- Aug 9, 2017
- 5,592
- 6,878
We all know each of the big AI companies uses search results as a basis for RAG queries, and as none of them have their own sizable proprietary index (yet!), they're using third parties for this purpose. Gemini is obviously powered by Google's index, OpenAI is heavily dependent on Bing (though I'm sure you'll have seen their crawler in the wild by now), and Claude uses the lesser-know Brave Search. The thing about Brave Search is that a great deal of the tech behind it is open source: https://github.com/brave/web-discovery-project
This X article has dived deep into the guts of Brave Search, and it makes for some interesting reading:
A few of the things mentioned certainly sparked my interest, such as lower HTML length ratio pages being discarded by the crawler - I tend to make my bulk sites as lean as possible on the html, so I will be testing with more added markup. Engagement signals also look automatable to some extent.
Remember, if you want to be cited by Claude, you most likely need to be in Brave Search's index!
This X article has dived deep into the guts of Brave Search, and it makes for some interesting reading:
A few of the things mentioned certainly sparked my interest, such as lower HTML length ratio pages being discarded by the crawler - I tend to make my bulk sites as lean as possible on the html, so I will be testing with more added markup. Engagement signals also look automatable to some extent.
Remember, if you want to be cited by Claude, you most likely need to be in Brave Search's index!