Keyword variations decide how much of a subject your page actually covers

A page that says “dispensary” ninety times has said one thing ninety times. A page that says dispensary, cannabis store, weed shop, recreational dispensary, budtender and dispensary menu has said six things, and Google has six ways to understand what the page is about instead of one. That difference is measurable, it shows up in correlation data going back to 2025, and it is the single easiest thing most SEO content gets wrong.

This article covers what keyword variations are, what the on-page correlation data says about them, and how the difference plays out in a real market. The worked example runs on cannabis retail, because the vocabulary gap in that vertical is unusually wide and the search volumes are public. The full term set is published separately as our cannabis keyword list.

What keyword variations are

A keyword variation is an alternate phrasing that carries the same search intent as the head term. “Cannabis store” and “weed shop” describe the same place. A shopper typing either one wants the same result, and Google knows it, which is why the two queries return overlapping results.

The families of variation keywords

Variations split into recognizable families. Synonyms swap the noun outright, so dispensary becomes cannabis store. Morphological variations change the form of the word, so dispensary becomes dispensaries. Stemming and inflected forms cover tense and number. Long-tail variations add qualifiers, so dispensary becomes recreational marijuana dispensary. Misspellings and alternate spellings sit in the same family, though they matter less than they did a decade ago. Question forms turn the term into a query a person would type or speak.

Google reads these as related rather than identical because its ranking systems stopped matching strings some time ago. RankBrain and BERT interpret a query’s meaning, and the Knowledge Graph connects the named things inside it, which is why a page about a cannabis store can rank for weed shop without ever using the phrase. That machinery is also why coverage pays: the more of a subject’s vocabulary a page carries, the more anchors those systems have to work with.

Where a variation stops and a new page starts

A variation is not a different keyword. This is the line that decides whether a term belongs on your page or on another one. “Dispensary near me” is a variation of dispensary because the intent is identical and only the local qualifier moved. “Cannabis stocks” shares two syllables and nothing else, and it belongs to a different page entirely. Intent, not string similarity, sets the boundary. The practical test: if a searcher typing the second term would be satisfied by the page you are writing, it is a variation. If they would bounce, it is a separate page.

What the correlation data says about variation keywords

Several tools measure on-page factors across the top 100 results for a query and publish how each factor correlates with ranking position. Cora SEO is the one we happen to use, and the figures below are from its exports of March 6, 2025 and August 2026. Treat them as one example of the shape rather than as the authority: no correlation tool publishes its sample size, every vendor’s dataset is skewed by the queries its own users choose to run, and the ordering matters far more than any single coefficient.

Reading the table

Correlations here run negative because a better position is a smaller number. Position 1 beats position 30, so a factor that rises as position improves shows a negative coefficient. The further from zero, the stronger the relationship.

Three things stand out, and they are what we would look for in any vendor’s table rather than in this one alone. Vocabulary variety occupies the top in both readings, and the three leading semantic factors all count distinct terms rather than repetitions. Placement is a real second-order signal, with variations in headings at −0.23, in anchor text at −0.22 and in ALT attributes at −0.19, which says the same term is worth more in a heading than buried in a paragraph. And backlink count sits 73rd at −0.13, present but nowhere near decisive, which is not an argument against links so much as a reminder that on-page vocabulary is the part you fully control.

Why exact-match repetition sits near zero

Density measured on the exact phrase registers at −0.02. Term frequency, which counts occurrences of the exact term rather than dividing them by length, registers at −0.21 and is real.

That gap is the finding, and it is the one thing in the table we would still believe if the numbers moved. Repeating your head term a fortieth time moves nothing measurable. Adding a term the page has never used moves the diversity factors that sit at the top of the table. Two pages can post the same density score while one covers thirty distinct terms and the other hammers eight, and only the first one reads as coverage.

The industry drew a lazy conclusion from near-zero density figures like these, namely that the core keyword no longer matters. That reading is wrong and acting on it costs rankings. The exact core keyword has to be present in the title tag, in the H1, and in the body. A page missing it from the title or the H1 is very difficult to rank however rich its surrounding vocabulary. What the data cannot tell us is the required amount, because no density threshold for the core term has ever been found and we do not assert one. Presence is the prerequisite. Variation is the driver.

The worked example: dispensary, cannabis store, weed shop

The search demand hiding outside the head term

Here is what the vocabulary gap looks like with real numbers. These are US monthly search volumes pulled from Semrush on September 6, 2026. Treat them as ordering, not as traffic forecasts, because vendor volume estimates are approximations.

A dispensary page optimized for “dispensary” alone addresses 201,000 monthly searches. The same page, written to cover the vocabulary above, sits in front of roughly 2.4 million. Most of that sits in the near-me family, which is a local ranking problem as much as a content one, but even setting the near-me cluster aside the non-local variations carry about 430,000 monthly searches against 201,000 for the head term.

Keyword difficulty falls as the phrasing gets specific

Read the difficulty column next to the volume column and a second pattern appears. “Dispensary” scores 91. “Weed shop near me” scores 24 at 27,100 searches, and “dispensary menu” scores 21 at 480. The variations a competitor ignored are usually the ones a new page can win first, and they feed the same commercial intent as the term nobody can rank for yet.

There is a third reason to place them that has nothing to do with the individual queries. Each of these terms teaches Google something about the page. A page carrying dispensary, budtender, cannabis club, recreational and medical is legible as a cannabis retail page in a way that a page carrying dispensary forty times is not. The variations rank for themselves and they also raise the page’s claim on the head term, which is what the diversity correlations describe.

How to find keyword variations

The variations belong to the market, not to your imagination, so pull them rather than brainstorm them. A brainstormed list cannot be failed, which is exactly what makes it useless as a target.

Start with the tools that report search volume

Semrush, Ahrefs and Google Keyword Planner each return phrase-match and broad-match reports around a seed term, with monthly searches and keyword difficulty attached. Moz and Ubersuggest cover the same ground. Run three or four seed phrasings rather than one, because each seed surfaces a different corner of the vocabulary: dispensary returns one set, weed shop returns another, and the union is the pool the page draws from. Record the volume with every term, since volume decides which variations earn a heading.

Then read the SERP itself

Google Autocomplete, the People Also Ask box and the related searches at the foot of the results page are Google telling you which phrasings it associates with the query. Google Trends separates seasonal spikes from durable search demand. Google Search Console is the most underused source of all: the queries report already lists terms your pages receive impressions for without ranking well, and those are variations you have half-earned and never placed deliberately. Answer the Public turns a seed into question forms. Bing Webmaster Tools reports a smaller but differently shaped query set.

Filter before you write

A raw pull is mostly noise. Strip the queries that share a string and not an intent, the competitor brand terms, the URL artifacts, and the conversational strings with no measurable volume. Hold the geographic modifiers back for city pages instead of stuffing them into a national one. Then cluster what remains by intent and decide, cluster by cluster, whether it belongs on this page. In the dispensary pull above, the near-me family is a real cluster with enormous search demand and it belongs on a local page, not on a national guide.

Where to place SEO keyword variations

The placement rows are worth reading as instructions. Variations in H1 to H3 correlate at −0.23, higher than the whole-page variation count itself. Variations in link anchor text at −0.22 outrank entities in H2 tags at −0.20. Variations in ALT attributes reach −0.19, which most content teams leave empty.

The surfaces that carry extra weight

Practically, that gives every page six surfaces where a term is worth more than it is in a paragraph: the title tag, the H1, the H2 and H3 headings, the anchor text of internal links, the alt attributes on images, and the meta description. Subheadings carry more of this load than most writers assume, and internal links carry it twice, since the anchor text describes the destination and the surrounding sentence describes the source. A dispensary page that puts “cannabis store” in an H2, links out with the anchor “recreational dispensary menu”, and captions its storefront image with alt text naming the product category has placed three variations in weighted positions without writing a single extra paragraph.

Volume order decides which variations get headings

The rule we work to: place the highest-volume variations in headings, in volume order, and stop when the headings are full. Terms below that cutoff go into body prose or nowhere. A heading exists to tell a reader what the section covers, and a heading written to hold a keyword rather than to describe a section fails at the only job it has. Schema markup does not replace any of this; it describes the page to a parser, while the headings describe it to a reader and to the ranking systems reading over that reader’s shoulder.

How this gets measured

Everything above is a claim about vocabulary breadth, which means it can be counted. We measure it with two figures. Contextual density counts every occurrence of the subject’s vocabulary against the page’s useful word count. Unique contextual density counts each term once, no matter how often it recurs. Divide the first by the second and you get a repetition index that tells you whether a page reached its score by covering a subject or by hammering a handful of terms. An index near 2 or 3 describes a page that varies its vocabulary as it goes. An index of 5 or 6 describes a page that returned to the same small set repeatedly.

The full method, the formulas, and the limits of what any of it proves are set out in our plain-language guide to contextual density and how a page’s vocabulary is measured. The short version is that keyword variations are one of three inputs, alongside entities and LSI terms, and that all three count distinct terms rather than repetitions.

Measure against the market, not against yourself

Before writing, we extract the vocabulary from the pages already ranking rather than inventing a list. A term list written from your own draft measures your own writing, which makes the comparison circular and the score meaningless. Pull the variations from a keyword tool with a volume figure attached to every term, extract the entities and supporting vocabulary from the top-ranking pages, then measure your draft against that merged list with the same instrument you used on the competitors. Anything else is scoring yourself against your own vocabulary.

Four mistakes worth naming

Confusing variation with keyword stuffing

Treating variation as keyword stuffing is the first. The two are opposites. Keyword stuffing repeats one phrase; variation replaces repetitions with distinct terms, which lowers term frequency while raising coverage. If adding a variation feels like stuffing, the sentence is probably wrong rather than the term.

Clustering by string instead of by search intent

Clustering by string similarity instead of intent is the second. “Cannabis store” and “cannabis stocks” share a word and nothing else. Group by what the searcher wants, and when a cluster’s intent diverges from the page’s, give it its own page instead of forcing it in.

Splitting variations across pages that then compete

Building a separate page for every variation is the third, and it produces keyword cannibalization: two pages of yours competing for one query, splitting the signals that would have ranked either of them alone. Variations of one intent belong on one page. When two near-duplicate pages already exist, consolidate them and set a canonical rather than optimizing both.

Padding to a word count

Padding to a word count is the fourth. Word count correlates at −0.21 and we report it because it is in the export, but we read the causation as running the other way. Covering a subject properly requires a certain number of terms, and placing those terms in readable sentences takes words. The word count of a strong page is the residue of its coverage, not the reason it ranks, which is why padding a thin page to hit a length target reliably does nothing.

What this does not prove

Correlation is not causation, and we would rather say so than imply more than the data carries. Pages rich in vocabulary tend to be better pages in ways a correlation tool cannot separate from the vocabulary itself. Every correlation dataset on the market, the one quoted above included, is built from the queries its own users choose to run, which skews toward the US and English-language search, and none of them publishes a sample size. A coefficient from any single export is a hint, not a measurement. Google changes its ranking systems constantly, reporting 3,234 launched improvements out of 654,680 experiments in the last year for which it published complete figures.

Two habits keep the work honest after publication. Track impressions and click-through rate per query in Search Console rather than position alone, because a page gaining impressions on twenty new variations is working even before any of them reach page one. And repeat the competitor research when the market moves, since content optimization against a snapshot from a year ago optimizes toward a SERP that no longer exists. Natural language processing on Google’s side keeps changing what counts as a related phrasing; the vocabulary that describes a subject changes more slowly, which is why the term list ages better than the tactics.

What survives all of that is the ordering, and it survives across sources rather than inside one. Across both exports quoted here, across the published studies we read alongside them, and across our own before-and-after measurement on client pages since 2025, the factors that count distinct terms sit at or near the top and raw repetition sits at the bottom. That pattern has outlived every algorithm update in between, which is the part worth building on. If a vendor’s next export moves a coefficient by three hundredths, nothing above changes.

---

If you want to know which variations your pages are already covering and which ones your competitors own, we run the measurement as part of a content audit, or you can start with a free SEO audit. For cannabis retailers specifically, our cannabis SEO work and the dispensary case study show what the vocabulary gap is worth once it is closed.

FAQ

What are keyword variations in SEO?

A keyword variation is an alternate phrasing of a keyword that carries the same search intent. Cannabis store, weed shop and dispensary are variations of one another because a searcher typing any of them wants the same result. Variations include synonyms, morphological forms such as plurals and tenses, long-tail phrasings, question forms, and common misspellings.

Do keyword variations actually help rankings?

In the on-page correlation data we track, the number of unique keyword variations a page uses correlates with ranking position at around −0.24, and variations in H1 to H3 tags at −0.23, while density measured on the exact phrase sits near zero. Those are correlations across pages Google already ranks, not proof of causation, and no vendor publishes a sample size. The ordering is what has held across readings eighteen months apart.

How many keyword variations should a page use?

There is no universal number, and anyone quoting one is guessing. The workable target is the coverage of the pages already ranking for your term, measured against the same list, plus a margin. A competitive commercial query may demand 180 terms across variations, entities and supporting vocabulary; a narrow local query may demand 60.

Are keyword variations the same as keyword stuffing?

No, they are opposites. Keyword stuffing repeats one phrase until the page is unreadable, which raises term frequency and moves nothing in the correlation data. Variation replaces repetitions with distinct terms, which lowers frequency and raises the diversity factors that sit at the top of the table.

Where should keyword variations be placed on a page?

The title tag and H1 carry the core keyword. Variations go in the H2 and H3 headings (−0.23), the anchor text of internal links (−0.22), the ALT attributes of images (−0.19), the meta description, and the body copy. Place the highest-volume variations in headings first, in volume order.

How do you find keyword variations?

Pull them from a keyword tool with volume attached to every term, using three or four seed phrasings rather than one. Semrush, Ahrefs and Google Keyword Planner cover the phrase-match and broad-match reports; Google Autocomplete, People Also Ask, related searches and Google Trends show what Google associates with the query; Google Search Console lists the terms your pages already receive impressions for. Then filter out the queries that share a string without sharing an intent.