Classroom and Canvas Describe No Paste Alert, but a Google Docs History Can Show a Block Arriving at Once

Not as an alert in anything these pages describe, though Google Docs keeps enough detail to show one after the fact. Google Classroom's originality reports compare a submission against web pages and books, and the help page describes no AI writing detection at all ¹. Instructure describes the AI writing report inside Canvas as part of the Turnitin Advantage Integration ².

Traces do exist, and separating them is most of the answer.

None of those rows is a paste alarm. The Google Docs row gets closest after the fact. Authorship gets closest while the writing is happening.

Length matters too. Turnitin raised its own minimum from 150 words to 300 words in May 2023 ¹⁰; under that rule a submission below 300 words was not evaluated.

The rumor lasts because the honest answer changes by platform. Picture a student on a school Chromebook writing in Docs, which leaves a version history that anyone with edit access can browse ³ or replay with an extension ¹². Now put the same student in Word on a home laptop, uploading the finished file at the end, and what's left is a timestamp and a file with no version history behind it. Both of them may have written every word themselves.

Earlier states of the file, each stamped with a time and an account. Only people who can edit a Google Doc can browse that record, so a teacher with view-only access can't open it, though the history is still there ³. Microsoft keeps it only for files stored in OneDrive or SharePoint. A document written offline in Word has none to show ⁴.

Two details decide how much survives. Over time, unnamed versions get folded into their neighbors, while a named version stays separate, and a document holds up to 40 of them ³. So how much survives is partly the owner's call: the owner of a Google Doc can delete unnamed history, and Google says that deletion is permanent and cannot be undone ³.

It can't settle authorship, though. A 2026 framework in Frontiers in Artificial Intelligence looked at assessment built to survive generative AI. It says process documentation should be read as evidence of engagement rather than proof of independent authorship, because process records are not immune to fabrication ⁹. A record shows that writing happened over time. It doesn't show whose thinking it was.

A paragraph that appears whole between two saves looks like a paste, and a replay extension will list it as one ¹². It looks just as much like a student who drafted in a notes app and moved the paragraph over, which is a normal way to write.

What a Docs history can and can't establish is its own question, worked through in what a Google Docs history proves.

Two kinds. One runs while the writing happens, and the other reads a Google Doc's history afterwards. Grammarly Authorship states that "it can distinguish between text typed directly into your document, text pasted from a browser-based source, and text pasted from an unknown source, such as a private browsing window" ⁵. Its record covers only what is written while it is on ⁵. It marks where pasted text came from ⁵, but not why the text was pasted or who wrote it first.

The second kind works from a finished Google Doc. Revision History, a Chrome extension that works on Google Docs and Slides, says it can "Identify all copy/paste events with timestamps" and "Distinguish between content pasted from external sources vs. within the document" ¹². Like Authorship, it can't say why text arrived in one piece.

Grammarly is unusually blunt about detection in general. Its support pages state that "none today can reliably and conclusively determine whether AI was used to produce text" ⁶. The same company offers a free AI detector of its own ¹¹.

Grammarly also shows up on the other side of this question, when a student's own editing gets read as machine writing. That case is covered in why Grammarly use gets flagged.

Often a percentage at the top, with highlighted sentences underneath. In June 2023 Turnitin put its sentence-level false positive rate at around 4 percent, which it explained as a 4 percent likelihood that a sentence highlighted as AI-written was written by a person, or 4 in every 100 highlights ⁷. The document-level figure it quoted then was under 1 percent, and it held only for documents read as 20 percent or more AI writing ⁷.

The instruction on that same page is one sentence: "use the information to initiate a conversation, not to draw a conclusion" ⁷. Turnitin also added an asterisk to its AI percentages under 20 percent, a range it called less reliable ¹⁰.

Nature reported in 2026 that Pangram's chief executive said a paragraph Pangram judges human in isolation could switch to AI when combined with other text in a segment ⁸. Neither verdict records who typed the paragraph.

None of this reads keystrokes. The detector readings described here are made from the finished text. They can't tell the order the words arrived in or the window they were typed into, and they can't tell whether anything was pasted.

Human publishes a false-flag figure of its own at human.olive.is, as of September 2026: 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%. It comes from Human's own in-house measurement on a held-out set, not an independent audit. It was measured on college-level academic writing, so applying it to any other kind of writing is an assumption. This is an estimate from our detector. A flag should be reason to look closer and not verified proof of cheating. Like every reading above, it describes the text a detector was given rather than the student who handed it in.

Where that column comes from inside a course site is a separate question, answered in what the Canvas AI column actually is. A student who wants to read a draft the way a detector will, before handing it in, can use a free detector a student can run.

Write where the history survives, and name the versions. In Google Docs a named version stays separate instead of being folded into its neighbors ³, and the document's owner can delete unnamed history permanently ³. Naming a version at each real stage of the draft costs nothing and answers a question no percentage can.

Word on a local drive keeps none of this, so a student drafting offline is drafting without a record ⁴. Saving into OneDrive or SharePoint turns it back on.

Say what was used, and say it in the paper. One sentence naming the tool and the stage of the work it touched turns a suspicion into a disclosed fact. That's cheaper than any argument that starts with a highlighted paragraph. Course policies differ on what's permitted, so the syllabus settles what the sentence has to cover, and how to write an AI disclosure statement sets out the wording.

A teacher who opens that conversation is rarely asking about the clipboard. The question is how the paper got written, and a student who kept a record can answer it.