Is Grammarly allowed at university?

Basic spelling and grammar correction is permitted essentially everywhere and has been for as long as spellcheck has existed. Grammarly's rewriting and tone features sit in a greyer tier: usually acceptable when applied to sentences you wrote yourself, but check whether your policy distinguishes editing from generation, because many were written before that distinction mattered.

Can I use AI to fix my English if it is my second language?

Usually yes. Language support has always been permitted and is often provided by institutions through writing centres and proofreaders. A growing number of policies now carve it out for non-native speakers by name. Confirm whether yours does, and whether a declaration is required, because using a permitted tool without a required disclosure is still a breach.

Is using a translation tool cheating?

It depends entirely on what you translate. Translating your own writing from your first language is usually treated as language support and is often explicitly permitted. Translating a source so you can read it is ordinary research. Translating generated content into English, or generating in English and translating out, is substitution with extra steps.

Do I have to disclose that I used a grammar checker?

Rarely, but where a policy requires a declaration you must comply. This is where most people actually fall foul: the use itself was permitted, but the required disclosure was missing. Check the wording of your policy, and if it is ambiguous, ask in writing before you submit and keep the reply.

Is ChatGPT allowed for homework?

It depends on how you use it and what your institution permits. Asking it to explain a concept you then write about yourself is research, the same as reading a textbook. Asking it to write your introduction is substitution. The test is whether the ideas and argument came from you, and whether you could defend every claim without the tool in front of you.

What happens if I break my university's AI policy?

Consequences range from a warning or a capped mark to failing the module, and in serious or repeated cases a mark on your record. The single largest aggravating factor is dishonesty during the process: institutions treat a denial that later unravels far more harshly than the original use, so precise disclosure is both the honest and the safer route.

Can I ask my tutor whether a tool is allowed?

Yes, and you should when the policy is ambiguous. Asking demonstrates good faith rather than raising suspicion. Do it in writing and keep the reply — an email from your convenor saying it is fine is worth more than any argument you could construct afterwards.

Which AI Tools Count as Language Support (And Which Don't)

Second-language writers have always used language help, and most policies allow it. A tier-by-tier guide to where the line falls and how to stay safe.

If English is not your first language and you are writing in it for assessment, you are in an awkward position that nobody designed and few policies address well.

Language support has always been permitted and often provided — writing centres, proofreaders, dictionaries, style guides. AI tools do the same job faster and better, which is why uptake among second-language writers is high and entirely reasonable. But the same tools sit on a continuum that runs into content generation, and the line between the two is rarely drawn clearly in the documents you are held to.

This is an attempt to draw it usefully.

The continuum

Thinking in tiers is more useful than thinking about which brand names are banned, because tools move between tiers depending on how you use them. The same product can be tier one in one session and tier four in the next.

Tier 1 — Mechanical correction. Spelling, punctuation, subject-verb agreement, article errors. Universally permitted, never disclosed, and part of writing since spellcheck existed. Nobody is coming for you over this.

Tier 2 — Expression of your own sentences. You wrote a sentence, it is awkward, a tool makes it read naturally. Your idea, your structure, your evidence; the machine adjusted how it lands. This is where most legitimate second-language use sits, and most policies that address the question at all permit it.

Tier 3 — Structure. Asking for an outline, or having a tool reorganise material you supplied. Genuinely grey. If your assessment includes whether you can construct an argument, having the structure supplied is closer to substitution than it looks — even though the content is yours.

Tier 4 — Content. The tool produced the argument, the examples, or the evidence, and you edited. This is not language support in any sense, whatever language you speak. It substitutes the thinking the assessment is measuring.

Tier 5 — Translation of generated content. Generating in English and presenting it, or generating in your first language and translating. Tier 4 with extra steps.

The useful test is not did I use a tool. It is did the ideas and the argument come from me, and could I defend every claim without the tool in front of me? If yes, you are in tiers 1–3. If no, no amount of language-support framing changes what happened.

Where common tools usually fall

With the caveat that all of them can be used at several tiers.

Spellcheckers and grammar checkers (Word, Docs, Grammarly's basic corrections) — tier 1. Safe everywhere.

Grammarly's rewriting and tone features, QuillBot's paraphraser, our own Text Humanizer — tier 2 when applied to sentences you wrote. These change how your writing reads without adding anything to it. They become tier 4 the moment you feed them text you did not write.

Translation tools (DeepL, Google Translate) — depends entirely on what you translate. Translating your own writing from your first language is usually tier 2 and often explicitly permitted; check, because some policies treat translation as generation. Translating a source to read it is tier 1 research. Translating generated content is tier 5.

General chat models (ChatGPT, Claude, Gemini) — anywhere from tier 1 to tier 5 in the same conversation. Asking "is this sentence grammatical?" is tier 1. Asking "explain this concept so I can write about it" is legitimate research, the same as reading a textbook. Asking "write my introduction" is tier 4.

Reference and dictionary tools — tier 1 always. Looking up a word has never been controversial.

Read your actual policy

The tiers above are reasoning. What you are actually held to is a document, and most people accused have never opened it.

Search your institution's site for "academic integrity," "academic misconduct," or "generative AI policy." Then look for four specific things:

Does it distinguish editing from generation? Many policies were written before this distinction mattered and are ambiguous about grammar tools. Ambiguity generally works in your favour, but it is better to know.

Does it name language support explicitly? A growing number of policies carve out language assistance for non-native speakers by name. If yours does, that clause is the most useful sentence in the document — note where it is.

Is disclosure required, and in what form? This is where most people actually fall foul. Using a permitted tool without a required declaration is a policy breach even when the use itself was fine.

Does it say a detector score alone is sufficient? Many now state explicitly that it is not. Worth knowing before you need it.

If the policy genuinely does not cover your situation, ask in writing before you submit and keep the reply. An email from your tutor saying "yes, that's fine" is worth more than any argument afterwards, and asking demonstrates good faith rather than raising suspicion.

The bind you are in

Worth naming, because it is real and rarely acknowledged.

Second-language writers have both the strongest legitimate reason to use these tools and the highest risk of being wrongly accused — and the two are causally connected.

Detectors estimate how statistically predictable text is. Writing in a second language is predictable for innocent reasons: a smaller active vocabulary means safer, higher-frequency word choices, and formal instruction teaches exactly the connective structures models overproduce. Language tools then nudge your text further toward the statistical centre, because that is their function.

So doing everything right — writing carefully, choosing words you are confident about, running a grammar check — produces the profile detectors flag. Research published in Patterns found detectors misclassified essays by non-native English writers at dramatically higher rates than native writers', and that the effect largely disappeared when the same essays were rewritten with richer vocabulary. The full mechanism is in why AI detectors fail non-native English speakers.

This is not your fault and it is not a reason to stop using legitimate support.

Using permitted tools safely

Practical habits that cost nothing.

Draft in Google Docs. The automatic version history shows composition happening over time, and it is the single most persuasive artefact if you are ever questioned. It also captures your revisions in your own words before any tool touched them.

Keep your first-language material. Outlines, notes, and searches in your first language are close to unforgeable evidence of authentic authorship. Generated English does not come with a Vietnamese outline behind it.

Log what you used. A short note: which tool, for what, on which sections. Two minutes, and it means you never have to reconstruct it under pressure.

Apply tools to your own sentences, one at a time. Rewriting a paragraph you wrote is tier 2. Pasting in three thousand words and accepting whatever comes back makes it much harder to say where your writing ends and the tool's begins — including to yourself.

Read every change. Rewriting tools sometimes swap a precise term for a vaguer one, or flip a hedged claim to an absolute one. On technical work that introduces errors you will be responsible for.

Do not run flagged work through a rewriting tool. If you have already been accused, changing the text afterwards is indefensible. Ours included.

Building the skill underneath

The tools are a bridge, not a destination, and the thing they are bridging is worth building directly.

What raises perplexity in your writing — and, not coincidentally, makes it better — is vocabulary range and rhythmic variation. Reading widely in English outside your field does more for this than any exercise. So does deliberately varying sentence length rather than writing uniform, careful, medium-length sentences. Our guides to sentence rhythm and finding your writing voice cover the mechanics.

This is not writing worse to fool a detector, which we would not recommend and which damages your writing. It is developing the range that the detector was a bad proxy for in the first place. The score improving is a side effect of the writing improving.

If you are accused anyway

Preserve your version history before editing anything, find the policy and its appeal deadline, and explain the mechanism rather than just asserting innocence — an assessor who understands why second-language writing scores high usually stops treating the number as evidence.

There is adaptable wording for exactly this situation in appeal letter templates, and the full first-24-hours sequence in falsely accused of using AI.