How to prove you didn't use AI when you're accused
To prove you didn't use AI, show your process, not a score. Export your document's version history, gather drafts, notes, outlines and sources, and offer to talk through the work in person. A detector result is a statistical estimate, not evidence, and a record of the work being built over time is much harder to argue with.
Getting accused of using AI on something you wrote yourself is stressful, and it happens more often than most schools admit. The good news is that the strongest evidence is usually already sitting in your files. This guide covers what to collect, how to handle the conversation, the mistakes that make things worse, and how to make the next accusation unlikely. It's written mainly for students, but the same steps work for freelance writers and anyone whose client or employer has raised the question.
Why do people get falsely accused of using AI?
Most accusations start with an AI detector. Detectors don't recognise ChatGPT. They estimate how predictable a piece of text is: how safe each word choice is, how even the sentences are, how much the rhythm varies. Careful, formal, well-organised writing is predictable by design, so it can score as "AI" even when a person wrote every word. We explain the mechanics in does my writing sound like AI? How to check.
Some writers are hit harder than others. A Stanford study found that popular GPT detectors flagged more than half of a set of essays written by non-native English speakers as AI-generated, while scoring essays by native-speaking students far more accurately (Liang et al., arXiv 2304.02819). Short texts, heavily templated assignments and writing that follows a strict rubric are also prone to false flags. The other trigger is a teacher's gut feeling, often set off by the same vocabulary we list in common AI words to avoid.
How do you prove you didn't use AI? Step by step
- Stay calm and ask what the evidence is. Ask which tool flagged the work, what score it gave, and which passages were marked. You can't answer an accusation you haven't seen, and asking is a reasonable request, not a confrontation.
- Export your version history. In Google Docs, open File → Version history → See version history. In Word, files saved to OneDrive or SharePoint have version history under File → Info. Screenshot or export the timeline that shows the document growing over hours or days, with deletions and rewrites along the way.
- Collect drafts, notes and sources. Outlines, handwritten notes, earlier drafts saved as separate files, library loans, research tabs, bookmarked articles, and messages where you discussed the topic with classmates or a tutor. Anything with a timestamp helps.
- Offer to talk through the work. Offer to explain your argument, why you chose particular sources, and what you cut. If it helps, offer to write a short piece on the topic under supervision. People who wrote something can usually discuss it easily; that's persuasive in a way no file is.
- Explain what detectors can and can't show. Calmly, and without attacking anyone. Point out that a detector score is a probability estimate, and that researchers have shown detectors can be unreliable (Sadasivan et al., arXiv 2303.11156) and biased against non-native writers.
- Use the formal process if you need it. Read your institution's academic integrity policy. Ask for the concern and the evidence in writing. If it goes to a hearing, bring an academic advisor or a student union representative, and submit your evidence as an organised packet with dates.
What evidence actually proves you wrote it?
Not all evidence is equal. Here's roughly how the common kinds compare, from strongest to weakest:
| Evidence | What it shows | How convincing |
|---|---|---|
| Version history (Google Docs, Word) | The text growing over time, with typing, deletions and rewrites | Strong, especially over several sessions |
| Talking through the work in person | You understand the argument, sources and choices | Strong, often decisive |
| Earlier drafts and outlines | The piece changing shape before the final version | Strong if dated |
| Research notes and sources | Where the ideas and quotes came from | Medium to strong |
| Past writing samples | Your voice and habits match earlier work | Medium |
| Messages about the assignment | You were working on it when you said you were | Medium |
| A low score on a different detector | That detectors disagree | Weak on its own |
If you typed the whole thing in one sitting, or pasted it in from another app, your version history will be thin. That doesn't mean you're guilty, but it does mean the conversation matters more. Lean on your notes, sources and your ability to explain the work.
What should you say if a professor accuses you of using AI?
Keep it short, factual and calm. Something like this works as a reply to an email or as your opening in a meeting:
Then follow up any meeting with a short email summarising what was discussed and what you shared. It creates a record, and it keeps everyone working from the same facts. If you're angry, and most people are, write the email, wait an hour, and cut anything that sounds like an accusation back.
What not to do after an AI accusation
- Don't rewrite or "humanize" the submitted work. Changing the text after the fact makes it look like you're hiding something, and it muddies the version history that's your best evidence.
- Don't rely on screenshots of other detectors. A low score on a second tool proves only that detectors disagree. Include it if you like, but lead with process evidence.
- Don't delete anything. Keep every draft, note and file exactly as it is, including the messy ones.
- Don't argue in public first. Posting about it before the process is over rarely helps and can make the meeting harder.
- Don't admit to something you didn't do to make it go away. If you're under pressure to accept a penalty quickly, ask for time and talk to an advisor.
How do you avoid being accused of using AI next time?
Most of this is about keeping receipts as you go, which costs nothing:
- Write in one document with version history on. Google Docs does this automatically. Avoid drafting elsewhere and pasting the final text in.
- Save your research trail. Keep a sources list or a notes document you add to while you read.
- Keep the messy drafts. "Draft 1" through "final-final" files are useful evidence, so don't clean them up.
- Read your final draft for flat passages. Tidy, predictable prose is what gets flagged. Vary sentence length, cut stock connectors, and add specifics from your own reading and experience. Our guide to make your essay not sound like AI walks through the fixes for school writing, and make your writing not sound like AI covers the same habits more broadly.
- Know the rules. If AI is allowed for brainstorming or grammar checks in your course, note what you used it for. If it isn't allowed, don't use it. No rewrite changes what your course permits.
Checking your own draft with a detector before you submit can help you spot the passages most likely to raise questions. Treat it as an editing prompt, not a verdict, because the same limits apply to every detector, including the one in our app. The wider picture of why writing gets read as AI is in our guide on how to not sound like AI.
Frequently asked questions
How can I prove I didn't use AI?
Show your process. Version history, earlier drafts, notes, outlines and sources show the work being built over time. Offering to explain your argument in person is often the most convincing step of all.
Can an AI detector prove I used AI?
No. A detector estimates how predictable the text reads. Research has shown detectors produce false positives, especially on writing by non-native English speakers, so a score on its own shouldn't be treated as proof.
What should I say if my professor accuses me of using AI?
Say plainly that you wrote it, ask what evidence the concern is based on, and offer your version history, drafts and a conversation about the work. Put the key points in a follow-up email.
Does running my essay through another AI detector help?
Not much. Detectors disagree with each other, so a low score on one tool doesn't cancel a high score on another. Process evidence carries far more weight.
How do I avoid being accused of using AI in the future?
Write in one document with version history on, keep notes and drafts, save your sources, and revise flat, formulaic passages in your own words before you submit.