In this article
The perfection trap
Making your English writing sound natural — not robotic, not textbook-stiff, not like AI wrote it — is the new challenge for ESL writers. For years, the goal was simple: eliminate every grammar mistake. Study textbooks, memorize phrasal verbs, polish prose until it is flawless.
That instinct served you well — until AI detectors arrived. Now, the cleaner your English, the more it looks like a machine wrote it.
AI detectors measure perplexity — how predictable each word choice is — and burstiness — how much sentence length and structure vary. Large language models produce low-perplexity, low-burstiness text by design. They pick the most probable next word and generate uniform output. Non-native English writers, especially those still building vocabulary, naturally produce the same kind of prose: safe word choices, standard grammar, consistent sentence length.
A Stanford study tested this directly. Researchers ran 91 human-written ESL essays through seven commercial AI detectors. Over 61% were incorrectly flagged as AI-generated. The detectors were not finding AI. They were punishing linguistic predictability.
The uncomfortable result: the harder you worked on your English, the more suspicious you look to a detection algorithm.
Grammatically correct is not the same as naturally fluent
Consider two ways to say the same thing:
- "I am writing to inform you that the report has been completed in accordance with your instructions."
- "Just wanted to let you know the report is ready — I followed the notes you sent over."
Both are correct. But the first version — the one most ESL courses teach — is a significant red flag for AI detectors. It is formal, predictable, and structurally uniform. The second version sounds like a colleague writing a quick update. It has contractions, a dash, an informal register, and varying rhythm.
Native English speakers switch between registers instinctively. They write differently in a Slack message, an email to a manager, a grant proposal, and a text to a friend. That variation creates high burstiness — exactly what detectors look for as a signal of human authorship. The Vanderbilt University acknowledged this problem when it disabled its AI detector, citing that the tools "have been found to be more likely to label text written by non-native English speakers as AI-written."
Non-native speakers often default to one register: formal academic. It is the safest zone. But safety is what gets you flagged.
Common phrases that sound robotic (and what to use instead)
Many ESL writers fall into what linguists call the "uncanny valley" — phrases that are technically correct but socially off. Here are patterns that make human readers (and AI detectors) pause:
"Kindly" in everyday emails
Textbook: "Kindly find attached the requested documents."
Natural: "I have attached the documents you asked for."
"Kindly" is formal British English from a different era. In contemporary American and international business English, it reads as stiff or automated.
Passive voice everywhere
Textbook: "The meeting was attended by all team members."
Natural: "The whole team was at the meeting."
Passive voice is not wrong — academic writing needs it. But when every sentence follows the "X was done by Y" pattern, the writing loses its rhythm and sounds like it was generated by a template.
Formal transitions in casual contexts
Textbook: "Moreover, it should be noted that the deadline has been extended. Furthermore, the scope has been revised."
Natural: "Also — the deadline got pushed back and the scope changed a bit."
"Moreover" and "Furthermore" belong in research papers. Using them in a Slack message or project update is the linguistic equivalent of wearing a suit to a barbecue.
Hedging with too many qualifiers
Textbook: "It would appear that there might potentially be an issue with the current approach."
Natural: "I think there is an issue with the current approach."
Hedging is a habit from ESL environments where directness feels risky. The problem: stacking qualifiers makes every sentence the same length and tone — low burstiness, high detector suspicion.
Seven strategies for writing naturally
1. Vary your sentence length deliberately
This is the single most effective change you can make. Read your draft out loud. If every sentence is 15-20 words, break some up. Make a few short. Let one run longer with a subordinate clause. The rhythm matters more than any individual word choice.
Short sentences create emphasis. Longer sentences carry complex ideas forward with more nuance and detail than a series of clipped fragments can.
Mix them.
2. Use active verbs
Active voice is not always better than passive. But defaulting to active voice forces you to name who is doing what, and that alone makes writing more specific and less generic.
Before: "The analysis was conducted and the results were compiled."
After: "We ran the analysis and compiled the results."
"We" makes it human. "Was conducted" makes it sound like a procedure manual.
3. Match the register to the context
Before you write, ask: who will read this, and where? An email to a colleague and a paragraph in a thesis need different language — even if the content is the same. Native speakers adjust register without thinking about it. For non-native speakers, it helps to decide consciously before drafting.
- Casual (Slack, internal chat): contractions, fragments, dashes.
- Professional (client email, report): clear structure, moderate formality, no slang.
- Academic (paper, thesis): formal transitions, citations, hedging where appropriate.
If you write everything at the academic level, everything sounds the same. And "everything sounds the same" is exactly what AI detectors flag.
4. Start sentences differently
A common ESL pattern: beginning every sentence with the subject. "The project is on track. The team has completed phase one. The next milestone is scheduled for June." This is correct but monotonous.
Try starting with:
- A time marker: "By Friday, we should have the first draft."
- A condition: "If the vendor confirms, we can move to phase two."
- A transition: "That said, the timeline depends on the review cycle."
- An aside: "Honestly, the hardest part was the data cleanup."
5. Use contractions where appropriate
In academic writing, avoid them. Everywhere else, contractions are normal. "I will" becomes "I'll." "It is" becomes "It's." "We have" becomes "We've."
Contractions do two things: they make the text feel conversational, and they break the mechanical rhythm that detectors associate with generated output. If your email has zero contractions, it reads as stiff — even if every word is correct.
6. Leave some rough edges
This sounds counterintuitive. You spent years learning to write cleanly. But natural writing is not perfectly smooth. It has asides, mid-sentence corrections, emphasis through repetition, and occasional fragments.
You do not need to introduce errors. You need to stop over-polishing. If a sentence communicates its point clearly, move on. Not every transition needs a connector. Not every list needs three items.
7. Draft in your native language first
If you think more clearly in Korean, Russian, Arabic, or Mandarin — start there. Capture the idea, the argument structure, the emotional tone. Then translate and adapt into English.
This is not cheating. It is the most natural workflow for a multilingual mind. The alternative — forcing yourself to think in English from word one — often produces the stiff, predictable prose that gets flagged. Starting in your native language produces writing that has your voice, your reasoning structure, your emphasis patterns. Those carry through translation in ways that pure English drafting often misses.
The flagxiety problem
There is a real psychological cost to this situation. Writers who have been falsely flagged — or who know someone who has — start second-guessing every sentence. Should I use a different word? Is this too clean? Am I going to get accused of cheating?
We call this flagxiety — the fear that your legitimate work will be misidentified as AI-generated. It is especially acute for ESL writers because the bias is structural: courts and universities are now recognizing that AI detectors systematically disadvantage non-native English speakers.
The answer is not to write worse English. The answer is to write naturally — with variation, register awareness, and your own voice — and to keep evidence of your writing process in case anyone questions it. If you are a student, read the full step-by-step guide to proving your essay is human-written. If you are a freelancer dealing with client accusations, see what to do if a client flags your writing as AI.
How Diglot helps you write naturally
Diglot is built specifically for the bilingual writing workflow. Instead of just checking grammar, it helps you move from your native language to natural-sounding English — preserving your voice, not replacing it.
- Draft in your language. Capture ideas in the language you think in.
- Translate and refine. Use the bilingual workspace to bridge the gap between your native draft and polished English.
- Adjust register. The AI writing assistant helps adapt tone — making academic prose more conversational, or casual notes more professional — without flattening your voice into a template.
- Prove authorship. The Authorship Certificate documents your entire writing process, from the first native-language draft to the final English version. If anyone questions whether you wrote it, you have cryptographic proof.
Write in your language. Publish in English. Sound like yourself — not like a textbook, and not like a machine.
Share
