How to Use Gemini to Organise and Search Google Shared Drives

The short answer: Gemini is excellent at finding and summarising what is already in your shared drives, and it will happily propose a cleaner structure. What it does not do is move and rename thousands of files for you — that part is an Apps Script or Drive API job, it is close to irreversible at scale, and it should never run without a dry run and a real backup behind it.

Every company that has been on Google Workspace for more than about three years has the same shared drive: forty top-level folders, six of them called some variant of Admin, a folder named New Folder (2) containing the only signed copy of a contract, and a naming convention that changed twice when the person who invented it left. Nobody can find anything, so people re-upload their own copies, and the sprawl compounds. Gemini genuinely helps with this — but which part of it Gemini helps with is the whole point of this article.

Why Shared Drives Decay Into Unsearchable Sprawl

Nobody owns the taxonomy. A folder structure is a shared decision that needs a maintainer, and it almost never has one. Every new project creates folders at whatever level the person who started it happened to be looking at.

Filenames encode context that only made sense that week. Final_v3_JL_updated tells you nothing eighteen months later, and there are ten thousand of them.

Search only works if you remember a word from inside the document. Drive's keyword search is good, but it cannot answer a question you can only phrase conceptually — the pricing approach we used for the Singapore retail client — unless those exact words appear in the file.

Duplication is the rational response to bad search. When finding the current version is slower than making a new one, people make a new one. That is not carelessness, it is a reasonable local decision that produces a terrible global outcome.

Departures leave orphaned structures. Files created in a personal My Drive and shared out disappear when the account is deleted. Shared drives fix this, which is why the migration off personal ownership is worth doing before anything else on this list.

What Gemini Can Do Across Drive

Semantic Search and Summarisation Versus Actually Moving Files

The genuinely strong capability, available today in Google Workspace with the right licensing, is retrieval and synthesis. You can ask the Gemini side panel or the Gemini app what our standard payment terms have been across client contracts this year, reference specific files, and get a synthesised answer with citations back to the documents. You can ask it to summarise a folder of proposals, extract every deadline mentioned across a set of files, or tell you which of four similar documents is the most recent substantive version. NotebookLM covers the same ground when you want a persistent, grounded workspace over a fixed set of sources rather than an ad-hoc question.

Critically, this all runs inside your existing permission model. Gemini surfaces only files the asking user already has access to, so it does not create a new exposure path on its own — though it does make existing over-sharing much more visible, which is usually a good thing and occasionally an uncomfortable one.

What Gemini does not do is bulk file operations. It will not move nine hundred files into a new folder tree or rename them to a convention. Ask it to and you will get a plan, not an action. That distinction is the single most useful thing to understand before starting this project.

Scripted Reorganisation With Apps Script or the Drive API — and Why It Needs a Dry Run

Doing the actual reorganisation means code: Apps Script for something quick and in-tenant, or the Drive API from a proper service if the volume is large. The useful division of labour is that the model classifies and proposes, and deterministic code executes. Feed Gemini an inventory — file name, path, owner, MIME type, last modified, last opened, size, and optionally a short extract — and ask it to propose a target folder and a normalised filename for each row, with a confidence score and a reason. Get that back as structured data, not prose.

Then treat the proposal as a change set to be reviewed, not a command to be run. Write it to a sheet. Sort by confidence. Have the person who actually owns that part of the business look at the low-confidence rows, because they are exactly the ambiguous ones where a wrong guess buries something important. Only then execute — in batches, logging every source and destination path so the operation can be reversed row by row.

A Safe Reorganisation Workflow — Inventory, Propose, Dry-Run, Apply

Inventory first, and look at it before doing anything else. Enumerate the drive via the Drive API into a spreadsheet. The inventory alone usually settles the strategy: if forty per cent of files have not been opened in two years, the answer is an archive folder, not a taxonomy.

Decide the target structure with humans. Ask Gemini to propose a structure from the inventory by all means, but the final tree is a business decision. Six to ten top-level folders that match how the company actually works beats a beautifully logical structure nobody recognises.

Archive before you reorganise. Move everything untouched for two-plus years into a dated archive folder in one operation. It is low-risk, reversible, and it typically removes half the problem before the hard part starts.

Generate the proposal as data. One row per file: current path, proposed path, proposed name, confidence, reason. Nothing executes at this stage.

Dry-run and review. Produce the diff — what moves where, what gets renamed, how many files land in each new folder. Spot-check the low-confidence rows and anything touching contracts, finance, or personnel. Have a second person review.

Apply in batches, logging everything. Start with one folder tree, verify it, then continue. Log the file ID, old parent and new parent for every move so you have a reversal path that does not depend on anybody's memory.

Communicate before, not after. Tell people the date, what changes, and what to do if something they need has moved. Most of the pain of a reorganisation is not technical.

Leave shortcuts where habits are strong. Drive shortcuts let a file live in one canonical place while still appearing where a team expects to find it. Used sparingly, they buy a lot of goodwill.

AI-Assisted Cleanup Versus a Manual Taxonomy Project

• Speed on volume — AI wins by an enormous margin. Classifying twenty thousand files by hand is a project nobody finishes; proposing twenty thousand classifications takes an afternoon and a modest API bill.
• Consistency of naming — AI wins clearly. A model applies the same convention to file eighteen thousand as to file one, which is more than can be said for a team of humans doing it in shifts.
• Understanding what the business actually cares about — Humans win, and it is not close. The model does not know that one badly named folder holds the only evidence for an ongoing dispute. That knowledge lives with people and has to be injected into the review step.
• Handling ambiguity — Humans win. The right response to a genuinely unclear file is to ask someone, and a model asked to classify will always produce a classification.
• Cost — AI-assisted is cheaper in labour by an order of magnitude, but it is not free: the review step is real work, and skipping it is what turns this from a cleanup into an incident.
• Risk profile — Manual is slow and safe; scripted is fast and, without a dry run, catastrophic. The workflow above exists to get the speed of one with the risk profile of the other.
• Durability — Neither approach survives without maintenance. A clean drive with no owner decays back within a year, which is why the naming convention and the owner matter more than the one-off cleanup.

The Irreversible-Mistake Problem

The genuinely dangerous part of this project is that Google Drive is a live system with no transaction boundary. Ten thousand moves are ten thousand independent operations, and there is no rollback button.

Permissions change when files move, even though links do not. This surprises people, so it is worth stating precisely: a Drive file keeps its ID when it is moved or renamed, so existing links keep working. What changes is inherited access. Moving a file from a personal My Drive into a shared drive transfers ownership to the shared drive and applies that drive's membership; moving between shared drives can silently remove access for people who had it through the old parent, or grant it to people who should not have it. A bulk move is therefore a bulk permission change, and it needs reviewing as one.

The trash is not an undo. Deleted items sit in trash for thirty days and then go permanently, and trash does not capture a move at all. If a script puts nine hundred files in the wrong place, nothing was deleted — so nothing is in the trash to restore.

Version history is not a backup. It preserves revisions of a document's content. It does not preserve folder structure, file location, ownership, or sharing state, which are exactly the things a reorganisation changes. Google Vault is a retention and eDiscovery tool, not a restore-my-structure tool.

A partially completed run is the worst outcome. A script that dies at file six thousand of ten thousand leaves a drive in a state that matches neither the old structure nor the new one. Batching and complete logging are what make that recoverable.

This is why a real, independent backup of the Drive data — one that can restore a point-in-time state including structure and sharing — is not optional insurance for this project. It is the thing that makes the project safe to attempt at all.

Getting This Right — Backup, Sharing Permissions, and When to Bring in IT

Three things to settle before the first script runs. First, backup: confirm you have a restorable copy taken before the change, and confirm it by actually restoring something from it. An untested backup is a belief, not a control. Second, permissions: run a sharing audit as part of the inventory, because a cleanup is the best opportunity you will get to fix files shared with anyone-with-the-link and external accounts that left two years ago. Third, ownership after the fact: name the person responsible for the naming convention and the structure, or you will be doing this again in eighteen months.

There is also a data-handling question if you send file content — not just metadata — to a model for classification. Within Google Workspace, Gemini processes your data under your Workspace terms rather than as public consumer input, but if you are calling an API from your own script, check which terms apply and record the decision. If you operate under Hong Kong's PDPO, Singapore's PDPA or the PRC's PIPL, the shared drive almost certainly contains personal data and the analysis is a processing activity like any other.

Brocent's cloud managed backup service exists for exactly this scenario — an independent, restorable copy of your Workspace data, including the structure and sharing state that Drive's own trash and version history do not protect. Our AI+ support practice builds the inventory, classification and dry-run tooling, and managed IT support handles the permission audit and the execution window. If your organisation is on Microsoft rather than Google, the equivalent work is covered in our guide to auto-tagging and organising SharePoint document libraries with Claude. Brocent has run managed IT engagements across Asia since our founding in Beijing in 2007, with headquarters in Singapore and a Hong Kong office since 2016.

Frequently Asked Questions

Can we undo a bulk AI file reorganisation?

Only if you planned for it. There is no rollback button in Drive, and the trash does not record moves. The two things that make a reorganisation reversible are a complete log of every file ID with its old and new parent, and an independent backup taken immediately before the run. With both, reversal is a script. With neither, it is a manual reconstruction from memory.

Does moving files break existing share links?

No — and this is the most common misconception. A Drive file keeps its ID, so its link survives moves and renames. What does change is inherited permission: the file takes on the access rules of its new parent. People who could open it through the old location may lose access, and people who should not see it may gain it. Treat a bulk move as a bulk permission change and audit it accordingly.

Can Gemini see files a user cannot?

No. Gemini operates within the requesting user's existing Drive permissions, so it will not surface a document that person could not already open. What it does do is make over-sharing visible — content that was technically accessible but practically buried now shows up in answers. That is a good reason to run a sharing audit alongside any AI rollout.

Is Drive version history a backup?

No. Version history preserves earlier revisions of a file's content. It does not preserve where the file lived, who owned it, or who it was shared with — which are precisely the things a reorganisation changes. Google Vault is a retention and legal-hold tool, not a structural restore. If you need to get back to yesterday's drive, you need a backup product.

Should we let AI rename files automatically?

Renaming is safer than moving, since the file ID and links are unaffected, but it still destroys information — the old name often encoded something. A good compromise is to apply the new convention while preserving the original name in the file description, so nothing is lost and search still finds the old string.

Where to Start

Do the inventory this week and do nothing else with it. Export the file list to a sheet and look at the age distribution, the duplicate names, and the sharing exposure. In most organisations that single spreadsheet reframes the problem — the answer turns out to be archive most of it, fix the sharing, and agree one naming convention, with the AI classification doing the tedious middle part. Then, before any script writes anything, confirm you can restore the drive to today's state. If you would rather have the backup, the audit and the reorganisation run as one controlled piece of work, get in touch.