Falsely Accused of Using AI on Your Essay? Here's What to Do

1. Preserve your evidence immediately

Do not rewrite files, delete history or create new "proof" — preserve what already exists. Your strongest evidence is usually the trail created while you were actually researching and writing:

• Original Word/Google Docs files with timestamps and version history
• Earlier drafts, even messy ones
• Outlines, handwritten notes, mind maps
• Research notes and downloaded articles
• Browser or library research history, where available
• Reference-manager records
• Track Changes, comments and feedback
• Assignment instructions and the rubric version you used
• If AI use was permitted: your prompts, outputs and disclosure

Make copies, but keep originals intact — provide evidence through the official channel rather than editing files to make them look more persuasive.

2. Read the exact rule you're alleged to have broken

AI policies vary by institution, module and assessment. Find the rule that applied on the submission date — was all generative AI prohibited? Was brainstorming allowed? Language editing? Was disclosure required? A general statement about AI isn't a substitute for the actual rule governing your task. Save a copy or screenshot of the relevant policy, tutorial letter or LMS announcement.

3. Ask what evidence triggered the concern

Respond respectfully while asking for clarity — was it an AI detector, a change in writing style, unusual citations, a lecturer's observation, metadata, or a combination? Ask what part of the work is in question and what process will follow. A detector percentage alone is not proof of misconduct — what matters is how the institution uses that signal under its own policy, and what corroborating evidence exists.

4. Build an authorship timeline

Put together a simple chronological record — when you received the task, what you researched, when you drafted and revised each section, when you submitted — with evidence attached only where it supports a point. A clear timeline beats a folder of hundreds of unexplained files.

5. Be ready to explain your own work

A genuine author can normally discuss the central argument, why particular sources were chosen, what a difficult paragraph means, and how the paper changed during drafting. Be ready to:

• Explain your thesis in plain language
• Identify your two or three most important sources and what you used them for
• Explain a paragraph the reviewer selects
• Describe a change between an early draft and the final version
• Be candid about any permitted tools you did use

6. Keep your response factual

Don't accuse the lecturer of lying, claim all AI detectors are useless, or rely on a different online detector's result — independent detectors often disagree. Your stronger argument is that automated detection has known limitations and your own authorship evidence should be considered alongside any indicator.

7. Use the formal review or appeal route if necessary

If it becomes a formal misconduct allegation, ask for the applicable procedure, evidence, deadlines and appeal rights. Meet each deadline. Respond to the allegation actually made. If consequences are serious, consider advice from your student representative body, ombud, academic adviser or an appropriate legal professional.

What not to do

• Don't fabricate drafts or backdate files
• Don't run the essay through a "humanizer" after the fact and present it as original
• Don't delete AI chats if AI use was permitted and legitimate
• Don't claim "0% on another detector proves I wrote it" — it doesn't
• Don't admit to conduct you didn't do just to make the issue disappear

Why false positives are a real concern

Research has documented substantial variation in AI-text detector accuracy, including false-positive risk. A widely cited 2023 study (Liang et al., Patterns) found serious bias in several detectors against non-native English writing. This doesn't mean every flag is wrong — it means a flag needs interpretation and corroboration.

The useful principle: the question isn't "what did the detector say?" but "what does the total evidence show?" Drafts, sources, version history, understanding of the paper and compliance with the applicable AI policy are all part of that evidence.