In this article
If you write in English as a second language, you have probably felt the new anxiety: you write something genuinely yourself, and a detector — or a teacher running one — calls it AI.
You are not imagining the unfairness. A Stanford study published in Patterns (Cell Press) tested seven commercial AI detectors and found they flagged 61% of TOEFL essays as AI-generated on average, even though the essays were entirely human-written. 89 of 91 of those real essays were flagged by at least one detector. Native-speaker essays were flagged far less often. The maths of these tools — they look for “predictable” word choice — punishes exactly the careful, standard English that second-language writers are taught to produce.
Universities have noticed. The University of Waterloo discontinued its AI detector in 2025, citing “unreliability and bias toward students whose first language is not English.” Vanderbilt did the same. MIT published guidance literally titled “AI Detectors Don’t Work.”
So the useful question is no longer “can a machine tell if AI touched this?” It is “did you do the human work of authoring it?” — and that question has a clear, checkable answer.
The work of authoring is the same everywhere
We read the academic-integrity guidance of the UK (QAA, Russell Group, Oxford), the EU (ENAI), the US (ICAI, Stanford, Columbia, MLA), Canada (Toronto, UBC, McGill) and Australia (TEQSA’s federal law, Monash, Sydney). Underneath the different wording, they converge on the same short list of human acts. If you do these, you have authored the work — regardless of what a detector says.
Here is the checklist.
1. Write your own thesis
The single thing no tool can do for you is decide what the essay argues. Brainstorming with an AI is fine; outsourcing the position is not. Oxford’s guidance is blunt: AI “cannot replace human critical thinking or the development of scholarly, evidence-based arguments.”
2. Write most of it in your own words
Authorship lives in the typing. Translation help, grammar fixes and word-choice suggestions are accepted everywhere — the University of Sydney explicitly says you do not need to acknowledge tools that only correct spelling and grammar. What is not accepted is pasting AI-generated paragraphs and presenting them as your own.
3. Find and read real sources
Cite papers you have actually opened. This matters more than ever because AI tools invent citations — confident references to papers that do not exist. Columbia’s policy puts the responsibility on you: “you are responsible for any content that you publish that includes AI-generated material.” Reading your sources is both good scholarship and your defence.
4. Cite every source you used
Attribute every borrowed idea, even one you paraphrased into your own words. This is the oldest rule in the book, and it is the easiest to evidence.
5. Make the analysis and synthesis your own
Assessors are trained to look for synthesis — connecting sources to each other and evaluating them — not summary. As writing guides at UNC and USC put it, a literature review should be “a synthesis of your sources, not a series of summaries.” This is human work by definition.
6. Paraphrase in your own words and structure
This is the trap that catches honest ESL writers most often. Patchwriting — keeping the source’s sentence shape and swapping a few synonyms — still counts as plagiarism where the source is not made clear, even with a citation. UBC defines it as writing that “does not demonstrate that the writer has synthesized the ideas themselves.”
The reassuring part: Oxford explicitly recognises patchwriting as a normal developmental stage for writers who are not native English speakers — something to grow past with practice, not a character flaw. The fix is to rebuild the idea from scratch in your own structure, not to relabel the original.
7. Disclose the AI help you used
Transparency is now the norm. Oxford requires “a formal declaration of any permitted AI-use.” Monash asks you to note the number of iterations and where AI output was “altered, adopted, integrated or built upon.” And an AI tool is never an author: the European Network for Academic Integrity states “an AI tool cannot be listed as a co-author” because accountability “always rests on a person.” The MLA agrees.
The honest line for ESL writers
Notice what this checklist does not punish: using translation, fixing your grammar, looking up a better word, asking an AI to explain a concept. Those are language mechanics, and they are exactly how multilingual writers do their best work. Scholars have criticised the “deficit narrative” that treats a second-language writer’s tool use as inherently suspect.
The defensible line is voice and idea ownership, not language mechanics. Your thesis, your sources, your analysis, your disclosure — that is what makes the work yours. The rest is just help, the same way a dictionary or a writing centre is help.
Two kinds of proof — and why it beats a detector
Some of these acts can be verified automatically, and some you simply declare. That distinction is the whole game:
- Auto-verified (signed): how much of the document you typed versus pasted, how many real sources you cited, which AI tools you used. A tool like Diglot can record these as part of a cryptographically signed certificate — a third party can check the signature in their browser, no account needed.
- Author-declared: “I verified each source exists,” “the analysis is my own.” These are your statement, clearly labelled as such — exactly like the AI-use declaration universities already ask for.
That combination — signed process evidence plus an honest declaration — is what regulators are moving toward, and it is far stronger than a detector score that is wrong about ESL writers more often than it is right. A detector guesses about the final text. An authorship record shows the human work that produced it.
How Diglot turns the checklist into a cockpit
This is the feature we are building into Diglot. As you write — researching sources, drafting in your own words, paraphrasing, citing, disclosing AI — the editor keeps a running authorship checklist: a cockpit that auto-detects what it can from your real work, points you to whatever is missing (open the citations panel, add a counterargument, write your AI-use note), and never scolds you for using translation or grammar help.
When you are done, you can attach a clearly-labelled “authorship practices” summary to your Authorship Certificate — the auto-verified items backed by a signature, the declared items marked as your statement. If your work is ever questioned, you do not argue with a detector. You share a link.
You did the work. This is how you prove it.
Want the background on why AI detectors fail non-native writers — with the court cases and university policies in full? Read AI Detection Lawsuits 2026: What ESL Writers Need to Know and How to Prove Your Essay Is Human-Written.
Share
