Your best evidence
already exists.

What evidence actually stands? — the short answer

The rule behind the chart

Independent beats self-made. Anything an outside system recorded on its own is strong. So is your own documented past. Anything only you could produce — and can now fake — is weak by itself.

Everything below is an application of this one line.

The strong column — independent evidence, in depth

Third-party timestamps — your single best exhibit

An outside system recording when you worked is the hardest evidence to wave away — because you couldn't have authored it. A reviewer can dismiss your drafts as self-serving. They can't say you forged your university's own submission log. What counts:

• LMS access and submission logs (Canvas, Blackboard, Moodle) — when you opened the assignment, when you submitted, sometimes how long the page was active.
• Sent-email timestamps — a draft you emailed yourself or a classmate, or a question you sent the professor mid-assignment.
• Cloud-sync timestamps — Google Drive, OneDrive, or Dropbox automatic save times. They sit on the provider's servers, not your laptop.

The move: name the system, not just the time. "The Canvas log shows I opened this nine days before the due date and saved twelve times." That's independent and checkable. Can't reach the logs yourself? Ask your school, in writing, to pull its own — they belong to the institution. Asking puts the burden of checking on the side that made the accusation.

Your prior-writing baseline — why a consistent voice protects you

A documented, consistent voice over time is a defense that predates the accusation. Reviewers in real cases have compared the accused work against the student's other writing — in Yang v. Neprash, faculty set the paper side by side with AI output and with his prior work. So if this paper reads like the three you wrote last semester — same formality, same tics, same level — the "this isn't your voice" theory collapses.

How to build and use it:

• Gather graded work from earlier courses, forum posts, peer reviews — anything with your name and a date on it.
• Show the through-line: the formal register or big vocabulary the tool penalized is simply how you've always written.
• A non-native or neurodivergent writer? This is your strongest single move. The "too clean / too formal" pattern the tool reacts to is your baseline, not a tell → Accused while writing in a second language →

The same checker run on your own pre-AI writing — turning the tool around

Run an AI checker on your old work from before these tools existed. If it flags that too, the tool is indicting your style, not your honesty. This is one of the few moves that attacks the accusation's foundation, not just this paper. A 2019 essay cannot have been written by a 2023 model. So a "high AI" reading on it is a self-administered demonstration: the instrument is unreliable on your writing specifically.

The limit, stated honestly: this is strong support, not a standalone win. Pair it with your baseline and timestamps. Why a checker's number can't carry a case on its own — the studies, and the vendors' own admission → Why a checker's number isn't proof →

The self-made column — when it holds, and when it turns on you

Genuine drafts and the messy path — why it's hard to fake

A real first draft is hard to fabricate — and that difficulty is exactly what makes it persuasive. False starts. A thesis that changed halfway. A paragraph you cut, a source you added late. That mess is convincing because faking it would mean knowing the finished paper before you wrote the middle. Real writing can't know that yet. So don't tidy your drafts up. The incompleteness is the proof. (The full mechanism, and how to build this on purpose, lives on the prevention page → How to keep a record before you're ever asked →)

Present them as a sequence, not a pile: "Here's the outline. Here's draft one, where the argument was different. Here's where I cut the third section." The exact sentences that make each exhibit land in a hearing → How to prepare for the hearing →

"Does version history prove I wrote it?" — the honest answer

This is the most-asked question, and the answer is not the simple yes most people expect. Version history can be your best self-made exhibit — or the thing that sinks you. The difference is brutal.

How it actually works. Google Docs records edits continuously and stores them on Google's servers. That makes the timeline harder to alter after the fact than a folder of manually saved Word files. Word's "version history" exists only if the file lived in OneDrive or SharePoint with versioning on. A local .docx mostly carries last-saved metadata, which is weak. So where you wrote changes how much the history is worth.

Why it can backfire — two ways:

The rule: bring revision history only if it's thick and genuinely shows the work. Even then, lead with your independent evidence — timestamps, baseline — not the history. If yours is thin, reach for the independent column instead. And never paste a finished paper in all at once. Write in the document, so the process exists at all.

What the cases say each piece of evidence is worth

The outcomes split on one line — a flag alone, or a flag plus something independent. And it's the evidence, not the argument, that decided each. Three worth knowing:

• A tool's number standing alone → annulled. In Matter of Newby v. Adelphi, one tool read a first-year's paper as 100% AI. Two others read it as human. The court threw the finding out as "without valid basis and devoid of reason." Nothing independent backed the number.
• A flag + a fabricated source + a thin history → the school prevailed in court. In Harris v. Adams, the paper cited a book that does not exist, and the edit history showed under an hour of work. The independent corroboration is what carried it.
• A flag + a side-by-side faculty comparison → the finding stood. In Yang v. Neprash, faculty compared the paper with AI output and with his prior writing. That human analysis — not just a tool — supported the finding, and the district court declined to second-guess it.

The pattern for you: don't try to out-argue the flag. Make sure the only thing on the table is the flag. The full record, who won, who lost, and why → The cases — coming soon

Special situations — when the usual evidence isn't available

"I wrote it in an exam box / a test field — nothing was saved"

This is the hardest position, and it's common — so lean entirely on independent corroboration. There's no draft trail. Build the case around what did leave a trace: the LMS log showing when you were in the test, and for how long. Any scratch paper, bluebook, or notes you brought. The consistency between this writing and your baseline. Anyone who proctored or saw you. And run the checker on your own past work, to show the instrument misreads your style. Be honest that this is thin — which is exactly why writing where a record is kept matters next time.

"My laptop was repaired / wiped and the timestamps are gone"

Device loss doesn't erase the independent record — go to the systems that never lived on your laptop. Cloud backups. Google account activity. Email drafts you sent yourself. The LMS autosave. Your phone's notes-app sync. The whole point of independent evidence is that it survives your hardware.

"I only have the final file"

You're not finished — and you must not build a fake history. Reach for the independent column: prior writing, the sources you actually used and can produce, classmates or tutors who saw you work, any messages about the assignment. A real but incomplete case beats a fabricated complete one. Every time.

What backfires — do not do these

• Don't manufacture or "clean up" a history. Re-running your draft through a humanizer, or building a fake edit trail, is a separate and heavier charge. Concealment after the fact is the independent corroboration the school was missing. Several companies sell both the checker and the "humanizer" built to beat it (Duey.ai · Undetectable) — so graders know the tells.
• Don't volunteer a thin revision history. Silence on it beats handing over evidence that argues against you.
• Don't over-explain in writing. A long written defense becomes part of the record and can be turned around. Move the substance to a meeting, where you show the work → What do I send back? →
• Don't fight a checker with another checker as your main move. "Six tools, six numbers" shows the tools are unreliable — useful. But it's a point about the instrument, not proof of your authorship.

Preserve everything — now, before it's gone

Evidence that shows process doesn't survive delay. Today, before any meeting:

• Turn nothing off and delete nothing — leave edit history on, keep every draft, outline, and note.
• Pull third-party timestamps into your own hands — download LMS logs, cloud version data, and sent emails, off your school account, before access can change.
• Assemble your baseline — gather older graded work that shows your voice over time.
• Run a checker on your own pre-AI writing and save the result.
• Write down the timeline — when you worked, where, and what you used, while it's fresh.

How to actually pull each one — do this tonight

Knowing what to save is useless if you don't know where it lives. The concrete steps:

• Google Docs version history — open the doc → File → Version history → See version history. Don't just screenshot it. Name the key points (right-click a timestamp → Name this version) so the progression is labeled. Then File → Download a copy, and File → Make a copy into a folder you control — the history travels with the copy.
• Word — a file kept in OneDrive/SharePoint has ⋯ → Version history online. A local-only .docx won't — capture the file's Created and Modified dates instead (right-click → Properties).
• LMS logs (Canvas / Blackboard / Moodle) — your open and submission times live in the course Access Report or the assignment's submission details. Can't see them? Ask the integrity or conduct office, in writing, to pull the log. It's their record — and asking puts the checking on the side that accused you.
• Sent email / cloud sync — search your sent mail for the assignment and any drafts you mailed yourself, and export them. Drive, OneDrive, and Dropbox each show a file's activity or revision history in the right-click menu.
• Save it where the school can't reach it — export everything to a personal device and a personal email, not your school account. A finding can change your account access overnight. The record you can't reach is the record you don't have.

One honest check before you screenshot anything: a thin history — a paper that appears in one sitting — reads against you. Look at what yours actually shows before you decide to lead with it. The chart above tells you which way each piece cuts.

You've gathered it — now what?

Gathering the evidence is half the job. The other half is using it without tripping over yourself in the room. Line up two things before any meeting:

• What to actually say — the exact sentence that makes each exhibit land, and the order to present them (lead with the independent ones) → How to prepare for the hearing →
• How the process works — who carries the burden, what a hearing actually decides, and when an advisor or attorney belongs in the room → What happens at the hearing →

You're not walking in to argue you're innocent. You're walking in to show a record and let it speak.

The hard part this page exposes

Read back up the chart and one pattern falls out: everything that holds up was either made by an outside system, or already existed before the accusation. Everything you can assemble afterward is the weak column. That's why it's weak. The strongest evidence isn't gathered once you're accused. It's a continuous, contemporaneous record of how the work was built — in plain words, a record made as you wrote, already there before anyone asks. The one thing you can't reconstruct under pressure.