UCSB Professor Retracts AI Accusation After Google Docs History

A UC Santa Barbara professor withdrew an AI cheating accusation after the student produced the Google Docs revision history for the flagged assignment. The retraction, which the student described publicly, shows what actually resolves a detector-based accusation: not an argument about the score, but a record of the writing process the detector could not see.

What happened

The accusation followed a familiar pattern. The instructor received an AI detection flag on a submitted assignment, treated the score as sufficient evidence, and informed the student that the work would be referred for an academic integrity finding. The student responded by opening the Google Docs version history for the file and sharing it with the professor, along with a short written explanation of how the assignment had been drafted.

The revision history showed the document being built incrementally over multiple sessions: outlining, drafting, cutting, rewording, and polishing. The professor retracted the accusation. UC Santa Barbara's Student Conduct Code requires a preponderance of evidence to sustain a finding, and a detector score sitting next to a timestamped record of human drafting does not meet that standard. For the broader context on this case and others at the same campus, see our coverage of UCSB AI accusation retractions.

Why Google Docs revision history is decisive

AI detectors analyze the finished text. They have no access to how it was produced. Google Docs revision history is the opposite kind of evidence: it shows the sequence of edits, the time between saves, and the shape of the document at every stage. A student who drafted the assignment themselves will have a revision history that looks like drafting, with sentences being written, cut, rewritten, and reorganized over hours or days.

A paste from a chatbot looks nothing like that. It appears as one or two large insertions, often with little subsequent editing. When a reviewer sees a revision history that is incompatible with the paste-and-submit pattern, the detector score loses whatever persuasive weight it had.

Why the detector flagged a human-written assignment

AI detectors score on statistical features of finished text, primarily perplexity and burstiness. Writing that is clear, structured, and uses common academic vocabulary tends to produce low perplexity, which detectors associate with AI generation. Students who revise carefully, follow a rubric, or write in the formal register expected in college coursework are at elevated risk of a false flag for exactly that reason.

Peer-reviewed research has repeatedly shown that detectors misclassify human writing at rates high enough to make the score, on its own, unreliable as evidence. A 2023 study led by Debora Weber-Wulff tested fourteen detection tools and concluded that none performed consistently enough to be used as the sole basis for an academic integrity finding. A Stanford study by Liang and colleagues (published in Patterns, 2023) found detectors flagged writing from non-native English speakers at particularly elevated rates.

What UCSB policy requires before a finding

Under the UC system's Policy on Student Conduct and Discipline and the UCSB Student Conduct Code, a finding of academic misconduct requires evidence establishing the violation by a preponderance standard. Instructors are expected to consult with the Office of Student Conduct before issuing formal allegations, and the office reviews whether the evidence meets policy requirements.

A detector score is a statistical probability, not a factual finding. When it is the only evidence, and the student produces contrary process evidence, the record is not sufficient to sustain the allegation. That is the procedural reason retractions happen at this stage rather than at a hearing.

Evidence that closes these cases

Revision history is the headline artifact, but it is strongest when paired with other process evidence that corroborates the same timeline. The combinations that tend to resolve cases without a hearing include:

• Google Docs version history exported with timestamps and change-level detail visible
• Browser history for the drafting period showing research activity, library database access, and source lookups
• Handwritten notes or outlines photographed with visible dates
• Citation manager entries (Zotero, Mendeley, EndNote) with import timestamps matching the drafting window
• Messages with classmates or writing center appointments that reference the assignment while it was in progress

For a complete walkthrough of what to preserve and in what order, see our evidence collection guide. The first 48 hours matter: version history retention windows, browser history defaults, and session logs all shorten with time.

If this is you at UCSB or a similar institution

The pattern that produced the UCSB retraction is reproducible. The specific steps that tend to work:

1. Request, in writing, the specific detector used, the score reported, and any human review notes attached to the referral.
2. Preserve the Google Docs revision history before touching the file. Export screenshots and download the document as .docx and PDF.
3. Collect secondary process evidence (browser history, notes, citation manager entries, messages) with timestamps.
4. Write a short cover note that explains how the assignment was produced and points to the specific artifacts that corroborate the timeline.
5. Review the procedural rights FAQ to confirm what you are entitled to request before any meeting or hearing.

If you are preparing a written response, NotBot generates a personalized defense package that addresses the specific detector that flagged you, organizes your version history and other process evidence, and reflects the policy your institution actually applies. If a finding has already been issued, the appeal package covers the procedural grounds that matter at that stage.