Claude gains Research and Google Workspace integration

Background

A chat assistant that answers from a single search has a narrow view: it sees whatever one query happens to return and then writes an answer from it. Questions that need several steps — comparing options, tracking down a detail inside a long document, or checking one claim against another — are poorly served by that pattern, because the second question depends on what the first one found.

Research in Claude was built around that gap. Instead of one lookup, the assistant carries out a chain of searches, each building on what the previous one turned up, and shifts its angle on its own rather than waiting for the user to redirect it^[1].

What launched

Anthropic introduced Research in Claude on 15 April 2025^[1]. The feature runs multiple searches that build on each other, explores different angles on its own and provides citations^[1]. According to Anthropic, the answers come with citations that are easy to check^[1] — the point being that a reader can follow a claim back to the page it came from instead of taking the model's word for it.

Using it does not require a separate product: users turn Research on by toggling the Research setting in chat^[1].

Connecting a workspace

Alongside Research, Anthropic shipped a Google Workspace integration that lets Claude search Gmail, Calendar and Google Docs, released in beta for all paid users^[1]. This changes what the assistant can draw on. A general web search reaches public pages; a connected workspace reaches the mail, schedules and documents a person or an organisation already has, which is usually where the relevant answer actually lives.

The rollout had an administrative step built in. Team and Enterprise admins first had to enable Google Workspace access domain-wide before individual users could connect their accounts^[1]. On top of that, Claude Enterprise admins could turn on Google Docs cataloging for the whole organisation, which uses a specialised index and secure retrieval-augmented generation^[1].

Retrieval-augmented generation, in plain terms, means the system fetches relevant passages from a document collection first and then writes its answer with those passages in front of it, rather than relying only on what the model absorbed during training. A specialised index and a secured retrieval path are what make that practical over an organisation's own files: the index decides what can be found, and the retrieval step decides what gets handed to the model.

Where it was available

Research launched in early beta for Max, Team and Enterprise plans in the United States, Japan and Brazil^[1]. At the same time, web search was extended to Brazil and Japan^[1]. The two changes arriving together meant that users in those two markets got a browsing capability and a multi-step research mode in the same release.

Why it matters

Two things distinguish this release. The first is the shape of the work: a research task is iterative, and a mode that plans several searches and follows them in sequence moves the assistant closer to how a person actually looks something up — start somewhere, read, then let the reading decide the next move.

The second is the source of the material. With citations attached to web results and with Gmail, Calendar and Google Docs reachable through a connected account, answers can be grounded in pages a reader can open and in documents the reader already owns. That makes the output easier to audit, and auditing is what separates a plausible-sounding paragraph from something a person can act on. The administrative controls — domain-wide enablement, organisation-level cataloging — reflect the same concern from the other side: an organisation deciding what the assistant is allowed to see before anyone starts asking it questions^[1].