Understanding AI Search Indexes

Effectively indexing large bodies of content involves choices by a solution developer; considerations include:

• Time required to index and reindex
• How current the index must be, especially if content is added frequently
• How computationally expensive the index operation is
• Which and how many data sources for the content are selected
• Whether the index will be used for general content search, AI, or both.

Azure AI Search offers three index types: keyword, vector, and hybrid, described in more detail below.

Keyword Index

Keyword indexes, as the name implies, rely on key terms in the query (such as “contracts” and “widgets” in the illustration). Keyword indexes have several advantages, including:

• Keyword indexing is very fast compared to vector indexing 
• Keyword indexing is especially useful for text-only datasets
• Keyword indexes can be consumed by Azure OpenAI for small Retrieval Augmented Generation (RAG) applications.

However, keyword indexes cannot be used to index nontextual data, such as images, and can be very difficult to scale efficiently to large datasets (say, social media feeds).