False positives · the evidence file, for authorship you genuinely hold
What actually proves you wrote something
You know you wrote it. The question this page answers is narrower and colder than that: what do you actually have, in hand, that someone else who wasn't there would find convincing. Not an argument. A file.
No piece of evidence proves authorship on its own, including the essay itself, and that symmetry is the thing worth understanding before you build your case. A detector cannot prove you used AI. You cannot prove from the final document alone that you didn't. What can shift a hearing is process evidence: a version history showing incremental changes over real time, drafts with your handwriting or your specific mistakes in them, a search or reading trail that predates the essay, and notes that show your thinking before it became prose. None of these are unforgeable, but together they describe a writing process, and a writing process is what a hearing is actually trying to establish, not a verdict on the text's style.
Why this happens
The finished essay carries no reliable signature of who typed it, because the same sentence can be produced by careful human drafting or by a model, and the surface text does not record which happened. Detectors try to infer authorship from statistical properties of that surface text, which is an inference about a pattern, not a direct read of who was at the keyboard, and a bare score cannot close that gap by itself. What text cannot show, process can: a document that grew over three weeks in visible increments is much harder to fake convincingly than a paragraph is to write convincingly, because faking a process means fabricating a timeline, not just a style. That asymmetry, style is cheap to imitate, process is expensive to fake, is why evidence-gathering focuses on process and not on the prose itself.
Where honest writing and machine writing genuinely overlap
These are real patterns, not coincidences. Each one is something a careful human writer in this situation has good reason to do, and something a model does for entirely different reasons.
- A single, uninterrupted draft with no revision history
- Both a rushed honest student and a copy-pasted AI draft can produce a document with no saved intermediate versions. The absence of history is genuinely ambiguous, which is exactly why it's worth avoiding going forward.
- A finished document opened and pasted in one sitting
- File metadata showing a document created and completed within minutes reads the same whether a person typed fast or copied text in. Metadata records timing, not typing.
- Research notes that don't match the final essay's structure
- Genuine research is messy and rarely maps cleanly onto the finished argument; notes that align too neatly with the final essay can look staged even when they're real, simply because real thinking doesn't usually arrive pre-organized.
- No searchable trail for cited sources
- A source cited in the essay but never appearing in your browser history or library account raises a real question regardless of authorship, since it suggests the source wasn't actually consulted the way the citation implies.
What actually helps
Google Docs and Word Online keep a full version history by default; check yours before you do anything else, since it may already exist without you having built it on purpose. If you write in a plain text editor or offline, save drafts as separate timestamped files going forward, and start now even if you have nothing for past work. Keep your outline, your research notes, and your browser history or reading list from while you worked; a scanned handwritten outline dated before the typed essay is stronger evidence than almost anything digital, because it is harder to fabricate after the fact. None of this is about building an alibi for a crime you didn't commit. It's about being able to show your work the way you'd show a math problem, because 'I know I did it' is not evidence to anyone who wasn't in the room.
What the flattening looks like
On the left, honest writing with the specifics drained out of it, which is the shape that reads as machine-written. On the right, the same content with the writer's own detail restored. The voice is The Cold Analyst
The evidence clearly demonstrates that I completed this assignment independently and without any unauthorized assistance. I have always been a diligent and honest student, and my work ethic speaks for itself. I am fully prepared to answer any questions about my process and stand firmly behind the authenticity of this submission.
The evidence shows I completed this assignment alone, without help I should not have had. I have always worked hard and honestly, and that record stands. I can answer any question about how I did the work, and I mean it when I say the submission is mine.
Steps worth taking
- Check what you already have Google Docs, Word Online, and most cloud editors keep version history automatically. Look before assuming you have nothing.
- Start saving drafts now, even mid-project A timestamped file saved today is still evidence for anything you finish from here forward, even if past work has nothing to show.
- Keep the mess, not just the outline Crossed-out notes, an abandoned first paragraph, a rejected thesis statement, all of these look more like a real process than a clean outline does, because real thinking is rarely clean.
- Match your citations to a real search trail If you can show you actually looked something up, browser history, a library database login, a saved PDF, that trail supports the essay far more than the essay supports itself.
Questions
How can I prove I didn't use AI?
You largely can't, not from the finished text alone. What you can do is show a writing process: version history, dated drafts, notes, and a research trail that together describe how the essay came to exist over time.
Can a professor prove I used AI?
Not from a detector score alone. A detector score is a statement about a text property, not about who typed it, and it cannot reliably distinguish machine text from certain honest human registers, formal ESL writing among them. Liang et al. (2023) found detectors falsely flagged 61.3% of non-native English TOEFL essays as AI-written.
Can a university prove you used AI?
Not from the finished essay alone, for the same reason a professor can't. What actually shifts a case is the process evidence you can put on the table yourself, version history, dated drafts, notes, a research trail, since that's evidence you control and a bare score isn't.
What if I genuinely have no drafts saved?
Say so plainly rather than trying to reconstruct a history after the fact. A late-arriving 'draft' created after the accusation is easy to spot and does more damage than admitting you write in one sitting.
Is version history enough on its own?
It's strong, but it isn't unforgeable; someone can paste a full document into an editor and make incremental edits to simulate drafting. It's strongest paired with something independent, like a handwritten outline or a research trail with an earlier timestamp.
underslop edits drafts for voice. It does not check, score, or contest anything, and nothing on this page will change a result you have already been given. What it can do is show you which patterns in a piece of writing read as machine-written, using the same open lint the rest of the site runs on.