The feature rolls out to U.S-based Claude Pro subscribers today.
The feature rolls out to U.S-based Claude Pro subscribers today.
Updated Mar 20, 2025, 4:50 PM UTC
STK269_ANTHROPIC_D
STK269_ANTHROPIC_D
Kylie Robison
Kylie Robison is a senior AI reporter working with The Verge’s policy and tech teams. She previously worked at Fortune Magazine and Business Insider.
Anthropic just gave its chatbot, Claude, a long-awaited upgrade: web search.
The company announced Wednesday that web search is now available for paid US users in a “feature preview,” and to enable it, users need to toggle web search on in their profile. Anthropic’s product lead for Claude, Scott White, told The Verge the the company plans to expand to free users and more countries in the “next couple of weeks.”
With web search integration, Claude can now pull in current data from across the internet, making it more useful for tasks that require up-to-date information. The system also adds citations to sources so you can verify where its information is coming from. In a video demo provided by Anthropic, an example query about building a web app triggers the model’s “search the web” feature. (White says you can toggle search on and off too.) It then highlights sources alongside the answer.
This feature has been a long-time coming, at least in AI years. ChatGPT launched search capabilities last July as a “prototype” before rolling it out more broadly over the course of the next few months. The demand for web search in Claude became so intense that some users created their own workarounds to give the AI internet access.
Last month in an interview, White asked what feature I wanted most in Claude. My answer was obvious: web search. It’s feels strange for chatbots to have knowledge cutoffs when competitors like Perplexity are already using the same models to search the web. Just before today’s announcement, I chatted with White again, and he explained why Anthropic took so long to implement this seemingly essential feature.
”We’re doing a lot of these things in parallel,“ White said. It’s a complex problem to work out, he added, thanks to how deeply integrated these features are with the model. When asked if these kinds of features (which are more expensive to operate than traditional search engines) would wind up costing users more money, White said no, but the company “might have different ways of pricing” for new features “in the future.“
What makes AI search tricky is the combination of two complex systems: probabilistic language models and web search. Unlike deterministic systems (like calculators) that always produce identical outputs for identical inputs, large language models generate different responses based on statistical probabilities — which means accuracy can vary wildly. When you combine this with web search, things get messy fast. We’ve already seen the spectacular failures this can produce, like when Google’s AI Overviews confidently recommended putting glue on pizza after scraping some troll Reddit comments. The consequence of failing can range from funny headlines to copyright lawsuits.
White emphasized Claude’s use of citations — so even if Claude gets it wrong, “we make it extremely easy for you to verify the content through that citation, to jump off and go directly to the source where we took that content from.” In other words, if it tells you to eat rocks, you can check if that came from WebMD or The Onion.
Most Popular