A Google Docs History Supports a Student's Account; Share It With Edit Access

The panel opens from the top right of the document. Google's help page on finding what changed in a file describes clicking Last edit, which opens a timestamped list of versions down the right side ¹. Each entry is stamped with a date and time, and names whoever edited. Clicking one shows the document as it stood at that moment.

Two habits make that panel worth more later, and both take seconds. The first is naming a version at the end of each writing session. Google's own description says unnamed revisions may occasionally be merged, so a long evening can collapse into one entry that looks like a single arrival ¹. Google states that a named version stops versions being merged ¹. A document holds up to 40 named versions, and a spreadsheet up to 15 ¹. Naming costs nothing, and it is the one thing that fixes an entry in place.

The second habit is copying an early draft before anything else happens. The same help page describes making a copy of an earlier version from the menu beside it, saved under a name and a location of its own ¹. A copy of the Tuesday draft sits in Drive as a separate file. Google is blunt about the risk to the original: deleting version history is permanent and cannot be undone ¹. The copy survives it.

Because the history rides on edit permission. Google states that browsing earlier versions of a file requires permission to edit that file ¹. A document shared as Viewer or Commenter hands over the finished essay and nothing behind it. That's how a student with a complete record can still turn up to a meeting with nothing to show.

The fix takes a minute. Open Share, add the teacher's school address as an Editor, and write one line saying the access is for the version history rather than for the text. Access can be pulled back the same evening.

If the school's accounts block outside sharing, or a teacher would rather not hold edit rights on a student's file, bring a laptop. The panel works as well at the meeting, and scrolling it in front of a person gets the record into the room faster than any link.

How a teacher reads the same panel names the entries a teacher looks at and what each one is taken to mean, and it's worth reading before the meeting.

Work spread across days, full of cuts and rewrites. Paragraphs move. A thesis changes in the middle. Sections vanish and come back a week later. There's no single correct shape, though. One essay might be a single sitting revised a few times across a week. Another might leave a handful of ragged entries across a phone, a library machine and a laptop, with almost nothing deleted between them.

In 2023 a UC Davis senior was referred for academic dishonesty after his professor ran the exam through AI detection software, GPTZero among it ³. He answered with an edit history showing the exam paper built across two or three hours ², alongside research on the detector's false flags ³. About a month later the university dropped the case, its letter citing conversations with him and his instructor and its own research into AI-text indicators ³.

The panel records versions: when text arrived, how it changed, and which account made the change, nothing more. So the claim it supports is narrower than it sounds, and still useful: this text was composed in this document over this stretch of days.

Say the limit out loud, or a teacher will say it first. A 2026 framework in Frontiers in Artificial Intelligence asks assessment designers for drafts and revision logs. It also writes that "process documentation should be interpreted as evidence of engagement, not proof of independent authorship, since process artifacts are not immune to fabrication" ⁴.

Support is the right word for that panel. Proof is not.

Then the panel shows one arrival. Better to say so first than to be asked. A paper written in a phone's notes app, on paper, or in Word and pasted into Docs produces a single large entry. Microsoft states that version history in Microsoft 365 works only for files stored in OneDrive or SharePoint ⁵, so a Word file kept outside those has no Microsoft 365 version history to hand over.

Everything else with a timestamp stands in for the curve. The notes file with its own edit dates. A photograph of the legal pad. The library database's access log. The messages to a roommate at 11pm arguing about the second body paragraph. No single piece is conclusive, but together they describe a week that one paste cannot.

Grammarly Authorship is one tool that keeps a record of where text came from. Grammarly states that it can distinguish text typed directly into a document, text pasted from a browser-based source, and text pasted from an unknown source such as a private browsing window ⁶.

Grammarly also states that its paid tiers add AI detection to Authorship reports, including on text that was not pasted in ⁶. The Docs panel draws no such line: a generated paragraph typed out by hand arrives there like any other typing. What a teacher can actually see when text is pasted in covers the same mechanics from the other side of the desk.

A single paste raises a question about where, not an answer about who.

An account of the argument, and a willingness to talk it through unprepared. Harvard College's guidance to faculty lists oral presentations as one way to evaluate comprehension, letting students explain their process, including AI application ⁷. Once a teacher can hear the reasoning behind a paragraph, the panel can stop mattering halfway through.

In a 10-minute meeting, three things travel well:

• The version history open on a laptop, 2 or 3 entries already picked out.
• One earlier copy, saved as its own document, showing a paragraph that later changed.
• A short account of where the sources came from, including any AI use the course allows.

A second reading on the text is a different instrument. Frame it as modestly as the panel. A student worried about how the finished essay reads can paste up to 1,200 words of it into Human at human.olive.is, one paper a day, with no account. Human's detector scores the text in memory and de-identifies any personal information for additional training. Human is tuned for college-level academic writing, reports how much of a piece reads as machine-written or machine-edited, and declines to answer at all below 50 words.

Human's own in-house measurement, as of September 2026 and on its own held-out sets rather than an outside audit, is 0 wrongly flagged by Human's default model of 600 human-written papers it had never been trained on; the ceiling on that is 0.498%. This is an estimate from our detector. A flag should be reason to look closer and not verified proof of cheating. In the meeting, a reading like that sits beside the version history as one more thing to talk through.

For the next paper, the loop open today is to check a draft, revise the sentences the detector flags, and check again. A writing tool in private preview reads a draft with the same detector, colors each sentence by what the detector reads it as, keeps quotations, numbers, citations and anything the writer marks as their own untouched, and helps rewrite the rest in the writer's own words; it never tells a person that their own writing is not theirs.

What to say to a professor after an AI accusation covers the conversation the panel sits inside.

The sentence that introduces the history can be plain. Here's the document and the week it was written across, and here's where the drafting happened before it landed in Docs.