Policies differ: Oxford treats undisclosed AI material in assessed work as plagiarism, while Sydney has allowed AI by default in open assessments since 2025.
| Setting | Documented position | Source |
|---|---|---|
| USC | AI-created work represented as the student's own, "whether paraphrased or copied verbatim", is plagiarism; instructors set course rules | USC Office of Academic Integrity |
| Oxford | Plagiarism includes material generated wholly or partly by AI, unless AI use was authorised in advance | University of Oxford |
| Sydney | AI allowed by default in open assessments from Semester 1 2025; exams and in-semester tests stay AI-prohibited unless allowed | University of Sydney |
| Elsevier journals | AI use allowed with oversight; a disclosure statement is required; basic grammar and spelling checks need none | Elsevier |
| Springer Nature | Language polishing with an LLM is "permitted" low-risk use; disclosure enhances trust | Springer Nature |
| Grammarly | Disguising restricted AI use with a humanizer may be cheating or plagiarism | Grammarly AI humanizer page |
| QuillBot | Follow your institution's AI guidelines; passing off reworded ideas as your own is unethical | QuillBot AI Humanizer page |
Is using AI cheating?
It depends on who sets the rules for the work and what they say. There is no single standard across schools, and many universities leave the decision to each instructor.
The University of Southern California's Office of Academic Integrity asks instructors to state in the syllabus whether AI is encouraged, permitted for specific assignments or prohibited. The default matters when a syllabus says nothing, and defaults differ between institutions. Some universities now start from "allowed": the University of Sydney announced on 27 November 2024 that from Semester 1 2025 AI use would be allowed in assessments by default, except exams, in-semester tests or where teaching staff opt out, and that students must acknowledge how they used AI.
The practical rule: read the course policy for each assignment. If it is unclear, ask the instructor in writing before you start, and keep the reply.
My take: the policy that governs you is the narrowest one that applies. A university can allow AI by default and a single instructor can still ban it for one essay. The assignment brief beats the university headline every time.
How I read a course AI policy
When I go through a syllabus or assessment brief, I look for five things, in this order:
- Scope. Does the rule cover the whole course, one assignment or one stage, such as brainstorming but not drafting?
- Which tools. Some policies name generative chatbots only; others cover paraphrasers, translators and "AI writing assistants" in general. A humanizer falls into the second group.
- What counts as use. Outlining, rewording, grammar fixes and full drafting are often treated differently.
- Disclosure. Whether you must declare use, and in what form: a statement, an appendix, a citation or a form.
- Evidence. Whether you may be asked for drafts, notes or version history, and whether there is an oral follow-up.
If any of those five is missing, that is the question I'd send to the instructor before starting.
Is it plagiarism to use AI?
Several universities define it that way when AI output is presented as your own. The University of Oxford's plagiarism guidance defines plagiarism as presenting work or ideas from another source as your own without full acknowledgement, and says this covers "material generated wholly or in part through use of artificial intelligence (save when use of AI for assessment has received prior authorisation)". Oxford adds that AI may be used in assessments only with prior authorisation or as an agreed reasonable adjustment for disability.
USC's policy goes further on rewording: AI-created material represented as the student's work is plagiarism "whether paraphrased or copied verbatim or in near-verbatim form." That sentence is the direct answer to whether running AI output through a humanizer changes anything. Under a policy like USC's, it does not.
My take: a humanizer changes the surface of the words, not where the ideas came from. Integrity policies are written about where the ideas came from.
Citation guides take the same view from the other side. The MLA Style Center's guidance on citing generative AI says to cite an AI tool whenever you paraphrase, quote or incorporate its content, and to acknowledge functional uses such as editing your prose or translating words.
Is Grammarly cheating? Is QuillBot cheating?
Grammar and spelling checks are rarely treated as cheating. Generative features are a different question, and the tool makers say so themselves.
- Grammarly. Its AI humanizer page says that in professional settings using AI to draft or polish is "appropriate", but in academic settings or any context where AI use is restricted or must be disclosed, using a humanizer to disguise AI use "may be considered unethical and could constitute cheating or plagiarism." Grammarly also offers Authorship, which records whether text was typed, pasted or AI-assisted so a writer can show their process.
- QuillBot. Its AI Humanizer page tells users to follow their institution's or workplace's AI guidelines on citing and acknowledging AI help, and says "trying to pass off someone else's idea as your own, even if it has been reworded, is unethical."
- Journals draw the line at substance. Elsevier's generative AI policy for journals says basic grammar, spelling and punctuation checks need no declaration, but when an AI tool makes substantive changes to sentence structure or organization, that should be disclosed.
So a spell check is almost never the problem. A paraphraser or humanizer that rewrites whole sentences is the kind of use policies ask you to disclose or avoid. Feature differences are in the Grammarly review and the QuillBot review, and the humanizers themselves are covered in my Grammarly AI Humanizer review and QuillBot AI Humanizer review.
Allowed or not: common uses at a glance
This is how I would sort the uses people actually ask about. It is a reading of the policies quoted above, not a ruling on your course; your assignment brief decides.
| Use | Under "AI banned unless authorised" | Under "AI allowed with disclosure" | Under "AI allowed" |
|---|---|---|---|
| Spell and grammar check on your own writing | Usually fine | Usually fine, no declaration (Elsevier) | Fine |
| Rewording your own paragraph for flow | Ask first | Disclose | Fine |
| Smoothing phrasing as a non-native speaker | Ask first, or arrange an adjustment | Disclose | Fine |
| Rewriting a chatbot draft so it reads naturally | Breach | Disclose the drafting and the rewrite | Fine; check facts |
| Rewriting AI output so it is not flagged | Breach | Breach, because the point is to hide use | Pointless at best |
| Paraphrasing a published source | Cite the source | Cite the source | Cite the source |
| Emails, reports and marketing copy at work | Follow your employer's policy | Follow your employer's policy | Fine |
My take: the row that never works is the one where the goal is concealment. Every other row has an honest version.
Where humanizers fall in integrity policies
Policies sort into three broad groups. Your course may sit in any of them, and it can change between assignments.
| Policy type | Example | What it means for a humanizer |
|---|---|---|
| AI banned unless authorised | Oxford assessed work | Using AI output, humanized or not, without authorisation is plagiarism |
| AI allowed with disclosure | Elsevier journals; Sydney open assessments | Allowed if you disclose it as the policy requires; hiding it is the breach |
| AI allowed or encouraged | Instructor-permitted assignments; your own business writing | Editing for readability is fine; you remain responsible for accuracy |
Two facts make the "hide it" approach weak even where it might seem to work.
Detection vendors target humanizers directly. Turnitin announced AI bypasser detection on 27 August 2025, describing humanizers as tools that "allow students to disguise AI-generated text as human-written." The feature is trained and tested for English only.
Detectors also misfire on honest writing. Perkins and colleagues tested six detectors on 805 samples and reported in a 2024 study that accuracy was 39.5% on unmodified content and 17.4% when evasion techniques were applied; they concluded the tools cannot currently be recommended for deciding integrity cases. USC tells its instructors the same: detector evidence alone is insufficient to find a student responsible. My guide to AI detector false positives covers what to do if your own writing is flagged, and how AI detectors work explains why scores move.
None of that makes undisclosed AI use acceptable. A policy breach is a breach whether or not software catches it.
My take: I would not build a plan on either side of the detector question. Detectors are unreliable enough that they should never be the only evidence against you, and good enough that hiding AI use is a gamble with your record. Disclosure removes the gamble.
If your honest work is flagged
False positives are documented, and Liang and colleagues found in a 2023 study in Patterns that seven detectors flagged more than half of 91 TOEFL essays by non-native English speakers as AI. If a score lands on writing that is genuinely yours, or on AI use you were allowed and disclosed, this is the checklist I'd work through. It is about showing your process, not about changing the text.
| Step | What to do | Why it helps |
|---|---|---|
| 1. Stay calm and read the policy | Find the section on AI and on how concerns are raised | Many schools, USC included, say a detector score alone is not enough |
| 2. Gather your process | Drafts, notes, outlines, browser history, version history in Google Docs or Word | Shows the work developed over time |
| 3. Gather your disclosures | Any AI statement you submitted, plus written permission from the instructor | Proves the use was allowed |
| 4. Ask what evidence exists | Request the report and the reason for the concern | You can only answer what you can see |
| 5. Offer to talk it through | Explain your argument and sources in a meeting | Knowing your own work is strong evidence |
| 6. Use the formal route if needed | Student advocacy office, union or ombuds | Keeps the process fair and documented |
What I would not do: rewrite the flagged work through another tool and resubmit it. That turns a false positive into a real problem.
Worked example: the same tool, two outcomes
Example (I wrote both versions to show the point; they did not come from any tool).
A student writes a paragraph herself: "The council's plan was late, underfunded and, frankly, confusing to the people it was meant to help." English is her second language and her first draft read: "The plan of council was coming late and it had not enough money and people was confused." She runs her own draft through a humanizer to fix the phrasing, and her course allows language tools with a one-line statement. She adds: "I used [tool] to correct grammar and phrasing in my own draft." That is ordinary editing, disclosed.
A second student asks a chatbot to write the whole essay, runs the output through the same humanizer so it reads less like AI and submits it under a policy that bans AI drafting. Same tool, same button. The first case is allowed; the second is plagiarism under every policy quoted above.
My take: the question to ask is never "which tool did I use?" It is "whose ideas are these, and did I tell the reader what help I had?"
Workplaces and publishers
Outside school, the same logic applies with different rule-setters.
| Setting | Who sets the rule | What I'd check |
|---|---|---|
| Journal or book submission | The publisher (Elsevier, Springer Nature and others) | Disclosure wording, whether AI can be listed as an author (it cannot under both policies), image rules |
| Client or agency work | The contract and the client | Whether the client bans AI copy or asks you to declare it |
| In-house content and email | Your employer's AI policy | Approved tools, data rules for confidential text |
| Freelance platforms | The platform's terms and the buyer's brief | Whether AI use must be stated on delivery |
| Your own blog or newsletter | You | Accuracy, and what your readers would expect to be told |
For business writing, a humanizer is mostly a style and productivity question. The two risks I'd watch are confidentiality, since you are pasting text into a third party, and accuracy, since rewrites can drop qualifiers or invent detail. My guide on whether AI humanizers work covers the research on meaning drift, and the field-by-field comparison sits in the ranking of the best AI humanizers.
Legitimate uses of an AI humanizer
- Your own drafts outside graded work: emails, reports, posts and product copy where AI help is normal.
- Non-native speakers smoothing phrasing in their own writing, where the ideas and structure are theirs. Springer Nature's AI policy lists "using an LLM to polish or refine the language" as permitted, low-risk use for authors.
- AI-assisted drafts where AI use is allowed, edited so they read naturally, and disclosed if the rules ask.
- Accessibility needs agreed with your institution, which Oxford names as an exception.
Before pasting coursework into any tool, my guide on whether AI humanizers are safe covers what vendors' terms allow on your text. For what a school can check, see my guides on whether Turnitin can detect humanized AI text and whether teachers can detect AI. For the editing itself, the manual method in how to humanize AI text by hand keeps you in control of every change, and the guide on humanizing vs paraphrasing explains where rewording still needs a citation.
How to disclose AI use
- Check the format your course or journal asks for. Some want a sentence in the submission, some a form, some a citation.
- Name the tool and model, what you used it for and how far you edited the result. Elsevier asks authors for the tool name, purpose and extent of oversight.
- Cite generated content you quote or paraphrase. The MLA Style Center's updated 2025 guidance recommends describing the prompt, naming the tool, model, company and date, and linking a shareable conversation URL where the tool provides one.
- Keep your process. Save drafts, prompts and version history. If a detector flags your work, these are your evidence.
A short disclosure that works in most settings: "I used [tool, model] to [suggest an outline / fix grammar / rephrase two paragraphs]. I reviewed and edited all output and take responsibility for the content."
Disclosure wording by use
| What you did | Sample statement |
|---|---|
| Grammar and spelling only | "I used [tool] to check spelling and grammar." (Often not required; include it if unsure.) |
| Rephrased your own sentences | "I used [tool] to improve the phrasing of my own draft in sections 2 and 3. The ideas and structure are mine." |
| Brainstormed with a chatbot | "I used [model, date] to brainstorm possible topics. I chose the topic and wrote the essay myself." |
| Drafted with AI where allowed | "Sections 1 and 4 began as drafts from [model, date] in response to the prompt '[prompt]'. I rewrote, checked and extended them." |
| Translation help | "I used [tool] to translate my notes from [language]. I checked the translation against the original." |
My take: over-disclosing costs you nothing. Under-disclosing is the thing that turns a permitted use into a misconduct case.
How we checked this
- Read on 29 October 2026: USC Office of Academic Integrity's generative AI page; University of Oxford plagiarism guidance; University of Sydney's 27 November 2024 policy announcement; Elsevier's generative AI policies for journals; Springer Nature's AI editorial policy; MLA Style Center posts on citing generative AI (2023 and the August 2025 revision); Grammarly's AI humanizer and Authorship pages; QuillBot's AI Humanizer page; Turnitin's 27 August 2025 announcement; Perkins et al. 2024 (arXiv 2403.19148).
- What I did: I read each policy and vendor page in full and compared the wording side by side. The use table above is my reading of those documents.
- Scope: quoted policies apply to those institutions and publishers only. Your own course or employer policy decides your case.
Related reading: the guide to the Turnitin similarity score, the Turnitin AI detection review, and the HumanRank guides hub.
Frequently asked questions
Is using an AI humanizer cheating? It is when it hides AI use your course or publisher does not allow or requires you to disclose. Where AI help is permitted and disclosed, or for your own emails and business writing, it is ordinary editing.
Is it plagiarism to use AI? Under policies like Oxford's and USC's, presenting AI-generated material as your own without authorisation or acknowledgement is plagiarism. USC says this applies whether the material is copied or paraphrased.
Is Grammarly cheating? Spelling and grammar checks are rarely treated as cheating. Grammarly itself says using its humanizer to disguise restricted AI use may be cheating or plagiarism. Check your course policy for generative features.
Is QuillBot cheating? It depends on the assignment's rules. QuillBot tells users to follow institutional AI guidelines and says passing off reworded ideas as your own is unethical. A paraphraser still requires a citation for the source idea.
Is an essay humanizer allowed for school? Only if your course allows AI help, and then usually with disclosure. If AI is not allowed for the assignment, rewriting AI output does not make it allowed.
Can schools detect humanized text? Turnitin added AI bypasser detection for English in August 2025. Independent studies also show detectors make errors in both directions, so schools are told to use other evidence too. Detection is a separate question from whether the use was allowed.
What should I do if my own writing is flagged as AI? Read your school's policy, gather drafts and version history, show any disclosure or permission you have and ask to discuss the work. Detector scores alone are not proof, and USC tells instructors as much.
How do I disclose AI use in an assignment? Use the format your course asks for. Name the tool, say what you used it for and how you edited the result, and cite any AI text you quote or paraphrase, following MLA or APA guidance.
REVIEW NOTEI checked this on 29 October 2026 against published policies from the University of Southern California, the University of Oxford and the University of Sydney, journal policies from Elsevier and Springer Nature, MLA citation guidance and the tool makers' own statements.