On October 6, 2026, Anthropic launched a public beta of "Claude for Google Workspace," which lets Claude edit documents and presentations directly inside Google Docs, Sheets, and Slides. It also released dedicated connectors that let users create and edit Google files from the Claude chat screen. Both are available on all paid plans. According to the official announcement, the sidebar inside Google apps gives access to the same models and work skills people already use in Claude.
However, some features, such as comments and charts, are handled differently depending on whether you work from inside a Google app or from the Claude side. Being able to edit a document is not the same as being able to hand off an entire collaborative workflow.
Editing scope depends on how you use it
To use Claude from the Google side, you install the add-on from the Workspace Marketplace and open Claude from the file's "Extensions" menu. Claude can read the contents of the file, including the text or cells you currently have selected, and make changes directly in that same file.
This reduces the back-and-forth of copying text into a chat and pasting Claude's revised version back into the document.
The Sheets examples Anthropic gave go beyond suggesting formulas. Claude can create pivot tables for aggregation and charts in Sheets. For merging and reshaping data, it can process a specified range with Python and write the results back to the sheet.
In Slides, Claude adds slides while carrying over the existing layout and theme, and it also checks for overlapping elements and content that spills off the slide. The design lets users do calculations and data processing in Sheets, and build presentations that match an existing design in Slides, all within the files themselves.
When you start from the Claude chat screen instead, you use dedicated connectors for Docs, Sheets, and Slides. You can pass a link to an existing Google file and ask for edits, or ask Claude to create a new Google document, spreadsheet, or presentation.
The existing Google Drive connector handles searching, uploading, and sharing files, so its role is separate from the dedicated connectors that edit file contents. According to the official connector help page, what Claude can access follows the existing sharing permissions of the connected Google account.
This difference lets you choose based on where you start. If you are revising a proposal that is already open, the sidebar inside the Google app is the easier starting point. If you are building a new Google-format document from information gathered in a conversation with Claude, the dedicated connectors are a better fit.
The sidebar alone cannot freely open other files in Drive or create new ones. To reference additional materials or files, you use the connectors separately.
Approval cards are not the same as Google's suggestion mode
In the sidebar, "Ask before edits," which requests confirmation before changes are made, is the default. You can also choose "Accept all edits," which proceeds with ordinary edits without asking each time.
Even when you review and apply changes under "Ask before edits," they are applied directly to the Google Doc itself. They do not remain as proposed edits the way they would in Google Docs' collaboration feature, suggestion mode.
In addition, the sidebar inside Google apps cannot handle comments in Docs, while the Claude-side connectors can. Comparing Anthropic's official help pages as of October 7, 2026, by where you operate, gives the following:
| Item | Claude sidebar inside Google | Claude-side Docs, Sheets and Slides connectors |
|---|---|---|
| Docs comments | Cannot read, reply to, or resolve | Can read and use |
| Docs suggestions | Claude's own changes are direct edits; cannot accept or reject existing suggestions | Described as able to read and use existing suggestions |
| Adding charts to Slides | Inserted as images; not linked to numeric changes in Sheets | Cannot add charts |
The sources are the sidebar's "Current limitations" section and the connector features and limitations. The table compares features available in the same public beta, not editing accuracy or processing speed. Also, the fact that the connectors can handle "suggestions" does not necessarily mean they can freely accept or reject every suggestion.
For example, having Claude rewrite text requires different functionality than revising a draft while reviewing comments left by colleagues.
The sidebar's approval card is a mechanism for users to review changes Claude makes; it does not replace Google's collaborative editing feature for multiple people to review proposed edits.
The same goes for charts in Slides. Producing an attractive static image calls for different functionality than a report that updates automatically as sales data changes.
Just before this, on September 30, Google added comment operations to the Docs, Sheets, and Slides APIs, and also announced a feature in Docs for submitting changes as "suggestions." The rollout is gradual, taking up to 15 days from that date, and comment functionality via MCP, which connects AI and external tools, is offered as a developer preview.
In other words, a feature being available in Google's underlying platform is a separate matter from whether a particular way of using Claude can access it.
The reasons the unsupported features are not yet supported have not been disclosed. Google's API expansion does not mean it can be assumed that Claude will support them soon.
Competition with Gemini extends to how work gets done
Gemini already offers document drafting in Google Docs, editing of selected text, and features for unifying style and formatting. Google's April 22 announcement also described a feature that generates documents using information from Drive, email, and elsewhere.
Having AI edit documents inside Google apps is not new with this release.
What matters for organizations that already use Claude every day is that they can bring the reference sources and work procedures they have set up on the Claude side directly into Google's editing screens.
Anthropic gave the example of creating a quarterly report from Salesforce customer information and call notes stored in Drive, saving that format as a skill, and using the same procedure from then on.
A skill packages instructions, such as how to write a report and what to check, in a reusable form. Which model to use is no longer the only selection criterion; which apps can run the procedures a team has built up also matters.
Google is moving in the same direction. In its September 30 skills announcement, it explained that publicly available instructions in the SKILL.md format can be copied over from other platforms. In Workspace, a phased rollout began on October 5.
However, skills are not synced between the Gemini app and Workspace, and each must be created separately.
Being able to move the instructions themselves does not guarantee the same work can be reproduced as is. If the internal systems that can be referenced and the editing features available in each app differ, the results can differ even with the same procedure.
Full compatibility between Claude and Gemini skills has also not been confirmed. Still, both companies share the effort to build routinely used work instructions into actual document work.
When comparing AI, factors will include not only the quality of the writing but also how much of an internal workflow can be migrated and how easily it can be maintained.
Enterprise adoption requires checking data retention terms
For Enterprise, features such as the Compliance API for retrieving activity logs, customer-managed encryption keys, and audit log export to OpenTelemetry also apply to this add-on.
However, not every setting used in an enterprise Claude environment carries over as is.
According to the sidebar's official help page, input and output data are deleted within 30 days, with some exceptions, but custom retention periods set by each organization do not apply.
This public beta also does not support Zero Data Retention (ZDR) operation, in which data is not retained, or inference through services such as Amazon Bedrock and Google Cloud Vertex AI.
Data not being used for model training is a different condition from data not being stored.
Management is also split between two parties. On the Google Workspace side, administrators control who can install the add-on. On the Claude side, they manage which models, connectors, skills, and so on are available.
Even if Google-side file permissions are set appropriately, you still need to separately confirm which data your own rules allow to be sent to AI and for how long it may be stored.
For organizations that start by trying document revision alone, this integration offers a new way to apply work procedures refined in Claude to actual Google files.
If they then expand to collaborative review or recurring reports, factors such as which route handles comments, how charts get updated, and whether data retention terms fit internal rules will inform the decision.
The race to build AI into Google documents is shifting from simply generating text to how deeply it can work its way into everyday tasks that people carry out while reviewing and confirming.