Anthropic opened Claude for Google Workspace to a public beta on 6 October 2026, putting a Claude sidebar inside Google Docs, Sheets and Slides. It is available on the Pro, Max, Team and Enterprise plans and edits the open file directly. Google cuts off any single operation after six minutes.
What Claude Can Do in Docs, Sheets and Slides
The add-on reads the file a user already has open and makes changes inside it, rather than returning text to copy and paste. One installation from the Google Workspace Marketplace covers all three applications, and the sidebar offers the same models, connectors and skills as the main Claude app.
Anthropic’s documentation describes a different set of jobs in each application.
| Application | What Claude does |
|---|---|
| Google Docs | Answers questions about the document with citations, edits in place, rewrites and restructures long documents, converts text into tables, checks for undefined terms, and fixes heading style and numbering. |
| Google Sheets | Explains formulas and flags hardcoded values, writes and fixes formulas, builds models on new tabs, normalises data formats, troubleshoots lookup errors, and adds tables and charts. |
| Google Slides | Drafts slides from notes, splits dense slides, makes titles consistent, adds charts matched to the deck’s colours, and restyles decks. |
Selecting text, a cell range or a slide first narrows Claude to that part of the file. Claude can also answer questions without editing anything.
The Two Edit Modes, and What Neither Will Do Alone
Claude runs in one of two modes, and the default is the cautious one. In “Ask before edits”, Claude describes each planned action in plain language and waits for the user to click “Allow”. In “Accept all edits”, it carries out ordinary content edits without pausing.
Three actions always require approval, in either mode, and Anthropic’s documentation states there is no always-allow option for them:
- Inserting an image from a web address: approval is needed every time.
- Adding a link: approval is needed every time.
- Removing anyone’s access to the file: approval is needed every time.
A further set of actions is refused outright, even if asked for. Claude cannot share files, grant anyone access, change ownership, create menus or scheduled automations, or load arbitrary web pages into a file. In Docs, larger rewrites can also be returned as suggestion cards that the user accepts or rejects one passage at a time.
Anthropic says the add-on treats document content as data rather than as instructions, labels hidden text it finds, and is designed not to act on a document that tells it to share the file, fetch a link or contact someone. The guidance still advises reviewing proposed actions before approving them, particularly in files that came from outside the organisation.
The Three Limits That Will Stop a Task
Most of the practical friction in this beta comes from three constraints, none of which produces an obvious error message.
- Six minutes: Anthropic’s documentation states that “a single long operation can run for up to six minutes before Google stops it“. The limit is Google’s, not Anthropic’s, and the suggested workaround is to ask Claude to work in stages.
- Firefox: the browser is not supported in this beta. Anthropic says the sidebar loads but actions do not complete, and directs users to Chrome, Edge or Safari instead.
- Closing the sidebar: shutting the panel stops whatever Claude is doing. It has to stay open until the task finishes.
Two further configuration limits apply. Sign-in must use a Claude account: API keys and cloud-provider sign-in do not work. And third-party inference through Amazon Bedrock, Google Cloud Vertex AI, Azure AI Foundry or an LLM gateway is not supported in this beta, which rules the add-on out for organisations that route Claude through one of those routes.
Attachments are capped at 20 files per message, 30MB per document and 10MB per image. Web search is off by default.
What Claude Cannot Do in Each Application
The per-application restrictions are the ones most likely to interrupt real work, and they differ sharply between the three apps. Across all of them, Claude works only on the file it was opened in: it cannot open, create or copy other files in Drive, and it cannot export the file as a PDF or download it. Voice dictation is not available in the sidebar, and the “work across apps” capability is a Microsoft 365 feature that does not exist in the Google version.
Beyond that, each application has its own gaps.
- Docs comments: Claude cannot read, reply to or resolve comments, and cannot accept or reject existing suggestions. Its own changes land as direct edits rather than as tracked suggestions.
- Docs tabs: Claude cannot create, rename or reorder document tabs.
- Connected Sheets: tabs backed by live BigQuery data cannot be read or edited at all.
- Sheets automation: Claude cannot create triggers, custom menus or scheduled refreshes.
- Slides charts: charts Claude builds are inserted as static images, not as linked Sheets charts, so they do not update when the underlying numbers change. Claude also cannot refresh an existing linked chart.
The Slides behaviour is the one worth planning around. A deck whose charts were generated by Claude will not reflect changes made later in the source spreadsheet, which makes the feature better suited to one-off drafts than to a recurring report.
Installing It, and When an Administrator Has To
Individual installation is short. Install Claude from the Google Workspace Marketplace, open a file, go to Extensions, then Claude, then Open Claude, allow the two permissions Google requests on first use, and sign in with a Claude account.
A greyed-out install button means the decision is not the user’s. In that case an administrator has to allow or install the app first. Super administrators can deploy it across the whole organisation, to selected groups or organisational units, or allowlist it so users can install it themselves. On Team and Enterprise plans, an owner or primary owner must separately enable the Docs, Sheets and Slides connectors before anyone can authenticate them.
Those connectors are a second, distinct feature released at the same time, also in beta. Instead of bringing Claude into a Google file, they let a user start in Claude, paste a Google file link or ask for a new document, sheet or deck, and work on it from the conversation. On supported setups the file opens in a pane beside the chat, and that built-in browser pane is off by default on Enterprise plans. Anthropic states that “Claude’s access matches your existing Google sharing permissions“.
What Happens to Your Document Data
The add-on requests two Google permissions: the ability to display third-party web content in prompts and sidebars, and the ability to view and manage the files it is installed in. Anthropic says it does not request access to Drive, Gmail, Calendar or contacts, and that Claude reads only the file that is open. Edits appear in the file’s version history under the user’s own name.
On retention, Anthropic says inputs and outputs are deleted from its systems within 30 days, and that this content is not used to train its models. Chat history is stored locally in the browser, per file, and does not sync between devices — so a conversation about a document will not follow the user to another machine.
Three compliance gaps are stated plainly in the documentation. Zero Data Retention is not available in this beta. The add-on does not inherit an organisation’s custom retention settings. And it is not covered for use with protected health information under HIPAA. Enterprise plans do get the Compliance API and customer-managed encryption keys applied to the add-on, which is the route to capturing session records before the 30-day window closes. For context on how much an assistant of this kind can see inside an application, Safari 27’s MCP server drew the same question about scope.
Status as of 8 October 2026
As of 8 October 2026, Claude for Google Workspace is in public beta, available on the Pro, Max, Team and Enterprise plans, and is not offered on the free tier. Anthropic has not given a date for general availability, has not said whether Firefox support is planned, and has published no benchmarks for the add-on’s accuracy.
The paid-plan requirement is the practical catch for most readers, since Google’s own assistant is already bundled into Workspace subscriptions. Anthropic’s pricing for this sits inside existing Claude plans rather than as a separate charge, a pattern that also governs which Gemini features need a paid plan.
Frequently Asked Questions
Which Claude Plans Include the Google Workspace Add-On?
Pro, Max, Team and Enterprise. There is no free-tier access during the beta, and sign-in must use a Claude account rather than an API key.
Why Does Claude Stop Partway Through a Long Task?
Google ends any single operation after about six minutes. Anthropic’s guidance is to break large jobs into stages rather than asking for everything at once.
Does It Work in Firefox?
No. The sidebar loads in Firefox but actions do not complete. Anthropic supports Chrome, Edge and Safari.
Can Claude Reply to Comments in Google Docs?
No. It cannot read, reply to or resolve comments, and it cannot accept or reject existing suggestions. Its own changes are applied as direct edits.
Is Zero Data Retention Available?
No. Zero Data Retention is unavailable in this beta, and the add-on does not follow an organisation’s custom retention settings. Anthropic says inputs and outputs are deleted within 30 days.
