A UC Davis student accused of using AI on a written assignment produced the one form of evidence that is almost impossible to argue with: a hand-written draft, dated, in the student's own notebook, written before the digital version was ever typed. The professor withdrew the charge. The case illustrates a pattern worth understanding, because most students do not have hand-written drafts, and the detector still flagged writing that a pen and paper had produced.
What the case shows
The core fact pattern is straightforward. A student typed up an essay from a hand-written draft. The typed version was submitted through Turnitin. Turnitin's AI indicator flagged a significant portion of the writing as likely AI-generated. The professor initiated an academic misconduct referral. The student produced the original notebook pages, showing the same sentences, dated across several study sessions, with visible revisions, crossings-out, and marginalia. The professor dropped the accusation.
UC Davis handles academic misconduct through the Office of Student Support and Judicial Affairs (OSSJA). When an instructor withdraws a referral before OSSJA opens a formal case, the matter closes without a finding on the student's record. That is what happened here. For UC-specific context on how these cases develop and close, see the UC Davis false-accusation pattern.
Why a hand-written draft is decisive evidence
AI detection tools operate on the finished text. They have no access to the writing process. When the process itself is documented on paper, dated, and physically predates the digital submission, the detector's statistical claim becomes untenable. A model cannot produce text that a student then hand-copies backwards in time.
The evidentiary weight comes from three things working together:
- Chronology: dates on the notebook pages that predate the typed submission.
- Continuity: the same sentences, same argument, same structure as the submitted essay.
- Process markers: crossings-out, arrows, marginal notes, and false starts that a finished AI output would not have.
None of these features can be back-generated from a Turnitin flag. That asymmetry is what makes hand-written process evidence so hard to dismiss.
Why the detector flagged the writing in the first place
This is the part most reporting on similar cases skips. The student did not use AI. The detector still flagged the essay. That is not a paradox: it is what the research on AI detectors predicts.
Peer-reviewed evaluations, including the 2023 Weber-Wulff study in the International Journal of Educational Integrity, found that no tested detector met a reliability threshold suitable for institutional decision-making. The Liang et al. Stanford paper in Patterns showed that carefully-worded, clear, or non-native English writing is disproportionately flagged. Formal academic prose, drafted slowly and edited for clarity (exactly the kind of writing a student produces when working from a hand-written draft), sits in the statistical zone detectors most often misclassify.
In other words: writing carefully, in longhand, then typing it up cleanly, is one of the writing processes most likely to trigger a false positive. The very care that produces academic writing is what the detector reads as AI.
If you do not have a hand-written notebook
Most students write directly in Google Docs, Microsoft Word, or Notion. That is fine. Those tools produce process evidence that functions the same way a notebook does, if you know where to look:
- Google Docs version history: shows every edit, every session, timestamped. Do not delete revisions before your hearing.
- Microsoft Word AutoRecover files and OneDrive version history: similar edit trails, especially when the file is stored in the cloud.
- Browser history: research sessions, library database access, sources consulted while writing.
- Reference notes, annotated PDFs, and outlines: anything that shows the argument developing over time.
The principle is the same as the notebook case. What matters is documented continuity between an early, messy version of your thinking and the final submitted text. Version history is the digital equivalent of crossings-out on paper.
What cases like this change
Each documented instance of a detector flagging demonstrably human writing weakens the argument that a Turnitin score is sufficient evidence of misconduct. The UC Davis case does not by itself settle policy. It does, however, sit alongside a growing body of published research and campus-level guidance (including 2023 memos from several UC teaching centers) advising faculty against relying on detectors as standalone evidence. Your procedural rights before a hearing include asking what standard of evidence the institution is applying and whether a human review took place before the referral.
If you are preparing a written response to an accusation, NotBot generates a personalized defense package that documents your specific writing process, cites the research on the detector that flagged you, and addresses your institution's policy language. If your case has already moved past the initial hearing, the appeal package covers the grounds most likely to succeed at that stage.
If the potential sanction is severe (suspension, expulsion, or consequences that affect visa status), consult an education law attorney before your hearing. Process evidence and research citations strengthen a defense; they do not replace legal counsel when the stakes are high.
Build your UC Davis defense package
A personalized response that documents your writing process and addresses the detector that flagged your work.
Get your defense package$49 one-time · Generated in 60 seconds