Anthropic gives Claude real-time web search, closing its biggest gap with ChatGPT
Anthropic added web search to Claude for paid users in the United States, working with Claude 3.7 Sonnet and enabled through profile settings, with Claude deciding on its own when a query needs current information. It returns direct citations so users can check the sources. Free-user and international access were promised for later.
Two details are worth keeping. First, the underlying engine appears to be Brave Search: Simon Willison traced the connection through Anthropic's subprocessor list, which added Brave Search on March 19, and demonstrated it by showing Claude's ten cited results matched a Brave query exactly, with 'BraveSearchParams' visible in outputs — a detail with real consequences for anyone who wants to block their site from being crawled on Claude's behalf. Second, Anthropic published no search accuracy benchmarks, and independent testing of LLM-based search assistants has found citation error rates around 60%.

WHY IT MATTERS
The knowledge cutoff was the single biggest practical reason to prefer ChatGPT for research, and removing it makes Claude a genuine substitute rather than a specialist tool. For builders, the lesson is that citations are not verification — a cited claim can still be a confabulated source, and this feature arrives without accuracy benchmarks attached. If you route Claude's search output into anything automated, treat the citation as a pointer to check, not as evidence.
