How to Prove You Didn't Use AI to Write Your Paper

how to prove you didn't use ai
Table of Contents

Knowing how to prove you didn’t use AI matters most on the day a detector flags your paper.

A detector score is not proof. Your writing process is the best evidence you have. This guide shows you what to do first, which records you already have, and how to reply calmly with a template.

Build your evidence before the meeting.

Key Takeaways

  • Protect the Record First: Don’t edit, rewrite, or delete anything, and ask for the report and deadline in writing.
  • A Score Isn’t Proof: Turnitin and OpenAI both say AI detection can be wrong and shouldn’t decide a case alone.
  • Show Your Process: Version history, dated drafts, notes, and feedback emails show a paper growing over time.
  • Know Each Weak Spot: Word history needs OneDrive, Authorship must be on, and big pastes show as jumps.
  • Explain It Out Loud: A five-minute walk-through of your thesis, sources, and cuts shows you know the paper.
  • Skip the Humanizer: Paraphrasing tools can get flagged themselves and break the match with your own drafts.
Writing help

Struggling with Clarity in your writing?

You're not alone. Many writers face this exact challenge.

Orwellix provides you with advanced writing tools specifically designed to overcome common writing hurdles. Our AI-powered platform helps you craft clearer, more engaging content with less effort.

Start for Free

The Short Answer: Show Your Process, Not a Score

You can’t prove a negative, so don’t try to prove you didn’t use AI. Show how you wrote the paper instead. Version history, dated drafts, notes, and your ability to explain your own argument carry more weight than any detector score, theirs or yours.

The steps below turn that into a plan: what to do today, which records your writing setup kept, how to present them, and what to say.

What to Do in the First 24 Hours

Your first job is to protect the record and get the facts in writing. Don’t argue yet. A calm, well-documented reply a day later beats a fast, angry one sent tonight.

The first 24 hours

Protect the Record Before You Reply

Don’t touch the fileNo edits or resubmits
Keep everythingDrafts, notes, chats
Ask in writingReport, passages, deadline
Read the rulesCourse and school
Start a logDates and messages
Orwellixorwellix.com
Freeze the evidence and get the facts in writing before you argue anything.
  1. Don’t touch the file. Don’t edit, rewrite, or resubmit the paper. Every change adds new entries to the history and blurs the timeline you’re about to rely on.
  2. Don’t delete anything. Keep old drafts, notes, outlines, chat logs, emails, and stray files, even the messy ones. Messy is good evidence.
  3. Ask for the details in writing. Reply by email and ask which tool flagged the paper, the score or full report, which passages were flagged, which policy you may have broken, and the deadline to respond.
  4. Read the rules that apply. Find the AI policy in your syllabus or course page. Then read your school’s academic integrity procedure: who decides, whether you can bring an advisor, and how an appeal works.
  5. Start a log. Make one folder for this case. Save every message and note each date, meeting, and deadline in a simple list.

Most schools have someone whose job is to help students through this, such as an academic advisor, a student advocate, or an ombudsperson. Contact them early. This guide isn’t legal advice, and they know your school’s process.

If you used AI in a permitted way: Say exactly how, in writing, before anyone asks twice. “I used ChatGPT to suggest search terms, as the syllabus allows” is a fact, not a confession. Here’s how to cite or disclose the AI help you used in the format your course expects.

Once the request is sent, you’ll usually have a few days. Use them to understand what the score does and doesn’t mean.

Why an AI Detector Score Isn’t Proof

An AI detector gives an estimate, not a finding. The companies that build these tools say so themselves, which is why their own words are the most useful thing you can quote. You aren’t attacking your instructor. You’re asking that the score be weighed the way its maker says it should be.

AI detector accuracy

What the Numbers Say About AI Scores

26%of AI text caught by OpenAI’s own toolOpenAI · 2023
9%of human writing labeled as AIOpenAI · 2023
61%of non-native writers’ essays flagged, on averageLiang et al. · 2023
~750of 75,000 papers possibly mislabeledVanderbilt · 2023
Orwellixorwellix.com
Sources: OpenAI’s AI classifier announcement (2023), Liang et al. in Patterns (2023), and Vanderbilt University’s Turnitin guidance (2023).

What Turnitin Says About Its Own AI Writing Report

Turnitin’s guide for instructors, Using the AI Writing Report, sets out these limits in plain terms:

  • Not a sole basis: The model “may not always be accurate,” so it shouldn’t be the sole basis for adverse action against a student. Human judgment and the school’s policy have to decide.
  • Low scores get an asterisk: False positives are more common between 0% and 19%, so scores from 1% to 19% show as an asterisk (*%) with no number or highlights.
  • Length and format matter: A report needs at least 300 words of prose. The model doesn’t reliably read poetry, scripts, or code, and it doesn’t detect bullet points, tables, or annotated bibliographies.

The AI score is also separate from the similarity score. Similarity shows text that matches other sources. The AI score guesses how text was produced. The two answer different questions, as this guide to plagiarism checkers for students explains. If you want the wider picture, see how AI-generated content and plagiarism overlap and where they don’t.

What OpenAI Found When It Built a Detector

OpenAI released its own AI text classifier in January 2023. By July 20, 2023, it had withdrawn the tool “due to its low rate of accuracy.” At launch, it caught 26% of AI-written text and mislabeled human writing as AI 9% of the time.

OpenAI’s help article for educators asks “Do AI detectors work?” and answers, “In short, not in our experience.” It adds that ChatGPT’s answers to “did you write this?” are random and have no basis in fact.

Who Gets Flagged Most Often

False positives don’t fall evenly. In a 2023 study published in Patterns, Liang and colleagues ran 91 TOEFL essays by non-native English writers through seven detectors. The detectors flagged them as AI-written at an average false-positive rate of 61.22%. The same detectors were near-perfect on essays by US eighth graders. If English is your second language, these tools are more likely to misread you. Look for AI writing help for non-native English writers that you can disclose and explain, not tools that hide your voice.

Schools have noticed. In August 2023, Vanderbilt University disabled Turnitin’s AI detector. It noted that Turnitin claimed a 1% false-positive rate. With 75,000 papers submitted in 2022, that rate could have meant around 750 papers labeled incorrectly.

Writing-tool companies say the same about their own detectors. Here’s what Grammarly and QuillBot say about their detectors in their own words.

How to use these facts: Quote one source, briefly, in your reply. For example: “Turnitin’s guidance says the AI score shouldn’t be the sole basis for a decision, so I’ve attached my drafts and version history.” Then let the evidence do the rest.

The Evidence That Shows Your Writing Process

Strong writing process evidence shows a paper growing over time: an outline, rough paragraphs, cuts, and changes after feedback. Which records you have depends on where you wrote. Here they are, strongest first, with the weak spot of each so nothing surprises you in the meeting.

EvidenceWhere to find itWhat it showsWeak spot
Google Docs history“Last edit,” top rightEdits across daysPastes show as jumps
Word version historyFile title, then Version historySaved versionsOneDrive or SharePoint only
Grammarly AuthorshipAuthorship reportA typing replayOnly if it was on
Draftback replayChrome extensionDocs history as a replayPaid, needs edit access
Research trailNotes, sources, emailsHow ideas formedNot tied to the file
Dated draft copiesEmail, cloud backupsSnapshots in timeGaps between copies

Google Docs Version History

Google Docs saves your edits as you type, so it’s often the best record you have. Google’s help page gives the path: open the doc and click Last edit at the top right. Only people with edit access can see the history. So you’ll usually show it on your own screen rather than send a view-only link.

  • Name key versions: Open the menu next to a version and choose “Name this version” (for example, “First full draft, Oct 3”). Named versions are easier to show.
  • Never clean up: Deleting version history is permanent and can’t be undone.
  • Know the weak spot: A large paste shows as one big jump. If you drafted a section in your notes or an email and pasted it in, say so and bring that source too.

Word Version History (OneDrive or SharePoint)

In current Microsoft 365, open the file, click the file title, and select Version history. In Word 2016–2021, go to File > History. This only works for files stored in OneDrive or SharePoint. A personal Microsoft account keeps the last 25 versions, while school accounts depend on how the library is set up.

Word Files Saved Only on Your Computer

A Word file saved only on your laptop has no version history. Don’t panic. Gather the next-best records instead:

  • Earlier drafts saved under different names or dates
  • Copies you emailed to yourself, a classmate, or a tutor
  • Backups in a cloud folder, with their upload dates
  • The file’s properties (created and modified dates), as supporting detail only, since dates are easy to question

Grammarly Authorship (Only If It Was On)

Grammarly Authorship records how text entered a document, typed or pasted, and can replay your writing. It works in Google Docs and Microsoft Word, and it’s available on Grammarly Free and Pro. It’s opt-in, so it only helps if you turned it on before you wrote.

Offer the replay, not just the headline percentage. In November 2025, Plagiarism Today found that AI text rephrased by Grammarly could appear in the report as mostly “typed by a human.” Grammarly’s owner later said it was changing how that text is displayed. A reviewer who watches the replay sees what actually happened.

Draftback Replays

Draftback is a Chrome extension that replays the revision history of a Google Doc like a video. You need edit access to the doc, and it now requires a paid subscription. It turns the same Docs history into something easier to watch in a meeting.

Your Research Trail

Your research trail shows the thinking behind the words. Collect:

  • Notes, outlines, and annotated sources
  • Paragraphs you cut, and early versions of your thesis
  • Feedback emails from your instructor, a tutor, or the writing center
  • Library or database search history
  • Shared AI chat links, if your course allowed AI and you used it

No single item proves authorship. Together, they show a person working through a problem over time.

Show You Know the Paper: The 5-Minute Walk-Through

The person who wrote a paper can explain it without notes. Prepare a five-minute walk-through and practice it out loud once:

  1. Your thesis in one sentence, and why you chose it.
  2. Why each key source is there, and what it added to your argument.
  3. What changed between drafts. For example: “I first argued that remote work cut costs, but the 2022 survey changed my focus to turnover.”
  4. A paragraph you cut, and why it didn’t fit.
  5. One weak spot you would fix with more time.

Then offer a meeting, or a short supervised writing sample on a related prompt. Offer it, don’t demand it. Your instructor may already have a process, and an offer shows confidence without pressure.

How to Present Your Evidence

Turn your scattered files into a one-page timeline. A reviewer should be able to see your process in under a minute. Here’s a sample for a fictional essay:

DateWhat changedEvidence file
Sep 28Notes on four sources01-notes.pdf
Oct 2Outline and working thesis02-outline.png
Oct 6First full draft03-history-oct6.png
Oct 9Revised after tutor feedback04-tutor-email.pdf
Oct 12Final edits, submitted05-history-oct12.png
  • Screenshot milestones: Capture the version history at the first outline, first full draft, after feedback, and final version.
  • Name files by date and order: Numbered names keep the timeline readable when someone opens the folder.
  • Show the live history: In a meeting, share your screen and open the version history in real time. A live record is more convincing than a screenshot.
  • Keep originals untouched: Send copies and screenshots. Never edit the original file to tidy it up.

Email to Your Professor About an AI Accusation

Your reply should be short, polite, and specific. State that you wrote the paper, show what you have, and ask for what you need. Copy this and fill in the brackets:

Subject: [Course code] – [Paper title]: my writing process

Dear Professor [Name],

Thank you for letting me know about the concern with my paper.
I wrote this paper myself, and I'd like to show you how it developed.

I've attached:
- A one-page timeline of my drafts
- Screenshots of the Google Docs version history from [dates]
- My notes, outline, and the feedback email from [tutor/writing center]

I'm happy to walk you through my process in a meeting, or by sharing
my screen and opening the full version history.

Could you please send me the AI report and the passages that were
flagged, and tell me which part of the course policy applies?
I want to make sure I respond by [deadline]. Please let me know if
that date is correct.

Thank you for your time.

[Full name]
[Student ID]

If you used AI in a way the course allows, add one plain sentence after the second line:

I used [tool] for [task, e.g. finding search terms], which the syllabus
allows. I've attached the chat link. All of the writing and analysis is my own.

Keep the tone flat and factual. Quote one source about detectors at most. Send it, then wait for an answer before you send anything else.

What Not to Do When You’re Falsely Accused of Using AI

When you’re falsely accused of using AI, panic pushes you toward moves that weaken a strong case.

What not to do

Seven Moves That Weaken a True Case

01Humanizers and paraphrasers
02Rewriting after the flag
03Faking drafts or timestamps
04Asking a chatbot
05Shopping for a score
06Denying allowed AI
07Accepting a disputed finding
Orwellixorwellix.com
Each of these breaks the match between your paper and your own records, or adds a new problem.

Avoid these:

  • Running the paper through an AI humanizer or paraphraser: Turnitin’s report flags text it thinks was altered by AI bypassers, paraphrasers, and word spinners. The new wording also stops matching your own drafts, which destroys your best evidence.
  • Rewriting the paper after the flag: The paper under review is the one you submitted. A new version looks like damage control.
  • Changing timestamps or making drafts after the fact: Version histories record when edits happened. Fabricated evidence turns a disputed score into real misconduct.
  • Asking a chatbot whether it wrote your essay: OpenAI says these answers are random. They prove nothing either way.
  • Shopping for a friendlier detector score: A second detector’s low score cuts both ways. If you rely on one tool’s number, you accept that the next tool’s number counts too.
  • Denying AI use that was allowed: If you used a permitted tool, say so. A small, honest disclosure protects everything else you say.
  • Accepting a finding you dispute just to end it: Agreeing can leave a record. Ask your advisor about your options first.

Before Your Next Paper: Build the Record From Day One

The best time to collect evidence is before anyone asks for it. A few habits make your next paper easy to defend:

  • Write in one tracked place: Draft in Google Docs or a Word file saved to OneDrive, from the very first note.
  • Name versions at milestones: Outline, first draft, after feedback, final.
  • Keep your trail: Save notes, sources, and every feedback email in one folder.
  • Keep AI records when AI is allowed: Save transcripts or share links, and disclose the use the way your course asks. If you’re unsure what’s allowed, compare your syllabus with this guide to AI writing tools for students.

Good writing habits help too, but not as a way to “sound human.” A clear thesis and analysis in your own words make the walk-through easy, because the argument really is yours. This list of common academic writing mistakes is a good place to check both.

A Note for Researchers and Authors

The same approach works for a thesis, journal manuscript, or book chapter. Version history, drafts shared with co-authors or supervisors, lab notes, and responses to reviewers all show your process.

Before you reply to an editor, read your journal or publisher’s AI policy. Check whether it asks you to disclose AI use, and where in the manuscript that disclosure goes. If a similarity report is also part of the question, compare similarity checkers for theses and manuscripts so you know what each report measures.

Check Your Own Draft in Orwellix (and Keep Your Record Intact)

Orwellix doesn’t verify authorship, detect AI, keep a writing replay, or help anyone pass a detector. Where it can help is self-review of a draft you wrote yourself, before you submit.

Keep your Google Docs or Word file as the record. Work on a copy in Orwellix, then make any changes yourself in the original file. Pasting a revised version back would show up as one large paste in your history.

  1. Paste a copy of one section into a new Orwellix document.
  2. Run the checks. Readability analysis, included on every plan, flags hard-to-read sentences. The grammar checker covers 5,000 characters a month on the Free plan.
  3. Ask questions in Ask Mode. It costs 1 credit per question and answers without editing your document. Try “Which paragraph’s claim has the least support?” or “Where does my argument skip a step?”
  4. Use Agent Mode only if your course allows AI editing. Check the policy first. Agent Mode proposes tracked edits, so you can review each AI edit and accept or reject it. Disclose that use the way your course asks.
  5. Make your changes in the original file by hand, so your version history stays a true record of your work.

You can start on the Free plan without a card.

Write smarter with Orwellix

Orwellix's AI capabilities help you craft clearer, more effective content.

Start for Free

Conclusion

An AI accusation feels personal, but it is mostly a records problem. Protect the file, ask for the report and deadline in writing, and read the policy that applies. Detector scores are estimates, and the companies that make them say a score shouldn’t decide a case alone. False positives also land hardest on non-native English writers.

Your answer is the paper’s history. Google Docs or OneDrive version history, dated drafts, notes, feedback emails, and an Authorship replay each show part of the story. Each also has a weak spot you should know before the meeting. A one-page timeline, a calm email, and a five-minute walk-through bring those pieces together. The moves to avoid are the ones that break the record: humanizers, rewrites, and made-up drafts.

As AI tools spread through every writing app, schools will rely less on scores and more on process. Students who keep a clear record from the first note won’t need to scramble. Tools like Orwellix can help you review a draft you wrote, as long as the final changes happen in your own tracked file.

Your process is the proof, so start keeping it today.

Frequently Asked Questions (FAQs)

1. Can a professor prove you used AI?

A detector score alone can’t prove it, and Turnitin says its AI score shouldn’t be the only basis for action against a student. A fair review weighs the score with other evidence, such as your drafts, version history, and how well you can explain the paper. That is why your process evidence matters more than the number.

2. Is Google Docs version history enough to prove I wrote my paper?

It is often the strongest single record, because it shows the paper changing over days. It works best with supporting evidence like notes, an outline, and feedback emails. Be ready to explain any large pastes, such as text you drafted in another document first.

3. What if I wrote my paper in Word without OneDrive?

A Word file saved only on your computer has no version history. Gather earlier drafts, copies you emailed, cloud backups, notes, and outlines instead. Then prepare a short walk-through of your argument, since explaining your choices is strong evidence on its own.

4. What is the best authorship verification tool for Google Docs?

Start with Google Docs’ own version history, which every document already has. Grammarly Authorship adds a typing replay, but only if you turned it on before you wrote. Draftback replays Docs history as a video, but it requires edit access and a paid subscription.

5. Can Turnitin detect paraphrased AI text?

Turnitin says its English AI report looks for AI text that was changed by paraphrasers, bypassers, and word spinners. That is why running your own paper through one of those tools is risky. It can get flagged, and it no longer matches your drafts.

6. Will using Grammarly get my paper flagged as AI?

It depends on how you use it. Accepting AI rewrites of whole sentences carries more risk than fixing a typo, because the wording is no longer fully yours. Check your course policy on AI editing tools, and keep your version history so you can show which changes you made.

7. Should I run my essay through an AI detector before I submit it?

It usually doesn’t help. Detectors disagree with each other, and a low score from one tool won’t change what your school’s tool reports. Your time is better spent keeping version history, notes, and drafts that show how you wrote the paper.

Try it yourself

Try Orwellix for Free

Experience Orwellix AI Agent's capabilities on the free forever plan, with 25 AI credits a month and no card needed. Need more? Try Pro or Premium free for 7 days.

Start for Free