LSI keywords are supposed to be words and phrases that are semantically related to your main keyword, which you add to a page so Google sees it as complete. SEO blogs and tools sell lists of them. The label is wrong, and the practice built on it needs a different name and a different method.
Where does the LSI keyword idea come from?
Latent semantic indexing (LSI) is an information retrieval technique from the late 1980s. It uses maths (singular value decomposition) to find hidden relationships between words across a fixed collection of documents, so a search for “car” can also return documents about “automobiles”. It was designed for small, static document sets, long before the modern web.
SEO writers picked up the name in the 2000s and 2010s. Their logic ran: Google finds related words important, LSI finds related words, so Google must use LSI, so you should add “LSI keywords”. Tool vendors then built “LSI keyword generators” around the term. The leap from “Google understands related words” to “Google uses LSI” was never backed by anything Google published.
What does Google actually say?
On 30 July 2019, Google’s John Mueller wrote on Twitter that there is no such thing as LSI keywords, and that anyone saying otherwise is mistaken. Search Engine Roundtable reported it the next day.
Google’s published documentation describes other systems for understanding language:
- BERT helps Google understand how combinations of words express different meanings and intent.
- Neural matching connects concepts in queries to concepts on pages.
- RankBrain helps Google understand how words relate to concepts, so it can return relevant content even when a page lacks the exact words in the query.
Google’s SEO Starter Guide adds that its language matching systems are sophisticated, so you do not need to match every query variation, and that repeating words excessively, even in variations, is against its spam policies.
So Google does understand related terms and synonyms, with newer methods than LSI. The word “LSI” in front of them is the myth. The word lists sold under that label are only lists of related terms, and a page does not need a quota of them.
What should you do instead?
Aim for topical completeness and natural language. A page about a subject should cover what a reader needs, which brings the related words with it.
- Search your main keyword. Read People also ask, Related searches and autocomplete. These are the related topics Google itself associates with the query.
- Open the top five pages. List the subtopics at least three of them cover. Those are the expectations for the topic.
- Decide what the reader needs. Add the questions that the top pages miss: a missing step, a price, a warning, a template, a local detail.
- Write headings for subtopics, not for terms. “How long does sourdough starter take to rise?” is a section. “Starter fermentation timeline” as a loose phrase added to a paragraph is not.
- Use words as your reader uses them. If customers say “cheese board”, use it next to “charcuterie board”. Take the wording from your emails, reviews and Search Console queries.
- Define ambiguous terms once. Say what you mean by a term that has two meanings, then move on.
- Check Search Console after publishing. The Queries tab shows the real related phrases your page ranks for, which is better data than any LSI list (see keyword research).
- Link to deeper pages for subtopics that need their own article (see topic clusters).
The page on semantic SEO and entities covers the larger idea: Google matches meaning, so clear structure and complete coverage serve it better than a word list.
What mistakes should you avoid?
- Buying or generating “LSI keyword” lists and pasting them in. Words added without a reason read badly. Google’s spam policies list repeating words or phrases so often that the text sounds unnatural as keyword stuffing.
- Adding related words with no new information. Each related term should come with a sentence that tells the reader something.
- Believing a tool’s “semantic score”. Vendor scores are their own models. They can point at missing subtopics, but they are not Google’s measure. Use them as a prompt for questions, then verify each against the results page.
- Forgetting the main intent. A page with many related terms and the wrong format still fails (see search intent).
- Counting. There is no target number of related terms (see keyword density).
Where can you learn more?
The free digital marketing module of the Fe/male Switch startup school explains how to write pages that match what customers search. Next steps on SEO100: apply the topic coverage method in content depth and length, and read semantic SEO and entities.