Whether editing tools constitute unauthorised AI

Short answer

It depends entirely on the specific tool, how you use it, and what your institution’s rules and the assignment brief actually say. There is no single rule across all universities and colleges in England and Wales. Some uses of Grammarly, translation software or paraphrasing tools are perfectly acceptable, while other uses of the very same categories of tool can be treated as unauthorised AI and therefore academic misconduct. The key questions are whether the tool substantively generates or rewrites content, and whether the particular use is permitted, restricted, prohibited or required for that assessment.

Why there is no universal answer

Academic misconduct rules are set by each institution, not by a single national law. There is no statute in England and Wales that defines “unauthorised AI” in academic work. Instead, each university or college has its own academic misconduct regulations and increasingly its own generative AI policy. Many providers have adopted a framework describing AI use in an assessment as prohibited, restricted, permitted or required. That means the same tool can be allowed for one assessment and banned for another. Because of this, the decisive documents are your institution’s academic integrity policy, its AI guidance, and the specific instructions on the assignment brief. If the brief is silent, the safest course is to ask the module tutor before you rely on any tool.

The distinction that actually matters

The line most institutions draw is between tools that support the mechanics of writing you have done yourself and tools that generate or substantively rewrite the content and ideas for you.

Tools that check spelling, punctuation and basic grammar, or that flag readability, are usually treated as legitimate assistive technology, in the same way that a spell-checker has been accepted for years. This is especially so where they are recommended for accessibility or for students with specific learning differences.

Tools, or features within tools, that write sentences for you, predict and complete sentences, generate new text, or substantially rewrite your text into different wording, are the ones commonly treated as generative AI. Where their use is not permitted for the assessment, that use is unauthorised AI and can be academic misconduct.

The important practical point about Grammarly is that it is not one single thing. Its basic spelling and grammar checking is generally treated very differently from its generative AI writing assistant, which can rewrite, expand or compose text. So you cannot say simply that “Grammarly is allowed” or “Grammarly is banned” — it turns on which features you use and what your institution permits.

Grammarly

Basic Grammarly correction of spelling, punctuation and grammar in text you wrote yourself is usually acceptable, and some institutions positively support it as assistive software. However, Grammarly now includes generative AI features that will rewrite sentences, improve arguments, expand points or draft text. University guidance in this area increasingly treats the generative rewriting features as not permitted unless expressly allowed, precisely because they change or produce content and make it harder for a marker to judge your own understanding. The concern institutions repeatedly express is authorship: if a tool has restructured your arguments or supplied wording, the marker cannot be sure the work reflects your own understanding.

Paraphrasing and rewriting tools

Dedicated paraphrasing or “spinner” tools, and generative rewriting assistants such as Wordtune or Paperpal, are the most likely of the three categories to be treated as unauthorised. Because they substantively rewrite your text, they raise clear authorship concerns, and they are also the type of tool most likely to be flagged by AI detection software. Using a paraphrasing tool to disguise material taken from a source, or to reword content so that it reads as your own, can be treated both as unauthorised AI use and as a form of plagiarism. If the aim or effect is to conceal the true source or authorship of the text, that is where the greatest risk lies.

Translation software

Using translation software raises a specific authorship question. Checking the meaning of a word or a short phrase is generally uncontroversial. Writing substantial passages in another language and then machine-translating them into English for submission is different, because it puts in question who actually produced the assessed English text and the analysis in it. Institutional policies commonly warn that translating large portions of text this way may be misconduct, and that using online translation to conceal the original source of a text is treated seriously. Many providers expect you to acknowledge or reference where content has been machine-translated. So occasional dictionary-style use is usually fine, but wholesale translation of your work, or translation used to hide a source, is high risk.

When ordinary use is not misconduct

It is worth keeping the risk in proportion. Using a spell-checker, checking a single word in a translator, or correcting your own grammar is a normal part of producing written work and is not, in itself, dishonest or generative AI use. Misconduct generally requires either that the tool has produced or substantively reworked the content, or that the tool has been used to disguise the source or authorship of the work, in circumstances where that use was not permitted. An innocent, minor and permitted use is very different from using a tool to generate or launder the substance of an assessment.

If you are already facing an allegation

If a university or college is alleging unauthorised AI use, the process is governed by that institution’s academic misconduct procedure, and you are normally entitled to fair procedure. That typically includes being told the specific allegation and the evidence relied on, a chance to respond, and often a viva or meeting where you can demonstrate your own understanding and authorship. AI detection tools are not conclusive proof on their own; they produce probabilistic indicators, and a well-run process should not treat a detector score as decisive. In one published case summary, a provider used a viva to test whether the student understood and had written the work, and the outcome turned on that rather than on a detector alone. Practically, you should keep and be ready to produce your drafting evidence, such as version history, saved drafts, notes, research materials and reference sources, because being able to show how the work developed is often the strongest defence to an authorship allegation.

Practical next steps

1. Read your institution’s academic integrity policy and its generative AI guidance, and check exactly what the assignment brief says about AI, editing tools and translation.

2. Identify which specific features you used, distinguishing basic spelling and grammar checking from any generative rewriting, paraphrasing or translation of substantial text.

3. If the brief is silent or unclear, ask the module tutor or programme lead in writing before submission, and keep the reply.

4. Where a tool has been used in a permitted way but has touched the content, follow any requirement to acknowledge or reference that use.

5. Preserve drafts, version history and research notes so you can evidence your own authorship if questioned.

6. If you are accused, request the specific allegation and the evidence, follow the institution’s stated procedure, and use any meeting or viva as an opportunity to demonstrate your own understanding.

What would change the answer

The outcome depends heavily on facts I do not have. The decisive variables are the wording of your institution’s policy, whether the assessment classified AI as prohibited, restricted, permitted or required, which specific tool and which feature you used, whether it generated or substantively rewrote content, whether translation was of short phrases or large passages, and whether any use had the effect of concealing the true source or authorship. Different combinations of these facts can move the same activity from clearly acceptable to clearly misconduct.