An AI-detector score is not authorship proof. Preserve your original materials, read the policy that applies to your case, and ask what evidence exists besides the score.
Start with the institution's own process
The University of Melbourne's staff guidance says a detector result alone is not sufficient evidence for a misconduct allegation. That does not decide another institution's process, but it is a primary example of a university requiring more than one automated score.
Turnitin's own AI Writing Report guide warns that its model may misidentify human-written, AI-generated, and AI-paraphrased text. The guide also suppresses exact scores below 20 percent because results in that range have a higher incidence of false positives. These are limits of that product, not a measured accuracy rate for every detector.
Gather evidence without altering the original
- Preserve the original file. Work on a copy rather than re-saving the only copy.
- Collect process records you already control. Drafts, notes, source annotations, dated messages, and the version history of the system where you wrote may provide context.
- Keep each artifact's limits attached. A Draft Record report describes markers in one DOCX package. It does not prove authorship, chronology, authenticity, or absence of assistance.
- Describe permitted tools accurately. Follow the policy that applies to spelling, grammar, translation, dictation, accessibility, or other assistance.
Ask focused questions
Ask which policy and evidence apply, whether human review occurred, and how you can explain your process. Do not upload confidential work to additional detector services merely to collect more scores.
No detector-evasion service
Draft Record does not provide rewriting, “humanising,” or detector-evasion features. Its purpose is to make a narrow file observation legible while keeping the observation inside its evidential limits.