How to keep version history in Word, Google Docs and Overleaf

Word keeps version history only for files stored in OneDrive or SharePoint. Google Docs keeps it for every document, but only people with edit permission can see it. Overleaf's free plan keeps the last 24 hours plus any versions you have labelled, and nothing else.

All three have to be in place before you need them. If your file has been sitting on your desktop for six weeks there is nothing to retrieve, and the answer is to work with what you actually have rather than to manufacture something.

Nothing below is an argument that a version history settles a case. It is a record. What weight it carries is decided by your institution's academic integrity procedure, and those differ.

Two of those rows are the ones people get caught by. A Word file that never left your hard drive has no history at all. And on Overleaf's free tier, edits you did not label drop out of the panel after a day.

Microsoft's support page states it plainly: version history works for files in OneDrive or SharePoint, and nowhere else. A .docx in your Documents folder, on a USB stick, or in a Dropbox folder that Word does not treat as a Microsoft cloud location has no version history for Word to show you.

Getting to it, once the file is in the right place:

• Microsoft 365 (Windows): click the file title in the title bar, then Version history.
• Office 2016 to 2021: File > History. If History is not on the navigation pane, Microsoft's note says you may actually be on a subscription version, and to check under Info.
• macOS: click the document name in the title bar, then Browse Version History. Older Office for Mac: File > Browse Version History.
• Word on the web: File > Version History.

Selecting a version opens it in a separate window. That separate window is the export route. You are looking at a whole document as it stood at a given timestamp, and you can save it out under its own name without touching the live file.

Two limits will decide whether this is any use to you.

The first is how often a snapshot is taken. Microsoft's AutoSave page says that when you open a file and make your first change, that change creates a version, and after that "new versions are only added to the version history periodically (about every 10 minutes) for the remainder of your editing session." So the history is a series of ten-minute stills, not a keystroke log. A paragraph that took you four minutes to write and rewrite appears as one jump.

The second is how many are kept. The same support page notes that signing in with a personal Microsoft account gets you the last 25 versions, and that with a work or school account the number depends on the library configuration.

Microsoft documents those two numbers in different places and does not do the arithmetic for you. If they interact the way they look like they do, twenty-five ten-minute snapshots covers an afternoon, not a semester. So if a dissertation has been in progress since March, do not assume March is still in there. Open the panel, scroll to the bottom, and look at what the oldest date actually says. Then save the early versions out as separate files while they are still there.

AutoSave itself is a Microsoft 365 subscription feature in the desktop apps, and it switches on when the file is in OneDrive, OneDrive for Business or SharePoint Online. Without a subscription you can still get version history by editing the file in Word on the web with a free Microsoft account, since the file is in OneDrive either way.

Google Docs records revisions for every document with no plan requirement. Open a doc, click Last edit at the top right, and the panel lists earlier versions with who changed what.

The thing that catches people is not finding the history. It is showing it to somebody. Google's help page says it twice: "To browse earlier versions of a file, you need permission to edit that file", and then again, "If you don't have permission to edit a file, you won't be able to see the version history." Sharing the document as view-only, which is the instinct when you are being asked to prove something, is exactly the sharing mode that hides the history.

So there are two routes:

1. Give the reader edit access to the document. They open the panel themselves and click through the history. This is the hardest form to argue with, because they are operating the tool rather than looking at a screenshot you sent them.
2. Make a copy of an early version. In the version history panel, next to the version you want, More > Make a copy. Google's page describes this as making "a copy and edit earlier versions of a file". You then have a standalone document containing your draft as it stood on that date, which you can download as a .docx or PDF and attach to an email.

Two more things about the Docs panel change how you should use it.

Versions get merged. Google's page says outright that "the revisions for your file may occasionally be merged", and that naming a version is how you "make sure your versions aren't merged". You get up to 40 named versions per document. If you are working on something that matters, spending ten seconds a day on More > Name this version, with a date and a note about what you did, converts an automatic blur into a labelled sequence you chose.

And the owner can destroy it. The panel offers Delete history and Delete this and older versions, and Google's warning is that "Deleting version history is permanent and cannot be undone." There is no reason for anyone reading this article to touch either.

Overleaf's History panel opens from the icon above the preview pane, with All history and Labels as the two views.

The free plan limit is documented and specific: "Full Project History is a premium feature. For users on our free plan, the History panel will show edits made in the last 24 hours as well as any labeled versions." The premium features page puts the same thing the other way round, describing full history as viewing the entire project "versus only 24 hours of history availability on a free Overleaf account".

Read that first quotation again, because the second half of it is the part that pays. The free plan shows the last 24 hours "as well as any labeled versions". Labelling costs nothing, takes about five seconds, and is the difference between a project whose visible history rolls over every night and one with a dated spine you put there yourself.

To label: open History, find the version, three-dot menu, Label this version, type the label, Add label. You can attach multiple labels to a version. The live state of the project is labelled Current state.

To get a version out: same three-dot menu, Download this version. Overleaf's docs say "A .zip file containing your project files is generated and downloaded to your device." That zip is a self-contained artefact with a filename and a date, which is a better thing to attach to an email than a screenshot of a panel.

Three Overleaf specifics that bite:

• Copying a project resets its history. Overleaf's docs: "A copied project starts with a completely new, fresh history—the original project's history is not copied to the new project." If your instinct on being accused is to duplicate the project so you do not break anything, you have just created a project whose history begins today.
• Track changes is not on the free plan. Overleaf's own comparison table lists track changes as unavailable on Free and Personal, available on Student, Standard and Pro. So the collaboration trail a supervisor might expect to see does not exist on a free account.
• Git, GitHub and Dropbox integration are premium. They are all listed under premium features, so the "just keep it in Git" advice below applies to a local repository you run yourself, not to Overleaf's built-in sync.

Not verified: whether a free-plan user can still download a labelled version older than 24 hours. The documentation says the panel shows labelled versions and separately says restoring versions older than 24 hours needs the premium feature. It does not spell out download. Test it on a throwaway project rather than assuming.

If you write in LaTeX, Markdown or anything else that is plain text, a local Git repository is the cheapest writing record there is. Initialise a repository in the folder, commit at the end of each session with a message describing what you did, and the commit log gives you an ordered, timestamped list of every commit with your own notes attached.

One caveat, and it is the reason to push somewhere. A commit's author date is written by your own machine and can be set to anything. A push to GitHub or GitLab, on the other hand, is recorded by a third party at the time it happens. If you are going to use Git as a record, push it to a private remote regularly, so there is a server-side timestamp that is not under your control.

For everyone else, the low-tech version still does the job: save a dated copy at the end of each writing session, chapter3-2026-03-04.docx, and keep it somewhere other than the folder you are actively editing in. It is not a version history. It is a sequence of dated files, which is most of what a version history gives you anyway.

Version history records the document. A process tracker records the act of writing, including where text came from.

Grammarly Authorship runs inside Google Docs, through the browser extension, and inside Microsoft Word, through the desktop app. Its own support page is unusually specific about where it stops working, so these are quoted rather than paraphrased:

• It is off unless you turn it on. "By default, Authorship in Google Docs and Microsoft Word does not track your writing activity." Nothing about it is retroactive.
• The free plan can produce and share reports. All plans including Free can "Generate an unlimited number of Authorship reports" and "Share an unlimited number of Authorship reports", see "Typed by a Human" analytics, and see colour-coded text in the report. Paid plans add attribution of where pasted text came from.
• Sharing is a URL. From the report you click Share and get "a unique URL that you can copy and share with whoever you choose. Your audience will see the same information available in the report."
• It is per device. "If you move the document to a different device and write anything there, your previous writing will be attributed as 'Unknown' because the new device has no Authorship data stored on it." Laptop at home plus a library desktop breaks the record.
• There is a length ceiling in Word. "There is currently a 20-page limit for Authorship in Microsoft Word when using the desktop application for Windows and Mac. After this limit is exceeded, the tracked history may become unreliable." That rules out most theses.
• It is not exact. Grammarly says it "may not capture 100% of cases accurately", listing fast typing, pastes from untracked desktop sources, and pastes from AI tools it has not yet categorised.

On the institution's side of the fence there is Turnitin Authorship, which is a different product from AI writing detection and answers a different question. Turnitin's FAQ describes it as using "metadata as well as forensic language analysis to detect if the submitted assignment was written by someone other than the student", and adds that "It will not be able to indicate if it was AI written; only that the content is not the student's own work." You do not run that one. Your institution does.

It can show a document changing shape over time, with dates. That is all, and it is not nothing.

What it cannot do is overrule a score by itself, because nothing overrules anything by itself in an academic integrity process. Turnitin's own documentation is more restrained about its detector than most of the discussion around it. On the AI Writing Report page: the model "may not always be accurate (it may misidentify human-written, AI-generated, and AI-paraphrased text), so it should not be used as the sole basis for adverse actions against a student", followed immediately by "It takes further scrutiny and human judgment in conjunction with an organization's application of its specific academic policies to determine whether academic misconduct has occurred."

A separate Turnitin guidance page for educators says the score "is not meant to provide definitive answers in isolation", and that when educators use it "as a single data point rather than a definitive response, then it is being used as intended".

Your version history is another data point in that same sense. So are your notes, the sources on your desk, and whether you can talk about your own argument for ten minutes without a script.

One more passage from Turnitin's documentation should be in front of you, because it is the company naming conditions under which its own score misfires. Turnitin's FAQ says false positives "can include content without a lot of structural variation, text that literally repeats itself, or text that has been paraphrased without developing new ideas", and then tells the reader of the score what to do about it: "If our indicator shows a higher amount of AI writing in such text, we advise you to take that into consideration when looking at the percentage indicated."

And a practical wrinkle. You may not be able to see the report you are being asked to answer. Turnitin's FAQ says "The AI writing detection indicator and report are not visible to students", though it adds that "with the PDF download feature, instructors can download and share the AI report with students." So ask for the PDF. Our walkthrough of how to read a Turnitin AI writing report covers what you are looking at once you have it.

We cannot tell you your institution will accept a version history, and anyone who does is guessing on your behalf. Policies differ by university, by faculty, sometimes by module. What we can do is point at what institutions are telling their own staff.

The University of Colorado Boulder's Center for Teaching & Learning has a page on addressing academic misconduct which lists, under documentation an instructor should gather, "Version history of documents or reports from process-oriented transparency tools (e.g., Turnitin Clarity)". The same page tells instructors to "Avoid AI detection tools, which are unreliable and unsupported by OIT at CU."

USC's Office of Academic Integrity, in its guidance to faculty on generative AI, suggests instructors "Gently inquire about their writing process and sources", specifically "when they started the assignment, what was their drafting/editing process, what tools were used". Elsewhere on the same page, work that appears to "Lack evidence of the student's learning process (e.g., missing drafts, outlines, or other required elements that demonstrate building upon ideas)" is listed among the things that prompt a closer look.

Neither page is a promise, and neither is your institution's formal procedure. They are two named institutions telling staff that the writing process is the thing to examine. That is the category your version history falls into.

The journal side is different, and thinner. We could not find a major publisher policy that names version history, drafts or revision records as evidence a corresponding author can be asked to supply. What publishers do have is disclosure requirements. Elsevier's policy on generative AI in writing asks for a "Declaration of Generative AI and AI-assisted technologies in the writing process" placed before the reference list, and puts the accountability on the authors for the content of the work. That is a rule about telling them, not about proving anything to them. We have compared several of these in publisher AI disclosure policies.

If you are a corresponding author dealing with a research integrity query rather than a course-level one, the request will come from the journal or the publisher's ethics team and will say what it wants. Answer that request, not this article.

If you have already been flagged, do these today, in this order.

1. Get the artefacts out of the tools before anything else happens. For each document involved: Word, open each old version and save it as its own file. Google Docs, make a copy of the two or three earliest versions and download them. Overleaf, download the versions you can reach as zips. Put the results in a folder that is not the account you were working in, and do not edit them. The point is to stop depending on a panel staying the way it is.

2. Turn on what only works going forward. Move local Word files into OneDrive. Start labelling Overleaf versions at the end of every session. If you use Grammarly and it fits your document, switch Authorship tracking on. None of this helps with the piece you are being asked about. All of it means the next piece is not this conversation again.

3. Find out what you are actually answering. Ask which report the flag came from and ask for the PDF, since students cannot see the AI indicator themselves. Then find your institution's academic integrity procedure, the actual document, and read the part about evidence and appeals. What you send should answer that procedure's questions in that procedure's order. A folder of files with no covering explanation makes the reader do work they will not do. We have a walkthrough of putting a response together in flagged, but you wrote it yourself.

Do not build a writing trail after the fact. Retyping your finished paper into a new Google Doc so that the history shows it being typed produces a document created after the date of the allegation, with a first version that is already most of a paper, and with a session pattern nothing like real writing. It is the one action in this whole area that turns a contestable score into a straightforward integrity finding, and it is a worse position than having no history at all.

Do not delete version history to tidy things up. Google's own warning is that it is permanent and cannot be undone.

Do not duplicate a project or document hoping to preserve it. Overleaf documents that a copied project starts with a fresh history. Copies are for extracting a single version you want to send, not for safekeeping.

If your honest position is that you wrote it on a laptop with no cloud sync and have nothing, say that. It is a normal way to have written a paper, and it is an answer you can defend. An invented record is not.

Does a version history prove I wrote it? No. It shows a document changing over time under your account. It is one input among several, in the same way Turnitin's own guidance describes its score as "a single data point rather than a definitive response". Which inputs carry weight is your institution's decision, not a vendor's and not ours.

My tutor has view-only access to my Google Doc. Why can't they see the history? Because Google requires edit permission. Their help page says "To browse earlier versions of a file, you need permission to edit that file", and repeats it as "If you don't have permission to edit a file, you won't be able to see the version history." Change their access to editor, or send them a downloaded copy of an early version instead.

I have been on Overleaf free for a month. Is my early history gone? Your History panel shows the last 24 hours of edits plus any labelled versions. If you never labelled anything, unlabelled edits from three weeks ago are not in your panel on the free plan. Start labelling now for what comes next, and check whether your institution has an Overleaf licence, which would give you full history without paying: Overleaf publishes a list of subscribing institutions.

My Word file was on my desktop the whole time. Do I have anything? Not version history, and AutoRecover is not a substitute for it. What you may have: earlier copies you emailed to yourself or a supervisor, files in a backup, the reading and notes behind the draft, and the sources themselves. Those are real. Assemble them instead of trying to reconstruct something that was never recorded.

Can I export the whole version history as one file? Not really, and it varies. Overleaf gives you a zip per version. Word lets you open a version and save it out as a file. Google Docs lets you copy a version and download the copy, but we found no built-in way to export the timeline itself. So in practice "exporting your history" means producing a handful of dated files, one per point you want to show, plus a short note saying what each one is.

Vendor documentation for Microsoft, Google, Overleaf and Grammarly was checked on 15 August 2026, and Turnitin's guides on 14 August 2026. These products change their plan limits without notice, so verify anything here that your case depends on against the vendor's current page before you rely on it.