Claude adds web search with citations for all users

Real-time smarts, finally

Claude’s days of living under a digital rock are officially over.

Anthropic just rolled out native web search capabilities to Claude, allowing the conversational assistant to tap into live internet data and return cited, up-to-date answers across tasks. The integration, running alongside Claude 3.7 Sonnet, determines automatically when an inquiry requires real-time information rather than relying strictly on training data.

Every search-backed answer includes explicit source citations. Instead of sending users off to dig through search result pages, Claude summarizes findings conversantly while leaving an audit trail you can click through to verify.

The company initially launched the feature as a preview for paid US users before expanding web search globally across all tiers, including free accounts, on May 27.

Better late than sloppy

If you follow AI tools closely, your immediate reaction is probably: what took so long? Competitors like ChatGPT and Gemini have offered live web retrieval for ages, and Perplexity built an entire company around the concept.

Scott White, Anthropic's product lead for Claude, addressed the delay in an interview with The Verge. He noted that the team was juggling multiple parallel initiatives and that fusing search deeply into the model’s reasoning framework is technically tricky.

Pairing a probabilistic language model with the wild web is a known recipe for disasters. Google demonstrated that hazard vividly when its AI Overviews told searchers to put non-toxic glue on pizza after scraping Reddit jokes. Anthropic is leaning heavily into transparent citations so users can immediately distinguish between genuine news outlets and satire sites. White also confirmed that while live retrieval drives higher inference expenses, Anthropic will not charge users extra right now, though alternative pricing tiers might emerge down the road.

Why it matters

In our view, web search closes the single biggest utility gap that kept Claude from running away with the daily driver crown. Claude 3.7 Sonnet already holds a stellar reputation for complex reasoning and long-form code, but being blinded by a fixed knowledge cutoff forced users into awkward multi-tool workflows.

This update is especially bad news for standalone AI search wrappers like Perplexity. When frontier model makers bake reliable, cited browsing directly into their flagship chats, the justification for paying twenty dollars a month for a separate search engine shrinks drastically.

Anthropic played the waiting game here, but it looks like the right bet. Rather than rushing out hallucination-prone scrapers, they delivered clean inline sourcing. The real test moving forward will be managing infrastructure costs without gatekeeping search behind higher subscription tiers while keeping spammy SEO link farms out of Claude's answers.