AI Search: What's different?

OpenAI is integrating search into ChatGPT, creating a multiple-horse race in the AI search market between Google, Perplexity, Bing, possibly Meta, and many more.

How’s traditional search different from AI search, and how can companies like Open AI and Perplexity — both with combined valuations less than Google’s 2023 annual revenue—compete with Google?

How Existing Search Works

There are several search platforms, such as Bing, DuckDuckGo, and others, but Google dominates search with a 90% share of the global search engine market.

So, let’s just focus on how Google search works.

To sidestep all the complexities of Google search, we can roughly divide it into three buckets: crawling, indexing, and the ranking algorithm.

Crawling

Google uses bots to discover and download web pages. These bots or automated programs follow links from one page to another, covering much of the web. During crawling, the bot reads the page content, including text, metadata, links, images, and structured data like schema markup.

Indexing

Once a page is crawled, its data is processed and stored in a massive database called an index. Indexing organizes the information on each page, making it quickly retrievable.

Algorithm & Ranking

When you ask a question or search for something, Google's search algorithm looks for pages containing keywords from the search query and assesses their relevance and intent. The goal here is to show the best web pages related to your question.

How AI Search Works

When people mention "AI search" today, they’re usually referring to its new role as an answer engine rather than a traditional search tool that outputs multiple blue links.

It combines language models with traditional web search capabilities to directly respond to a query with citations.

On the backend, it leverages a technology called Retrieval Augmented Generation (RAG). This allows you to augment the capabilities of a large language model (LLM) like ChatGPT with facts obtained from external sources.

Here’s how it works:

When you ask a question, it is forwarded to a traditional search engine. The web pages related to your question are extracted, and the relevant content from those pages is fed into an LLM.

The LLM then does a job similar to that of a virtual assistant by reading through all the content and giving you a concise, summarized answer to your question. Importantly, the LLM also includes the appropriate sources to support the answer.

Let’s say you ask, “What is AI?” it will:

• Scan relevant websites and blogs

• Compile key facts into a helpful summary

• Provide a concise answer with citations

Do Perplexity and Open AI really have a chance?

It’s easy to see how an answer engine can save time and effort in filtering through irrelevant content.

With platforms like Perplexity AI, Chatgpt and similar products, you can engage in natural, back-and-forth dialogues to find the answers you need.

Instead of reading through multiple links, one can simply ask questions in everyday language and receive detailed, context-aware responses. This conversational approach makes searching feel more intuitive.

However, Google Search also seems to be fully embracing this new paradigm. They recently started rolling out the AI overview feature into Google Search in other countries outside the United States to provide summarized answers.

One could argue that this new form factor for search could significantly increase the total volume of search queries, with some of those done on specific applications that are good at certain things.

For example, Perplexity is flexible and great at using multiple models to understand queries and output great answers, so it is not limited to a specific LLM. Whereas Open AI and Google are limited to just their own trained models.

Also, Google potentially has the most to lose here. Perplexity, Open AI and others can only gain market share or pivot into some other AI product if they are unable to gain significant market share. But Google’s main business is search, and any sustained loss in search volume could lead to a spiral into the abyss like Yahoo.