complete solution to a problem that has no intellectual content and wastes disproportionate time.

Frequently Asked Questions