Academic Integrity Policy AI: What Counts as Misrepresentation

Most university academic integrity policy on AI does not ban the tool. It bans the claim. Assisted drafting — using AI to brainstorm, outline, tighten sentences or check grammar — is usually permitted or tolerated. Misrepresentation — submitting AI-generated work as your own unaided effort, or hiding the fact that you used it when disclosure was required — is where misconduct findings come from. The line moves with the assignment, the course, and the instructor, which is why reading the syllabus matters more than reading the policy.

How do university AI policies draw the line between assisted drafting and misrepresentation?

Policies rarely publish a clean definition. What they do instead is define a prohibited act — usually “submitting work that is not the student’s own” or “misrepresenting the authorship of submitted work” — and then leave instructors to decide what that means for a given task.

In practice, three things separate permitted assistance from misrepresentation:

1. Who generated the substance. If you produced the argument, the evidence and the structure, and AI helped you express it, you are inside the assisted-drafting zone. If AI produced the argument and you cleaned up the prose, you are outside it.
2. Whether you disclosed it. Many policies make disclosure a condition of use. Using AI where it was allowed, then not saying so, can itself be the violation — even if the work is genuinely yours.
3. What the assignment asked you to demonstrate. An essay that assesses your reasoning treats AI differently from a lab report that assesses your data handling, which is treated differently again from a reflective journal that assesses your own experience. The same tool use can be fine in one and disqualifying in another.

A useful test: could you defend every claim in the document in a ten-minute conversation, without notes, with someone who disagrees with you? If yes, the work is yours in the sense the policy means. If you would be explaining someone else’s reasoning, it is not.

What does disclosure actually look like in practice?

Disclosure is not a confession. It is a short, specific statement of what you used and what you did with it. Most policies that require it do not prescribe a format, so the burden falls on you to be clear.

What works, across assignment types:

Assignment type	Typical disclosure approach	What to include
Essay or report	Footnote or methods note	Tool name, what you asked it to do, which sections it touched
Lab or data assignment	Methods section	Whether AI was used for analysis, code, or write-up only
Reflective or personal writing	Usually prohibited entirely	Check the syllabus before you start; this is the most common hard line
Code or problem sets	Comment block or README	Which functions or solutions AI suggested, and how you verified them
Group project	Team agreement, then a note	What each member used, and how the group checked the output

A disclosure sentence that survives scrutiny looks like this: “I used ChatGPT to suggest alternative phrasings for three paragraphs in section 2 and to check my citations. The argument, sources and conclusions are my own. I revised every suggested sentence before including it.”

That is specific, it is verifiable against your revision history, and it does not overclaim. Vague disclosure — “AI was used in the preparation of this work” — is worse than none, because it tells a reviewer you knew disclosure mattered and chose not to be clear.

If you want to see what a detector actually flags in your own draft before you submit it, the AI detector scores the whole document and highlights each sentence by AI probability, so you can see which lines carry the signal and fix them yourself.

How do I keep my process defensible if I’m questioned?

Defensibility is built before the accusation, not after. The students and faculty who come through misconduct processes well are the ones who can reconstruct what they did.

Keep these, as a habit rather than an emergency measure:

• Version history. Word’s tracked changes, Google Docs revision history, or dated files. This is the single most persuasive artefact you can produce, because it shows a draft developing over time.
• Your notes and sources. Rough notes, annotated PDFs, a reading list. They show the thinking that preceded the writing.
• Your prompts, if you used AI. Save them. A prompt that says “here is my outline, suggest tighter transitions” reads very differently from one that says “write me an essay on this topic.”
• A short process note. Two or three sentences in your own files describing how the piece came together. If you are ever asked, you have the answer already written.

If you are questioned, the worst response is to become vague or defensive about the tool. State plainly what you used, show the history, and explain your reasoning. Faculty are generally trying to establish authorship, not catch you out — and a clear account of a defensible process usually ends the conversation.

One thing worth knowing: detector output is a probability estimate, not proof of authorship. A high score is a reason to look closer, not a finding. If you are facing a flag on writing you did yourself, our guide to false positives and what to do when you’re accused walks through the response step by step.

Where the rules are ambiguous — and how to interpret them

Most policy gaps fall into four patterns. Knowing which one you are dealing with tells you who to ask.

The policy is silent. No AI clause at all. The default is the older rule: work must be your own. That does not prohibit AI assistance, but it does mean undisclosed AI-generated substance is still risky. Ask the instructor before the deadline, not after.

The policy is broad and the syllabus is specific. Many institutions set a permissive or restrictive baseline and let instructors tighten it. The syllabus wins. If your syllabus bans AI for a specific task, the university’s general permission does not override it.

Grammar tools are exempted, generative tools are not. This is the most common split, and it is genuinely confusing, because modern grammar checkers suggest rewrites that look a lot like generative output. The safe reading: if the tool changes meaning or supplies content, treat it as generative and disclose it.

Group work is unaddressed. Policies usually speak to individual submissions. If your group used AI in any form, agree the disclosure as a team, and put it in writing before submission. Disagreement between group members about what was used is a common trigger for referrals.

When in doubt, email the instructor with a specific question: “I’d like to use AI to check grammar and suggest transitions in my draft — is that acceptable for this assignment, and how would you like it disclosed?” That email is itself evidence of good faith, and it takes two minutes.

Using AI as a tool versus submitting AI as your work

The distinction is not about how much AI was involved. It is about who is answerable for the content.

Using AI as a tool looks like this: you have a thesis, you have sources, you ask for help expressing a point more clearly, you read the suggestion, you decide whether it is accurate, and you rewrite it in your own voice. You could defend the result without the tool ever existing.

Submitting AI as your work looks like this: you paste a prompt, receive a finished piece, skim it, and submit. The prose may be fluent. But the ideas, the structure and the conclusions came from somewhere you cannot account for, and you cannot stand behind them.

Here is a worked illustration. Take a flat, machine-shaped sentence:

Before: “The implementation of renewable energy policies is a multifaceted issue that requires careful consideration of various stakeholders.”

After: “Renewable energy rules fail when they ignore the people who pay for them — installers, grid operators, and households facing higher bills.”

The second version is not better because it dodges a detector. It is better because someone decided what the sentence was actually claiming.

That matters for how you read any score. In our own measurement on 2026-09-13, we scored 23 hand-written DeAIze guides with our detector at a 30% flag threshold: mean score 21%, median 22%, range 0-46%, and 5 of the 23 were flagged. Unedited model output scored higher on average — 12 passages, mean 33% — but the range overlapped almost completely with the human sample. A score tells you a text has machine-like patterns. It cannot tell you who wrote it, and it should never be treated as a verdict on its own.

If your own writing is picking up a flag and you want to understand why, how AI detection works explains the signals behind the score, and AI-assisted writing for students covers where the common rule boundaries sit.

A short checklist before you submit

• Does the syllabus or assignment brief say anything about AI? Read it again, specifically.
• Can you name what you used AI for, in one sentence, without hedging?
• Have you disclosed it in the format the assignment expects?
• Do you have version history or notes that show the work developing?
• Can you defend every claim and conclusion as your own reasoning?
• If the answer to any of these is no, fix it before the deadline, not after.

The policy is not there to catch you. It is there to protect the value of the work you actually did — which is the part no tool can supply.

---

Measurement note: figures in this article come from our own detector run on 2026-09-13 — n = 23 hand-written guides versus n = 12 unedited model outputs, median AI score 22% and 34% respectively. Method: /methodology/.