Can Google Docs detect AI? What it records instead

Google's own documentation for Docs lists the tools the editor contains: spelling and grammar checks, word count, translation, voice typing, version history and, for Workspace customers, a comparison tool. None scores text for AI. The version history help page describes what Docs does record: "You can see changes that have been made to a document", who was the last person to update it and when, and, for anyone with edit permission, every earlier version (Google Docs Editors Help, find what's changed in a file, read 9 November 2026).

Google Classroom's originality reports are the feature most often mistaken for an AI check. Google states that when a teacher turns them on, "the reports use Google Search to compare a student's Google Docs, Google Slides, and Microsoft Word files against webpages and books on the internet", and against previous submissions from the school if an administrator enables school matches (Google Classroom Help, how an originality report is created, read 9 November 2026). The report flags passages similar to a web page the student did not cite and links to the source.

Google also states the limits: Google Scholar is not searched, the report does not judge citation formatting, and the teacher decides the severity of any flagged passage, read 9 November 2026. On the Education Fundamentals edition a teacher can enable reports on 5 assignments per class, reports stay available for 45 days, and students can run up to 3 reports on their own work before turning it in, read 9 November 2026. Nothing in those pages mentions AI. A passage written by a chatbot that matches no web page will not be flagged, and a human-written passage that quotes a web page without citation will.

Schools that want AI detection in Classroom add a third-party tool. GPTZero lists Google Classroom and Canvas among its integrations (GPTZero for educators, read 9 November 2026). The same arrangement, platform plus licensed add-on, is how detection reaches Canvas, Blackboard and Moodle; the guide does Canvas detect AI explains that plumbing.

Version history is a timeline. Google states that a user with edit permission can view earlier versions, see who updated the file and what changed, restore a version, copy one, and name up to 40 versions per document so they are never merged, read 9 November 2026. For a student accused of using AI, that timeline shows the document growing: an outline on Tuesday, two paragraphs on Wednesday, a rewrite of the introduction on Friday. A single paste of 1,200 words at once shows up just as clearly.

Three limits apply, all stated by Google, read 9 November 2026. Only someone with edit access can see the history, so a student usually has to open it for the teacher or share the file. Google may merge revisions, so very small edits can be grouped; naming versions as you go prevents that. And an owner with a personal Google account can delete version history permanently, which means a clean history is evidence and a deleted one is a question.

This is the evidence GPTZero and Turnitin themselves point to. Turnitin states that its percentage "should not be used as the sole basis for action" and that it provides data for an educator's decision, not a finding of misconduct (Turnitin, AI writing detection FAQs, read 9 November 2026). A detector reads the final text once; version history shows how the text got there.

Two products record the writing process inside Docs, and both work by watching keystrokes and pastes rather than by scoring the finished text.

Grammarly Authorship is an opt-in feature on Grammarly Free and Pro that, once enabled through the browser extension, "automatically tracks your writing process in Microsoft Word, Google Docs, or the Grammarly Editor", categorising text as typed, pasted or AI-assisted, and produces a report the user alone decides to share (Grammarly Authorship, read 9 November 2026). Grammarly's page contrasts this with detectors, which "only guess from the final draft". GPTZero sells a Google Docs writing replay for educators that lets them "watch documents being written, view copy and pastes, and check for unnatural typing patterns" (GPTZero for educators, read 9 November 2026).

Both are process records, and both have the same caveat as version history: they show how text entered the document, not where it was composed. Text drafted in a chatbot and retyped by hand looks typed. The guide on what to do if you are flagged for AI sets out how to use any of these records in a dispute.

Open version history with them, in the document, and walk through it. Ask which detector produced the reading, the exact figure and the date; the guide to whether AI detectors work explains what those readings can and cannot show, including the 2023 finding that no detector tested reached 80 percent accuracy. If your course allowed AI assistance, show where you disclosed it.

No detector score should be the only evidence in a misconduct case. Google Docs gives both sides something better than a score: a dated record of the work. Keep it, name your versions, and never delete the history of a document you may have to defend. The rest of this series is in the guides hub.