# AI and Google Drive. URL: https://levystreet.com/integrations/google-drive/ Updated: 2026-10-19 An AI Google Drive integration guide for NZ: Gemini in Drive, Drive API scopes and quotas, the Drive MCP preview, and source-linked document Q&A design. ## Short answer An AI Google Drive integration lets a model answer questions from your Drive files and link to the exact document it used. For New Zealand teams on Google Workspace, Gemini in Drive and Gemini Notebook cover many needs natively. A custom build makes sense when answers must also cite sources outside Google, appear in another channel, or follow your own access and logging rules. - Gemini in Drive is listed from Business Standard (NZ$21 per user monthly), not Business Starter. - The drive.file scope is non-sensitive; drive.readonly is restricted and needs extra verification. - Google's Drive MCP server is a developer preview that inherits each user's permissions. - New Drive API quotas apply to Google Cloud projects created on or after 1 May 2026. - Drive change notifications carry no details, so you poll the changes list after each one. ## What Gemini already does in Drive (October 2026) Google has included Gemini in Workspace Business and Enterprise plans since January 2025. On [Google's NZ pricing page](https://workspace.google.com/pricing?hl=en_nz), as listed on 19 October 2026, Business Starter is NZ$10.50, Business Standard NZ$21 and Business Plus NZ$33 per user per month. "Gemini in Drive" (find, summarise and analyse files) appears from Business Standard; Starter's Gemini is limited mainly to Gmail. Gemini Notebook, renamed from NotebookLM on 16 July 2026, answers from a chosen set of sources with citations. A useful rule: for one team asking about a fixed set of documents, such as a policy manual and employment templates, a Gemini Notebook is faster than any build. For thousands of changing files, answers inside another system, or sources outside Google, build. Our [Gemini](/tools/gemini/) and [Gemini Notebook](/tools/notebooklm/) guides compare the native options. ## Integration routes and when to use each | Route | What it gives you | Use it when | |---|---|---| | Gemini in Drive and Gemini Notebook | Summaries and cited answers inside Workspace | Staff ask in Google's apps and the plan includes it | | Drive API v3 | Files, metadata, permissions and export of Google Docs to text | You need a custom index, metadata extraction or answers outside Google | | Changes feed and push notifications | `changes.list` from a saved start token; `changes.watch` alerts with no change details | You want an index that stays current without re-crawling | | Drive MCP server (developer preview) | A remote MCP endpoint per product that inherits the user's permissions; needs the Workspace Developer Preview Program | Prototyping an agent, not production | | Apps Script, n8n, Make or Zapier | Triggers and simple flows | Small automations inside one domain | ## Automation 1: source-linked document Q&A - **Inputs**: Files in one shared drive, such as "Operations manual", with file ID, name, owner, modified time and link. - **Indexing**: Export Google Docs to text and extract PDFs; chunk with file ID and heading. Keep the index current from a saved `changes.list` token. - **Model step**: Answer only from retrieved passages, with a link to the file and section after each claim. Where passages conflict, show both with their modified dates. - **Access control**: Index only a shared drive whose membership matches the audience, or check the asking user's access to each result before display. - **Human review**: Questions answered "no source found", and answers users mark unhelpful, go to the document owner weekly. - **Failure and recovery**: If the start token is lost or invalid, re-enumerate the shared drive and reconcile by file ID. Citations are what make this trustworthy, as covered in our lesson on [checking AI answers against sources](/learn/checking-ai-answers-against-sources/) and our [RAG development](/services/rag-development/) service. ## Automation 2: a contract register with approval 1. **Detect.** A new PDF in a "Signed contracts" folder appears in the changes feed. 2. **Extract.** The model returns counterparty, start date, end date, renewal notice period and value, each with a page reference. 3. **Validate.** Dates must parse, the end must follow the start, and the notice period must be a number. Missing values are flagged, never guessed. 4. **Approve.** Rows land on a "Pending" tab of a register Sheet with a link to the PDF; the contracts manager checks and moves them to "Approved". 5. **Write back.** Only approved rows drive renewal reminders. The source PDF is never edited. ## Scopes, permissions and approval design [Google's scope guide](https://developers.google.com/workspace/drive/api/guides/api-specific-auth) (updated 3 September 2026) is the main design input. `drive.file` is non-sensitive and covers only files a user opens with your app or picks through the Google Picker, which suits "ask about this document" tools. `drive.readonly` and `drive` are restricted: they need restricted-scope verification, are limited to certain app categories, and apps that store or transmit the data on servers must pass a security assessment. A service account with domain-wide delegation can act as any user within its granted scopes, so treat it as an administrator credential. The [Drive MCP server](https://developers.google.com/workspace/guides/configure-mcp-servers) inherits each user's permissions and is configured with `drive.readonly` and `drive.file`. Under the Privacy Act 2020, the Privacy Commissioner expects a privacy impact assessment before AI tools handle personal information, and you stay responsible for Drive content a provider processes for you. ## Costs and limits - **Quotas**: for Cloud projects created on or after 1 May 2026, the [Drive API](https://developers.google.com/workspace/drive/api/guides/limits) allows 1,000,000 quota units per minute per project and 325,000 per user. Google plans charges above a daily threshold of 400,000,000 units, with at least 90 days' notice. - **Errors**: over-quota calls return 403 "User rate limit exceeded" or 429; use truncated exponential backoff. - **Model cost, worked example**: 1,500 questions a month at 5,000 input and 400 output tokens on Claude Sonnet 5.5 (US$2 and US$10 per million tokens) is 1,500 × US$0.014 = US$21 a month. ## Where it goes wrong - **Old copies.** "Copy of Policy v3 FINAL" gets cited instead of the current file. Exclude archive folders and show modified dates in answers. - **Thin notifications.** Push alerts say only that something changed; your code must call `changes.list` and handle paging. - **Permission drift.** A file unshared yesterday is still in the index today unless you re-check access at answer time. - **Spreadsheets.** Exported Sheets lose formulas and context. For numbers, query the Sheet directly; see [analysing a spreadsheet with AI](/learn/analysing-a-spreadsheet-with-ai/). Most Drive projects sit inside a wider Workspace rollout; see our [Google Workspace guide](/integrations/google-workspace/). ## Questions people ask ### Can Gemini answer questions across all our Google Drive files? Gemini in Drive can find, summarise and analyse files a user can open, on Workspace plans that include it; Google's NZ comparison lists it from Business Standard. For a fixed set of documents with citations, Gemini Notebook is often the better tool. Neither will answer from systems outside Google, which is the usual reason to build a custom index. ### What is the difference between the drive.file and drive.readonly scopes? drive.file is non-sensitive and covers only files the user opens with your app or picks through the Google Picker. drive.readonly lets an app view and download every file the user can see, so Google classes it as restricted. Restricted scopes need additional verification, and apps that store or transmit that data on their servers must pass a security assessment. ### Can ChatGPT or Claude connect to Google Drive? Yes. ChatGPT's apps (formerly connectors) include a Google Drive app, and Anthropic's connector directory lists Google Drive for Claude. Both act with the signed-in user's access. In ChatGPT Business, apps are on by default and admins control availability, so decide your policy before staff connect company drives to personal accounts. ### Is our Drive data used to train Google's models? Google's Workspace privacy hub says customer content is not used for other customers and is not human reviewed or used for generative AI model training outside your domain without permission. That covers Gemini in Workspace and Gemini Notebook on qualifying editions. If a custom integration sends Drive content to another model provider, that provider's terms apply instead. ## Sources - [Google Workspace pricing (NZ)](https://workspace.google.com/pricing?hl=en_nz) - [NotebookLM is now Gemini Notebook](https://blog.google/innovation-and-ai/products/gemini-notebook/notebooklm-gemini-notebook/) - [Generative AI in Google Workspace Privacy Hub](https://support.google.com/a/answer/15706919?hl=en) - [Choose Google Drive API scopes](https://developers.google.com/workspace/drive/api/guides/api-specific-auth) - [Retrieve changes](https://developers.google.com/workspace/drive/api/guides/manage-changes) - [Usage limits (Drive API)](https://developers.google.com/workspace/drive/api/guides/limits) - [Configure the Google Workspace MCP servers](https://developers.google.com/workspace/guides/configure-mcp-servers) - [Connected apps in ChatGPT](https://help.openai.com/en/articles/11487775-connected-apps-in-chatgpt) - [Artificial intelligence and the Information Privacy Principles](https://www.privacy.org.nz/assets/New-order/Resources-/Publications/Guidance-resources/AI-Guidance-Resources-/AI-and-the-Information-Privacy-Principles.pdf) - [Working with third-party providers](https://www.privacy.org.nz/resources-and-learning/a-z-topics/working-with-third-party-providers/) - [Pricing](https://platform.claude.com/docs/en/about-claude/pricing) Published by Levy Street (https://levystreet.com), Wellington, New Zealand.