Accused of Using AI: How to Actually Defend Your Work
When an assignment or a piece of writing is suspected of being AI-generated, some responses work and most don't. Here is what to say, in what order, and what to have ready before it happens.
It usually starts with one line in an email or a message. "Was this assignment written with AI?"
If you wrote it yourself, it stings. If you did use a tool somewhere, it's unclear how much to say. Either way, a bad response makes things worse. Let's take it in order.
First: the responses that don't work
Better to clear these out of the way, because the reflex reactions tend to hurt you.
Bringing back a result from a different detector. The most common attempt, and it almost never lands. The other side already knows tools disagree. Worse, "I ran it through several tools" can make your intent look questionable.
Reacting emotionally straight away. "Why would I ever do that?" contains no information. The person checking is usually working through a procedure, not trying to corner you.
Over-explaining. Answer things nobody asked and your account gets long, and long accounts develop inconsistencies. Answer the question that was asked.
Quickly editing the text. Changing the file after you've already submitted makes the situation dramatically worse. The modification timestamp stays.
What does work: the process
You have to flip the direction of proof. "I didn't use AI" is proving a negative, which is close to impossible by design. Instead, show "here is the process I went through."
In rough order of how much they help:
1. Writing history
Version history in a cloud document is the strongest single item. Hours of sentences accumulating, being deleted, being reordered is hard to fabricate. Even with local files, drafts and intermediate versions saved by date are convincing enough.
2. Traces of your materials
- The collection of links you consulted
- Photos of a handwritten outline or mind map
- Interview notes, lab notebooks, photographs you took
- Library loan records
Material that didn't make it into the final text is especially persuasive. Having gathered things you ended up not using means you actually did the research.
3. Explaining the content
This is the definitive verification in the end. If you can answer "why did you choose this example in section 3?" or "where did this statistic come from?", most cases close there.
To prepare, write one line per section about why you ordered it the way you did. The person who can explain the structure is the person who wrote it.
Handling specific situations
A school assignment
- Answer with facts, not feelings. "I wrote it myself. I can show you materials from the writing process."
- Organize everything and submit it in one go. Delivering it piecemeal erodes trust.
- If a meeting is offered, take it. A lot gets resolved face to face.
- If a formal procedure starts, find out what the procedure is. Your institution's rules will set out your right to respond and how to appeal.
If you did use a tool
Hiding it is the riskier path. If it surfaces later, the concealment becomes a bigger problem than the use.
- Be specific about where and how you used it. "I used it to sketch the outline and wrote the body myself" and "I had it write everything" are entirely different matters.
- Check your institution's policy. Most university guidelines don't ban AI outright — they require disclosure and accountability. If you were inside the rules, say so.
- If you did breach the rules, identifying exactly which part moves things along faster than defending yourself.
Freelance and client work
A contract changes the approach.
- Check first whether the contract has an AI-use clause.
- Send an organized package of your process: staged draft files, research material.
- For future work, agree the scope of AI use in writing so it doesn't recur.
What to do in advance
Suspicion arrives without warning. Three habits cut the cost of responding enormously.
- Write in cloud documents. Version history accumulates automatically.
- Keep one folder of materials. Everything you consulted for a piece, in one place.
- Check once before submitting. Running it through an AI detector shows you which paragraphs read mechanically. Not to lower a score — to know where the questions are likely to come from.
Why writing you did yourself scores high is covered separately in why your own writing gets flagged as AI. Knowing which kinds of writing attract false positives also makes it easier to explain why you were questioned in the first place.
Finally
An AI detection result is a reference AI-probability signal, and no tool can identify an author. Concluding anything from a score alone is a risky call for the person checking, too. Pointing that out calmly is fair — but don't try to end the conversation with it. Showing the process is far faster.