Keyword Research · Guide

What Is an LSI Keyword in SEO?

A term the industry uses for related words, which does not describe anything search engines actually do. Google has said so directly and more than once. If that is all you needed, you can stop after the first section.

Updated: August 2026
Written by: Andrew Odgers, Managing Director
Reading time: 11 minutes
The answer, before anything else

It Is Not A Thing Search Engines Use

LSI keywords are not something Google uses to rank pages. That is not a matter of opinion or a grey area. John Mueller of Google stated in July 2019 that there is no such thing as LSI keywords. He restated in January 2023 that they have no effect.

Why we lead with it. Most readers want one answer.

Somebody has been told to add LSI keywords to a page and wants to know whether that is real. It is not. Burying that behind eight paragraphs of history would be a waste of their afternoon.

Why the question keeps being asked. It is still sold.

Tools offer generators for them. Agencies list them on proposals. Articles explain how to use them. The demand for the phrase is genuine even though the thing it names is not.

What we are not saying. That related words are pointless.

They matter a great deal, which block four covers. The problem is the label, the mechanism it implies and what people do as a result.

Why this page exists at all. The demand is real.

People search for this term and deserve a straight answer rather than another article that repeats the myth in order to rank for it.

What it costs to believe it. A worse page.

The term is not merely harmless nonsense. Acting on it means editing vocabulary into a page rather than improving what the page says, which is time spent making something read worse.

What follows. The history, then the practical part.

The history is genuinely interesting. The practical half is block five onwards.

The history, kept short and fair

Where The Term Came From

Latent semantic indexing is a real technique. It emerged in the 1980s as a method for finding relationships between words across a collection of documents. It was a genuine advance in information retrieval. None of that is in dispute. What is in dispute is the leap from there to modern web search.

What it was built for. Small collections.

Sets of documents, in a library or a research archive. The scale it was designed for is a very long way from billions of pages that change daily.

Why it never made the jump. The maths does not scale.

The technique becomes impractical at web scale. Search engines developed other approaches to the same problem, which is understanding what a page is about rather than merely which words it contains.

How it entered the industry. A borrowed name.

The SEO usage took the term without the method. Somebody needed a way to describe using related words, this phrase sounded rigorous and it spread.

Why it stuck. It sounded scientific.

An idea with a technical name is easier to sell than one without. That is the whole mechanism of its survival. It is worth recognising in other terms too.

Why the history is worth knowing. Pattern recognition.

The same shape recurs constantly in this industry. A real observation, a scientific sounding name, a tool built to sell it and a decade of people repeating it. Recognising the shape is more useful than knowing this particular example.

What is fair to say. It was not a scam originally.

Most people using the term believed it. It described something they had correctly observed, which block four covers. They simply attached the wrong explanation to it.

Careful, because the output is not worthless

What The Tools Are Actually Selling

A generator producing LSI keywords is generally pulling related terms out of the pages that already rank for your subject. That is a reasonable thing to do. It is simply not latent semantic indexing. The list it produces is not a set of required words.

How they are usually built. From the results.

Take the pages currently ranking, extract words appearing frequently across them, present the result. Nothing mysterious, nothing to do with the technique whose name is attached.

What that list is genuinely good for. A prompt.

It tells you what other people writing about this subject thought worth mentioning. As a check against having missed something obvious, that is useful.

What it is not good for. A checklist.

The words are not requirements. Nothing rewards you for including them and nothing penalises you for leaving them out, which is the difference the label conceals.

Why we name no tools. A programme decision.

We do not name keyword tools anywhere on this site, paid or free. The criticism here is of a label rather than of any particular product. Naming one would suggest the others are different.

What to ask if you are sold this. One question.

Ask what happens if you leave the words out. Anybody who can answer that clearly is describing a prompt. Anybody who cannot is selling the label.

The constructive half

The Real Observation Underneath It

There is something true buried in the myth, which is why it survived. Content that covers a subject properly does contain the related language. Pages that cover a subject properly do tend to perform better. Both observations are correct. The explanation attached to them was not.

What people noticed. A correlation.

Pages doing well contained more of the vocabulary surrounding a subject. That was real and it was observable in any set of results.

What they concluded. Cause and effect, backwards.

They decided the words caused the performance, so adding words would produce it. In fact the coverage caused both the words and the performance.

Why the distinction matters commercially. Different work.

One version means editing vocabulary into a thin page. The other means knowing the subject well enough to write about it fully. Only the second produces anything.

How search engines actually handle it. Meaning.

Modern systems interpret what a page is about rather than counting which words appear. Our advanced material covers that mechanism. It is the genuine version of what the myth gestured at.

What that means for you. Write completely.

A page answering everything a customer would reasonably want to know contains the related language automatically, because you cannot explain a boiler replacement without mentioning flues, pressure and warranties.

The test. Would a tradesman recognise it.

If somebody who does the work for a living would read your page and think it was written by somebody who knows the job, the vocabulary is already right.

The practical warning

Do Not Insert Terms From A List

Working through a list of related words, finding places to fit each one into a page, is keyword stuffing with a newer name. The activity is identical to the practice everybody agrees is counterproductive. The only thing that changed is the list.

What it looks like in a page. Awkward sentences.

A paragraph that mentions three related terms it did not need. A sentence bent to accommodate a phrase. Any reader can feel it, even if they could not say why.

Why it fails mechanically. The words prove nothing.

Their presence does not demonstrate knowledge of the subject, which is the thing being assessed. A page containing the vocabulary and none of the substance is exactly what modern systems are built to see past.

The version that catches good writers. Editing afterwards.

Somebody writes a decent page, is handed a list and goes back to work terms in. The second version is worse than the first. The process felt like improvement throughout.

Where the wider argument lives. The stuffing guide.

Why keyword stuffing hurts SEO covers the modern forms of this, including the version specific to local businesses inserting town names.

What to do with a list you were given. Read it once.

Check whether it contains anything you genuinely forgot. Then close it and write the page. That gets the benefit without the damage.

The replacement

What To Do Instead

Cover the subject as somebody who knows it would explain it to a customer. That single instruction replaces the entire exercise, produces the same language and produces a better page while doing it.

What that means practically. Answer the real questions.

What does this cost, how long does it take, what goes wrong, what should I look out for, what are the options. Those are what a customer wants and they cover the subject naturally.

Who should write it. Somebody who knows.

The person doing the work has the vocabulary already. Getting half an hour of their explanation is worth more than any generated list.

The check worth running. Compare against the results.

Read what currently ranks and ask whether your page answers anything theirs does not. That is a useful exercise and it is not the same as copying their vocabulary.

What genuinely helps beyond words. Specifics.

Actual detail from actual jobs. Prices where you can give them, timescales, what varies and why. That is what distinguishes a page written by somebody who does the work.

Where the numbers get examined. The metrics guide.

If you are being sold this alongside difficulty scores and volume figures, our metrics guide covers what those numbers actually are.

Where the rest sits. The hub.

All twenty-two guides are on the keyword research guide.

Google said so in 2019, then again in 2023

You are being
sold a technique
that does not
describe anything.

The term borrows the name of a genuine method from the 1980s that was built for small document collections and never scaled to the web. What the generators actually produce is related words pulled from pages that already rank, which is a reasonable prompt and is not a list of requirements.

What we do in place of it:

Half an hour with whoever does the work The questions customers actually ask Costs, timescales and what varies What goes wrong, described plainly Coverage rather than vocabulary No lists worked through afterwards Checked against what already ranks No tool names on any report

If somebody hands you a list of terms to include, ask what happens if you leave them out.

The full guide series

Every guide.
One subject.

What the words mean, what the numbers in a keyword tool actually are, how to find the searches worth targeting, how to turn a list into a publishing plan and what to do when a page you relied on slips.

Questions people ask

LSI Keywords, Briefly

What is an LSI keyword?
An industry term for words related to your main subject. The label borrows the name of a genuine technique from the 1980s that search engines do not use for this, so what people call an LSI keyword is simply a related term with a scientific sounding name attached.
Does Google use LSI keywords?
No. John Mueller of Google stated in July 2019 that there is no such thing as LSI keywords. He restated in January 2023 that they have no effect. Both statements are public and neither has been contradicted since.
Are LSI keyword tools useless then?
The output is generally not worthless. Lists sold under that label are usually related terms gathered from pages that already rank, which can be a reasonable prompt about what a subject involves. The label is the problem rather than the list.
Should I use related keywords in my content?
You already will, if you cover the subject properly. Related language appears naturally when somebody who knows a topic writes about it. What you should not do is take a list and work through it, which is keyword stuffing under a newer name.
What is latent semantic indexing?
A real information retrieval technique developed in the 1980s for finding relationships between words across small collections of documents. It was never built for the scale of the web. The SEO usage borrowed its name rather than its method.
What should I do instead of looking for LSI keywords?
Write the page as somebody who does the work would explain it to a customer, covering what they would actually need to know. That produces the related language without any exercise. It produces a better page as well.