Exploring Brave's Latest Innovation: AI-Powered Search Engine Unveiled

Brave has recently introduced a new AI search engine called Answer with AI, which prioritizes privacy. This search engine operates with its own index of billions of websites. With their current search engine handling 10 billion queries annually, Brave's AI-powered search engine is now among the largest online.

Despite concerns from the search marketing and ecommerce communities about the impact of AI search engines on the future of the web, Brave's AI search engine still displays links. Importantly, it does not automatically respond to commercial or transactional queries with AI. This should come as welcome news for SEOs and online businesses. Brave is committed to preserving the web ecosystem and will closely monitor website visit patterns.

Answer With AI Is Powered By Brave

Search Engine Journal recently had a chat with Josep M. Pujol, the Chief of Search at Brave. He shared insights on the search index and how it collaborates with AI. Most importantly, he provided valuable advice for SEOs and business owners looking to boost their rankings.

Brave's AI search engine is different from other AI search solutions because it relies solely on its own search index of crawled and ranked websites. All the technology behind the search index, Large Language Models (LLMs), and Retrieval Augmented Generation (RAG) technology is developed in-house by Brave. This not only prioritizes privacy but also ensures that Brave's search results are one-of-a-kind, setting it apart from other search engines.

Search Technology

Our search engine is developed internally. Josep M. Pujol, Chief of Search at Brave, explains that we have immediate access to over 20 billion web pages. This allows us to extract various information such as schemas, tables, snippets, and descriptions in real-time. We are also able to choose specific data to use, ranging from entire paragraphs or texts to individual sentences or rows.

Retrieval Augmented Generation (RAG)

With a full search engine available to us, our main concern shifts from retrieval to the selection and ranking of information. Not only do we have access to pages in our index, but we also have access to important ranking information like scores and popularity. This access is crucial in determining which sources are the most relevant.

The search engine operates by using a search index along with large language models and Retrieval Augmented Generation (RAG) technology to provide up-to-date and fact-based answers. When I inquired about RAG, Josep verified that this is indeed how the search engine operates.

Our New Feature Utilizing RAG

You are right about our new feature utilizing RAG. In fact, we have already employed this technique in our previous Summarizer feature launched in March 2023. However, with this new feature, we are enhancing both the amount and quality of the data utilized in the content of the prompt.

I asked about the language models in use in the new AI search engine and how they’re deployed.

“Models are deployed on AWS p4 instances with VLLM.

We primarily use Mixtral 8x7B and Mistral 7B as our main LLM model.

In addition to that, we employ several custom trained transformer models for tasks like semantic matching and question answering. These models are smaller in size to meet the strict latency requirements of 10-20 ms.

The selection of data for our feature is crucial because it determines what will appear on the final LLM prompt. This data can include text snippets, schemas, tabular data, or structured data from our rich snippets. The focus is on selecting the most relevant candidates to add to the prompt context.

For example, when processing the query "presidents of France by party," 220KB of raw data is analyzed. This includes 462 rows selected from 47 tables and 7 schemas. The prompt itself consists of around 6500 tokens, with the final response being a mere 876 bytes.

In short, one could say that with “Answer with AI” we go from 20 billion pages to a few thousand tokens.”

How AI Works With Local Search Results

I then inquired about how the new search engine will show local search results. I asked Josep to provide some scenarios and examples of queries where the AI answer engine will display local businesses. For instance, if I search for the best burgers in San Francisco, will the AI answer engine give a response with links to relevant businesses? Will this feature be beneficial for individuals planning business trips or vacations?

Josep responded:

The Brave Search index contains over 1 billion location-based schemas, allowing us to gather data on more than 100 million businesses and points of interest.

Answer with AI is a comprehensive term that encompasses Search, LLMs, and various specialized machine learning models and services. This combination helps us retrieve, rank, clean, merge, and present information effectively. It's important to note that LLMs are not solely responsible for all decisions. Currently, we primarily use them to analyze both unstructured and structured data, whether it's during offline processes or when responding to queries.

Tips For Ranking Well

Sometimes, the outcome may show a strong influence from LLM. This happens when we think the user's question can be answered with a single Point of Interest, such as "checkin faro cuisine." Other times, their impact is more discreet, like with "best burgers sf," where they create a business description from various web sources or group the business into a specific category in a uniform taxonomy.

I next asked if using Schema.org structured data was useful for helping a site rank better in Brave and if he had any other tips for SEO and online businesses.

He answered:

When creating the LLM prompt, we focus on incorporating schema.org structured data. It is ideal to include structured data related to the business using standard schemas from schema.org. The more detailed these schemas are, the more precise the response will be.

Although our Answer with AI can also provide information about the business not covered in those schemas, it is recommended to present information in various formats. This ensures a more comprehensive and accurate response.

Some businesses solely depend on aggregators like Yelp, Tripadvisor, and Yellow Pages for their business information. However, there are benefits to incorporating schemas into the business website, even if it's just for crawling bots.

Embracing AI Search in the Brave Browser

Brave shared that at some point in the near future they will integrate the new AI search functionality directly in the Brave Browser.

Josep explained:

We are excited to announce that we will soon be integrating the AI answer engine with Brave Leo, the AI assistant built into the Brave browser. This will allow users to easily send the answer to Leo and seamlessly continue the session there.

Other Facts

Brave also revealed some interesting details about their new search engine.

Brave Search goes beyond just providing text answers. By deeply integrating the index and model, it can incorporate online, contextual, named entities enrichments. This process adds more context to a person, place, or thing, resulting in answers that include generative text, informational cards, and images.

The Brave Search answer engine is able to gather information from its index and geo local results to offer detailed insights on various points of interest. Currently, the Brave Search index contains over 1 billion location-based schemas, allowing us to access data on more than 100 million businesses and other points of interest. With these extensive listings, the answer engine can quickly provide comprehensive results for points of interest worldwide.

Experience the capabilities of the new AI search tool by visiting http://search.brave.com/

Editor's P/S:

Brave's introduction of Answer with AI, a privacy-focused AI search engine, marks a significant step in the evolution of search technology. By combining its own search index with AI capabilities, Brave aims to provide users with comprehensive and up-to-date information while safeguarding their privacy. This move also challenges the dominance of traditional search engines and raises questions about the future of the web ecosystem.

For SEOs and online businesses, Brave's commitment to preserving the web ecosystem and monitoring website visit patterns offers a sense of stability. While Answer with AI prioritizes privacy, it does not automatically respond to commercial or transactional queries with AI, ensuring that businesses can still reach their target audience through organic search results. The advice provided by Josep M. Pujol, Chief of Search at Brave, emphasizes the importance of using structured data and providing detailed information to improve ranking. As Brave integrates Answer with AI into its browser, users can expect a seamless and enhanced search experience that combines the power of AI with the privacy and control they value.