Claude can now search the web and cite what it finds, closing the gap that sent reps to a second tab for anything that happened this quarter.
Pre-call research fails on recency. A model trained months ago cannot tell you about the funding round, the reorg or the earnings call that is the actual reason to reach out. Search plus citations means the research and the reasoning happen in one place, and you can check the source before you put a claim in an email.
Until this shipped, Claude answered from what it learned during training. Anything after that cutoff date was invisible to it. Web search means Claude can go and look, then quote the page it used.
The citation part matters more than the search part. A model that searches but does not show its sources has moved the verification problem rather than solved it. Citations let you click through and confirm before the claim reaches a buyer.
The short version: Claude can now look things up and show you where it got them, so current-events research and analysis stop being two separate jobs.
Anthropic launched web search for Claude on 20 March 2025, allowing Claude to search the internet and return answers with direct citations.
Availability at launch: Paid Claude plans, United States only.
Later expansion: The same announcement carries a 27 May 2025 update taking web search global and extending it to the free plan.
Activation: Users toggle web search on in profile settings.
Price: Not stated — included in existing plans rather than metered separately.
Rate limits: Not stated in the announcement.
What it doesn't do: it does not make Claude a research database. It searches the open web, so anything behind a paywall, inside a CRM or discussed privately remains invisible. And the launch headline overstated reach — this was paid-tier, US-only for its first ten weeks, which is worth remembering when a vendor announcement says a feature is simply 'available'.
SDR — Directly useful and immediately. Trigger events — funding, leadership changes, product launches — are the whole basis of a relevant first touch, and this is the fastest path from company name to a defensible reason for the call.
AE — Use it for the pre-meeting sweep rather than for deep account planning. It is strongest at 'what changed since we last spoke' and weakest at judgement calls about org dynamics, which are not on the public web.
Manager — The coaching point is verification, not speed. Citations exist so reps check them; a rep who pastes a cited claim without opening the link has adopted the risk without the control.
RevOps — Consider where research output lands. If reps are gathering current account intelligence in a chat window and nothing reaches the CRM, you have created a second shadow system rather than closed one.
The failure mode in AI-assisted research is a brief full of facts that are true and useless. Force the output down to the handful that would actually change how you open the call.
Two minutes, once a week. What changed in AI, and what to run because of it.