Latent Semantic Indexing is a mathematical technique from 1988, originally developed for information retrieval, that maps relationships between terms and documents based on which words tend to co-occur, its own kind of statistical guess at meaning rather than true language understanding. In SEO circles, "LSI keywords" became shorthand for related terms and synonyms worth including alongside a primary keyword to signal topical depth.
Here's the accuracy gap worth knowing: Google has stated directly, more than once, that it doesn't use LSI as a ranking technique, and modern search relevance runs on more sophisticated systems built around language models and entity understanding rather than 1980s-era term co-occurrence math. The advice "LSI keywords" usually points toward, covering a topic thoroughly with genuinely related terms and concepts rather than repeating one exact phrase, is still sound. It's just not actually LSI doing the work; it's closer to how Google evaluates topical completeness and entity relevance today. Chasing a list of "LSI keywords" from a tool is a weaker mental model than simply writing content that covers a subject the way someone who actually understands it would.
