30 percent. That's the share of genuine professional texts that guides warn may be falsely flagged by AI detectors. The practical defence for anyone accused of machine writing is procedural: assemble a time-stamped audit trail, preserve research and drafts, capture the act of composing, and be ready to explain the thinking aloud. Together these elements turn a bare assertion of authorship into demonstrable chronology, the single strategy institutions now find most persuasive.

30 percent. That number explains why organisations are asking authors for process evidence rather than a simple attestation. When an automated classifier returns a suspicious verdict, the question that decides the outcome isn't whether the text looks human. The question is whether the author can document how the text came to be, step by step.

1. Collect a time-stamped audit trail

The immediate priority is a Time-stamped audit trail that shows incremental development. Cloud platforms such as Google Docs and Microsoft Word 365 record edit histories and can display a chain of small additions, deletions and rewrites. That pattern matters because generative systems typically surface as relatively large, clean insertions rather than a sequence of small edits. If your work lives in the cloud, export or screenshot the version history and any track changes metadata. If you use a local editor, preserve file system timestamps and any autosave or recovery files. Where available, include dated backups from OneDrive or Google Drive to demonstrate continuity.

First, prioritise immutability. Export the version history to PDF or take time-stamped screenshots. Second, keep multiple copies: one archive on cloud storage and one local copy. Third, annotate the exported history briefly, noting which file corresponds to which dated milestone in your drafting process.

Worked example: imagine a project that began with a one-page outline. Save that outline as "outline 2026-03-01", keep the Google Docs version history cover image, and export the first two revision snapshots. The assessor can then see an outline, an interim draft and the final submission in sequence, rather than a single stand-alone file with no provenance.

2. Preserve the research trail and bibliographic metadata

Next, gather the research materials that produced the argument. The guide recommends precise bibliographic records and preservation of sources.

Follow the University of Alaska Fairbanks guidance on references as a practical anchor: for books include author, title, place of publication, publisher and year; for journal articles include author, article title, journal title, volume, issue, date, pages and DOI or permalink; for web pages include author, title, URL and date accessed.

Save bookmarks, export browser history, and keep PDF copies of articles. Take screenshots of webpages with visible timestamps and retain annotated readings. If you logged search queries while researching, export that history too. Handwritten notes, photographed whiteboard diagrams and early outline files are especially persuasive because they show messy, non-final thinking that generative systems don't produce as part of a single, instantaneous output.

First, list every source you consulted and attach the bibliographic fields the University of Alaska Fairbanks recommends. Second, collate copies or screenshots so each citation links to a dated file. Third, keep any annotated PDFs or notes that show how a source informed a paragraph or claim.

Worked example: if paragraph three of your essay draws on a journal article, include the article PDF, the DOI, a screenshot showing the access date and the marked-up passage you used. That chain shows both source and use.

3. Capture the act of composing

Keystroke-level logs and visual replays provide the clearest behavioural trace of authorship. Some document platforms retain detailed edit logs; where possible, use them. The guide recommends browser extensions such as Draftback to visually replay writing in Google Docs, showing text being typed over time rather than appearing fully formed. Where keystroke logs aren't available, assemble a sequence of dated intermediate files with distinct filenames and preserved modification timestamps.

If you used collaborative tools, include comments, suggestions and replies between co-authors that show back-and-forth drafting. Where helpful, supplement these with screenshots or short screen recordings made while composing a section. These audiovisual artefacts aren't compulsory in every case, but they're potent when questions about process arise.

Mastheads

First, check whether your editor retains an edit log you can export. Second, if you used Google Docs, run a Draftback-style visualisation and export it as a file or screen recording. Third, save intermediate drafts with clear, dated filenames and keep the files in a folder that records modification times.

Worked example: if you saved drafts as "v1.docx", "v2.docx" and "v3.docx" with timestamps two days apart, an assessor can see how a section expanded and was revised, which is far harder to fake than a final file alone.

4. Prepare to explain the work verbally

Institutions increasingly ask for an oral explanation to verify mastery of the subject matter. Be ready to walk an assessor through your line of reasoning, explain why specific sources were cited, and justify stylistic choices such as metaphors, analogies and structural turns. Practice explaining how key paragraphs were developed, why particular sentences follow others, and how data or citations were interpreted. That granular exposition is difficult to produce convincingly if you didn't originate the material.

First, rehearse a short recorded explanation that ties each contested paragraph to the notes, sources and draft snapshots you collected. Second, prepare to demonstrate the evolution of a paragraph: where the idea began, which sources shaped it, and how it moved into the final wording. Third, if available, provide contemporaneous meeting notes or calendar entries that show you discussed the topic during the drafting period.

Worked example: record a five-minute walkthrough of two central sections. State which note sparked each section, which article provided the key fact, and why you chose the concluding sentence. Attach this recording to the dossier when you submit it to an assessor.

5. Build a rebuttal to the detector

AI detectors are tools with measurable limits. They can struggle with formal, non-native or unusual styles and with texts that show low variation in sentence length and structure, sometimes called low burstiness. When a detector flags your work, request the detailed report and the specific passages it identified as suspect. Where possible, obtain the score breakdown or methodology from the body that applied the tool. If a third-party service was used, ask for the output and the thresholds that produced the verdict.

Assemble comparative examples of your prior writing so assessors can see your stylistic fingerprint across documents. Point out features that plausibly trigger a false positive: non-native phrasing, dense technical terminology, or repeated structural patterns. Present your audit trail alongside the detector output so the assessor can compare the flagged passages with the version history and source notes.

First, demand transparency about the detection method and the threshold applied to your text. Second, prepare a file that juxtaposes the detector's highlighted passages with the draft history and source annotations. Third, include earlier work that shows the same stylistic habits.

Worked example: if a detector marks long, formal sentences as likely machine written, provide two earlier essays with similar sentence structure and the version histories that show those sentences evolved over time.

6. Corroborate with related outputs and attestations

Corroborating context strengthens the dossier. Include earlier assignments, blog posts, forum contributions or publicly dated writing in the same voice. If you collaborated with peers or mentors, ask them for contemporaneous attestations that describe your drafting process. Preserve meeting notes and calendar invites that show you were working on the subject across a period of weeks or months.

First, collect public, dated writing that matches your style. Second, request short attestation emails from collaborators that describe their interactions with you during the drafting process. Third, add calendar entries or meeting notes as additional time-stamped corroboration.

Worked example: a dated blog post that uses the same turn of phrase as your disputed paragraph helps prove you habitually write in that way. A short email from a supervisor confirming they reviewed an early draft is further corroboration.

7. Follow a calm, document-first dispute process

When you have the evidence, present it methodically. The guide recommends a calm, document-first approach: assemble the version history, research materials, keystroke or draft evidence, and an oral explanation; present them to the assessor or academic integrity office; and ask for a review that includes the detector output and the rationale for its verdict. If the institution has a formal appeals process, follow its deadlines and evidence rules. Where no clear policy exists, request a written explanation of the decision-making standard and the detector's role. Discuss detector accuracy with the assessor, because tools don't guarantee 100 percent precision and have documented false-flagging behaviour.

First, submit a single dossier containing all evidence and a short cover note that maps each disputed passage to supporting files. Second, ask for the detector report and the thresholds used. Third, follow the formal appeals timetable if one exists and keep copies of all correspondence.

Worked example: submit a dossier with the exported Google Docs history, the Draftback replay, annotated source PDFs, dated drafts and a three-minute recorded explanation. Ask the office to include the detector's report in the review file and to confirm the appeals timetable in writing.

In Short

1. Export and save version histories and track changes, with screenshots or PDFs of the timeline. 2. Preserve precise bibliographic metadata for every source using the University of Alaska Fairbanks fields as a template. 3. Capture the act of composing with Draftback visualisations, keystroke logs or dated intermediate files. 4. Record a short oral walkthrough that ties notes and sources to final paragraphs. 5. Request the detector output and present a side-by-side rebuttal with prior work and corroborating attestations.

Related Articles

Thirty percent is the figure that frames the risk. If a detector flags your work, present a single dossier: exported version histories or Draftback replays, dated intermediate drafts and annotated sources, plus a short recorded walkthrough, and attach the detector's report so an assessor can compare each flagged passage against your process.

This article was created with AI assistance.