LSI keywords are the related words and phrases SEO tools suggest adding to a page, named after latent semantic indexing, a document-retrieval method from the late 1980s. Google doesn't use them: Google's John Mueller wrote in July 2019 that "There's no such thing as LSI keywords." What helps is covering the related terms, entities and subtopics a complete answer needs.
Key takeaways
- Latent semantic indexing is a real 1980s retrieval technique, but it isn't among the ranking systems Google documents.
- "LSI keywords" is an SEO label for related terms: covering related terms is sound, the theory is not.
- Google matches meaning with RankBrain, BERT and neural matching, and understands entities through its Knowledge Graph.
- What works: cover the subtopics and terms the top-ranking pages share, in natural prose.
- Repeating terms to hit a count is keyword stuffing under Google's spam policies.
What are LSI keywords?
LSI keywords, in SEO usage, are words and phrases related to a page's main keyword: synonyms, variants and terms that tend to appear alongside it. For a page about cold brew coffee, a tool might suggest "coarse grind", "steep time", "concentrate" and "French press".
The theory behind the label is that Google uses latent semantic indexing to judge relevance, so adding these terms lifts rankings. The theory is wrong, though covering related terms still helps. "Semantic keywords" or "related keywords" describe the idea more accurately.
What is latent semantic indexing (LSI)?
Latent semantic indexing is a method for retrieving documents by the concepts they share rather than by exact word matches. Researchers at Bell Communications Research filed a patent for it in September 1988, "Computer information retrieval using latent semantic structure" (US 4,839,853), and described it in the 1990 paper "Indexing by Latent Semantic Analysis" in the Journal of the American Society for Information Science (Microsoft Research listing).
The method builds a large matrix of which terms appear in which documents, then uses singular value decomposition to compress it into about 100 factors. Documents and queries that score on the same factors count as related, even when they share few words. LSI was built for indexing document collections, a decade before Google existed.
Does Google use LSI keywords?
No. On Twitter (now X) on July 30, 2019, Google's John Mueller wrote: "There's no such thing as LSI keywords -- anyone who's telling you otherwise is mistaken, sorry." (John Mueller on X). Google's guide to its ranking systems doesn't mention latent semantic indexing.
Instead, the guide describes systems that match meaning rather than exact strings:
System | What Google says it does |
|---|---|
RankBrain | "helps us understand how words are related to concepts" |
BERT | understands "how combinations of words express different meanings and intent" |
Neural matching | understands "representations of concepts in queries and pages" and matches them |
Passage ranking | identifies individual sections, or passages, of a page |
Google also describes a "sophisticated synonym system that allows us to find relevant documents even if they don't contain the exact words you used" (How Search works). You don't need to sprinkle synonyms for Google to understand a page.
Why do related keywords still matter for SEO?
Related keywords matter because they are evidence that a page covers its topic, not because of LSI. Google's own example: when you search for "dogs", "you likely don't want a page with the word 'dogs' on it hundreds of times," so its algorithms assess whether a page has other relevant content, "such as pictures of dogs, videos, or even a list of breeds" (How Search works).
Related terms
Related terms are the words a complete answer naturally uses: a cold brew guide that never mentions steep time or grind size is missing something readers need.
Entities
An entity is a specific thing, such as a person, place, product, organization or concept, rather than a string of letters. Google introduced its Knowledge Graph in 2012 to understand "real-world entities and their relationships to one another: things, not strings" (Google, May 2012). For entity SEO, name things precisely (the product model, the city, the author), state their attributes, and add structured data where it fits; Google uses structured data "to understand the content of the page" (Intro to structured data).
Subtopics
Subtopics are the questions a full answer has to cover, and they matter more with AI search. Google says AI Overviews and AI Mode "may use a 'query fan-out' technique — issuing multiple related searches across subtopics and data sources — to develop a response" (AI features and your website). A page that answers each subtopic under a clear heading can be found for those searches. When a subtopic is big enough for its own page, link the pages into topic clusters.
LSI keywords vs what actually works
Approach | What it really gives you | Verdict |
|---|---|---|
"LSI keywords" from a generator | A list of related phrases, not an LSI effect | Fine for brainstorming |
Sprinkling synonyms | Nothing Google's synonym system doesn't already do | Skip |
Covering terms the top pages share | What readers expect a full answer to include | Do it |
Naming entities clearly | No doubt about who or what the page is about | Do it |
Answering subtopics | A chance to rank for the sub-questions, including AI fan-out searches | Do it |
LSI keyword examples, and what they really are
Here is a typical "LSI keyword" list for "how to make cold brew coffee", sorted by what each term really is:
Type | Examples | Why it belongs on the page |
|---|---|---|
Variants and synonyms | cold brew recipe, cold-brewed coffee | Google matches these anyway; use whichever reads naturally |
Related terms | coarse grind, steep time, coffee-to-water ratio, concentrate | A complete recipe can't avoid them |
Entities | French press, mason jar, arabica beans | Specific things readers own or buy |
Subtopics | how long to steep, cold brew vs iced coffee, how long it keeps | The questions readers ask next |
How to find the related terms top pages use
The best source of related terms is the pages already ranking for your keyword. Only compare pages with the same search intent as yours.
- Search your keyword in the country you target and open the top-ranking pages.
- Write down each page's H2 and H3 headings, and mark the subtopics most of them cover.
- Note the phrases that recur under those headings, plus People also ask and related searches.
- List the subtopics and terms your page doesn't cover yet.
- Add sections or sentences where a reader would expect them, never a list of terms. Google's spam policies list "repeating the same words or phrases so often that it sounds unnatural" as keyword stuffing (Google spam policies).
How to do this in RankTera
RankTera's KW Research automates steps 1 to 3. Enter the keyword and country, choose which Google results to analyse (up to 5) in Select SERP Results, and RankTera extracts the phrases they use under their H2–H6 headings with each phrase's average count (AVG COUNT), plus the pages' average word count and heading structure. AI Clean-up removes junk keywords. For a published page, Improve Existing Article finds the keywords you're missing versus the top pages and proposes edits you accept or discard.
What we found: the terms top pages use for "lsi keywords"
We ran "lsi keywords" through RankTera's KW Research on October 15, 2026, using US Google results. The pages it analysed (backlinko.com, lsigraph.com, keysearch.co, semrush.com and answersocrates.com) averaged 1,777 words, and it found 77 keywords. Selected terms, in the order RankTera ranked them:
Term | Average count per page |
|---|---|
LSI keywords | 10 |
related terms | 3 |
search engines | 2 |
unstoppable traffic | 2 |
content silo | 2 |
main keyword | 2 |
search intent | 2 |
semantic keywords | 2 |
"Related terms" is the second most used term and "semantic keywords" also makes the list, which suggests the pages describe the practice in those words. "Unstoppable traffic" comes from LSIGraph's sales page, a reminder to clean raw term lists. Our separate analysis of six top Google results the same day (median 1,737 words) found three keyword-generator tool pages and three articles; none used a table, and only one had an FAQ section.
Worked example: closing the related-term gap on a cast-iron guide
This example is illustrative; the counts are not real data. A kitchenware blog's 1,100-word guide to "how to season a cast iron skillet" ranks on page two. The writer compares it with the top-ranking pages:
Term or subtopic | Average count on top pages | The guide | Action |
|---|---|---|---|
seasoning | 9 | 7 | None; already covered |
oven temperature | 2 | 2 | None |
smoke point (which oil to use) | 3 | 0 | Add a section: "Which oil is best for seasoning cast iron?" |
rust | 2 | 0 | Add a section: "How to fix a rusty skillet" |
enameled cast iron | 1 | 0 | Add one sentence: enameled pans don't need seasoning |
The fix adds two sections and about 350 words without repeating "seasoning" once more. The gap was subtopics, not keyword density.
Frequently asked questions
Are LSI keywords and semantic keywords the same thing?
They describe the same practice: adding words and phrases related to your main keyword. "Semantic keywords" is the more accurate name, because Google doesn't use latent semantic indexing. These terms are useful because they show a page covers its topic fully, which is why the pages that rank tend to share them.
Do LSI keyword generators work?
They work as brainstorming tools, not as a ranking method. A generator gives you related phrases; it can't tell you which ones your page needs. Compare its suggestions with the subtopics the top-ranking pages actually cover, keep the ones a reader would expect, and ignore any claim that they trigger an LSI effect.
What are NLP keywords?
NLP keywords is another SEO label for related terms and entities, named after natural language processing, the field behind systems like BERT. Google does use language models to understand queries and pages, but a list of NLP keywords isn't a ranking formula. The advice is the same: cover the subtopics and entities a complete answer includes.
How many LSI keywords should I use per page?
There is no target number. Cover the subtopics and terms most top-ranking pages share, as long as they fit your page's intent, and use each term where the reader needs it. If you're inserting a phrase only to hit a count, stop: Google treats unnatural repetition as keyword stuffing. To check demand for a term, use Google Keyword Planner.
Do related terms and entities help in AI search?
Yes, indirectly. Google says AI Overviews and AI Mode may run related searches across subtopics before answering, and that no special optimization is needed. A page that names its entities clearly and answers each subtopic under its own heading has more chances to be retrieved for those searches. See our keyword research guide for the full process.
See the terms the first results actually use
Instead of chasing LSI keywords, look at what the pages that rank really cover. RankTera's KW Research reverse-engineers the terms the first results use on Google and in the pages ChatGPT and Gemini cite, and the AI Writer can turn that research into a draft. Start your 14-day free tier — no card needed; it includes 3 keyword researches and 3 AI articles. See pricing for paid plans.