"LSI keywords" is a term some SEO tools and guides use for words related to a main keyword. It borrows the name of latent semantic indexing, an information retrieval technique from the late 1980s. There is no evidence Google uses LSI, and Google representatives have said so publicly. The underlying advice, cover a topic with the vocabulary an expert would use, is sound; the label is a myth.
Where the term came from
Latent semantic indexing was developed to find relationships between terms in a fixed collection of documents. It was designed for small, static collections and does not suit an index of the whole web that changes by the second. At some point, SEO marketing adopted the name for any list of related words, and tools started selling "LSI keyword" suggestions.
What search engines actually do
Modern search uses language models, entity recognition and many other systems to understand what a page is about and how it relates to a query. These systems recognize synonyms and related concepts without needing a particular list of words to appear. What still matters is that a page genuinely covers its topic, and covering a topic naturally means using its vocabulary.
What to do instead
Treat related-term lists as a coverage check, not a quota. Look at what top-ranking pages and the People Also Ask box discuss, list the subtopics an expert would expect, and make sure the page addresses them where relevant.
Example
Reframing a writer brief
Hypothetical CRM vendor Mossgate's content briefs had a section labeled "LSI keywords: include all." Writers forced awkward words into drafts. The team renamed it "Subtopics to cover" and listed questions instead of words: how pipeline stages work, what to do with stale deals, how forecasting differs by team size. Drafts became more useful and still used the related vocabulary naturally.
Common mistakes
Inserting term lists mechanically
Dropping in words from a tool's list without adding the ideas behind them makes copy worse and does not improve relevance.
Questions
The phrase is used widely, but Google does not use latent semantic indexing. Related terms matter only as part of covering a topic well.
