Claude Can Now Edit Your Google Docs, Sheets, and Slides Directly — What a Business Needs to Configure First

Two things shipped, not one

It is easy to read “Claude in Google Workspace” as a single feature. Per Anthropic's own launch article, two distinct things shipped together. The first is a sidebar add-on: one installation covers Docs, Sheets, and Slides, and it places Claude alongside whichever file is already open, where it can read the document, see what text or cells are selected, and make changes in place. The second is a set of connectors, in beta, that work the other direction — a user pastes a Google file link into a Claude chat, or asks Claude to create a new Doc, Sheet, or Slide deck, and Claude works on it without the file ever being open in a browser tab (Anthropic's launch article). Both are available on every paid plan — Pro, Max, Team, and Enterprise — and both are documented in Claude's Help Center.

The practical capabilities match what a business actually drafts in Workspace: tightening wording and restructuring a Doc, building a financial model or analyzing a dataset live in Sheets, and generating slides from existing layouts while flagging overlap in Slides, per Claude's product page. PDF, CSV, and Office files can be dropped into the sidebar directly, and a saved workflow (a “skill”) can be reused across files instead of re-explaining the same instructions every time.

The admin step that gates it — and is easy to miss

Nothing in Google Workspace connects to a third-party AI tool by default, and Claude is no exception. Per Claude's own Help Center article on the connectors, a Google Workspace admin has to explicitly trust Claude as a third-party application — in the Google Admin console, under Security > Access and data control > API controls > Manage third-party app access — before any user in the domain can authenticate it. On Team and Enterprise Claude plans, there is a second gate on top of that: an Owner or Primary Owner of the Claude workspace must separately enable the connectors at the organization level before individual users can turn them on.

That two-gate design is the right shape for a business, but it only works if someone actually walks through both gates on purpose. The failure mode worth naming is the opposite of over-caution: an admin approves Claude in the Google Admin console because an individual contributor asked for it, without anyone separately deciding which teams should have the Claude-side toggle on. A trusted-app approval is a yes/no switch for the whole domain; it is not, by itself, a decision about who should be editing finance models or client-facing decks through it.

The setting that decides how much control you keep

The detail most worth reading before rollout is the edit-approval setting, because it is the one choice that determines whether Claude proposes changes or simply makes them. Per Anthropic's launch article, the default is “Ask before edits”: Claude presents proposed changes as approval cards, and a person confirms each one before it lands in the file. The alternative, “Accept all edits,” lets Claude apply changes straight through without stopping. For a first draft nobody has seen yet, that is a reasonable speed trade. For a live shared model or a deck already circulating for client review, it is the setting that turns a drafting tool into something editing a document other people are already relying on, unsupervised.

Access itself is scoped sensibly: per Anthropic, “Claude sees only the file you open it in, plus any connectors you turn on,” and it inherits whatever Google sharing permissions already exist on that file — it does not grant Claude a new, broader view of a shared drive. On Enterprise plans, the controls a business would already expect carry over to the add-on: the Compliance API, customer-managed encryption keys, and OpenTelemetry audit export, per the same launch article. Those controls answer who can see what happened. The “Ask before edits” setting is the one that answers who approved it happening in the first place— and it is the one a business actually has to choose, not just inherit from a platform default.

Rolling it out without a surprise

Where Skoor fits

Skoor implements Claude's own apps on a business's own devices and treats a connector like this one as a configuration decision, not a checkbox: connectors are allowlisted per role, against a written profile for what that role is permitted to touch, with a published, versioned rubric and a dated, attributable record kept in a store the business controls — the pieces described on Skoor's security page. Anthropic built the approval controls into the product; deciding which roles get “Accept all edits” on which file classes, and reviewing that decision on a cadence, is the governance layer a business still has to own — and the one Skoor implements alongside the connector itself.