Anthropic adds web search to Claude 3.7 Sonnet, available now in preview for paid US Claude users, with support for free users and more countries coming soon
Anthropic's AI-powered chatbot, Claude, can now search the web — a capability that had long eluded it.
TechCrunchKyle Wiggers
Context & Ripple Effects
Claude began as a consumer and business chatbot with paid tiers, then broadened its product surface through interactive Artifacts. Web search adds current-information retrieval to that existing assistant experience rather than creating a separate product.
The paid-US preview was an initial distribution step: the capability later reached developers through web search in Anthropic's API, while a later Chrome preview extended Claude toward acting inside the browser.
First-order effects
Paid Claude users in the US can test web-grounded responses in Claude 3.7 Sonnet immediately; free users and additional countries remain pending.
Anthropic gains a new premium-preview feature that makes Claude more useful for questions requiring information beyond its built-in knowledge.
Second-order effects
The preview creates a path from consumer testing to developer use, as shown by the later API rollout with per-search pricing; customers can begin treating fresh web information as an application input rather than only a chat feature.
Search-enabled answers make source access and attribution more consequential for publishers whose material may be surfaced through AI assistants.
Third-order effects
If web retrieval becomes standard across assistant tiers and APIs, differentiation shifts from static model knowledge toward the quality, cost, and trustworthiness of the retrieval-and-citation layer.
The feature is one step toward Claude becoming an assistant operating layer: subsequent browser action capabilities suggest a product arc from answering questions to completing work across web-based contexts.
The trend: AI chatbots are evolving into web-connected work surfaces that combine model reasoning, live retrieval, and eventually action-taking.
It isn't really Claude's “fault” here, since its search did come up with two of my posts, the most recent and then one I wrote back in December. It then uses that info as context for the answer. The issue is that the AI “trusts” the search result too much. Same issue with RAG. [i…
Claude now has a good version of web search, which is really useful, but it also has the same issue as all non-agentic (that is, not Deep Research) AI uses of web search or RAG, which is the model only gets a small amount of context from search & then sometimes over-extrapolates.…
Claude has browsing. Claude has browsing. Claude has browsing. Claude has browsing. Claude has browsing. Claude has browsing. Claude has browsing. Claude has browsing.
Excited for this release! Claude can now search the web for you and provide citations. This is an important feature for getting the most up-to-date and reliable information through Claude.
Wrapping up our week of Claude Code updates with a much-requested feature: web fetch. This eliminates a major context-switching pain point. Here's how it works: [image]
So with @AnthropicAI enabling “search” in Claude, and @OpenAI already offering this capability — is “search” now just a feature of a bigger shift in how we interact with information?