Contextual density, explained: how we measure whether a page really covers its subject

Ask a periodontist to talk about dental implants for two minutes, then listen to the words that come out. Osseointegration. Bone graft. Titanium post. Abutment. Sinus lift. Immediate loading. Peri-implantitis. She never planned that vocabulary. It arrived because she genuinely knows the subject.

Now ask someone who skimmed one article about implants last night. They will say “dental implants” a great many times.

Anyone listening can hear the difference in about fifteen seconds. Google cannot hear it, because Google is not listening the way a person does. It is counting. Contextual density is our attempt to count the same thing a human ear picks up instantly: whether the page in front of you was written by someone who owns the subject.

What contextual density measures

Contextual density measures how much of a page’s meaningful text is built from the vocabulary of its subject.

Three kinds of words count toward it.

Keyword variations are the core phrase and its relatives: word combinations, synonyms, singular and plural forms. For a Montreal implant clinic: dental implants, dental implant, single-tooth implant, implant dentistry, tooth replacement.

Entities are recognized names and concepts that belong to the subject, the ones with established recognition: brands, places, people, tools. Nobel Biocare. Straumann. CBCT scan. All-on-4. Our guide to entity SEO covers what entities are and where they belong on a page.

LSI terms are the supporting contextual words, the actions, processes, descriptions and technical terms that surround the subject. Osseointegration. Bone graft. Healing cap. Abutment. Crown. Edentulous.

Count how often all of those terms appear on the page. Divide by the page’s useful words, meaning every word except the connective tissue that carries no subject meaning (a, an, the, of, in, on, and, or, is, are). Multiply by 100.

The result is a percentage. It answers one question: how much of this page is actually about its subject, rather than words arranged around it.

The formula, exactly

Contextual Density (CD) = (Σ(KV) + Σ(RE) + Σ(LSI)) / (TW − SW) × 100

Σ(KV) is the total number of times keyword variations appear, repetitions included. Σ(RE) is the total number of times relevant entities appear, repetitions included. Σ(LSI) is the total number of times relevant LSI terms appear, repetitions included. TW is the page’s total word count. SW is its stop-word count.

Those three sums count every occurrence, not every distinct term. That distinction matters more than it looks, and the section on repetition below is where it becomes a problem worth solving.

Two pages, the same score, and only one worth reading

Here is where a single number starts to mislead, using arithmetic simple enough to check on paper.

Two implant pages, each 1,000 useful words. On both pages, subject vocabulary appears 90 times. Both therefore score a contextual density of 9%.

Page A got there with 30 different terms, each appearing about three times. It walks through the consultation, the CBCT scan, the bone graft, the titanium post, osseointegration, the abutment, the crown, and what peri-implantitis looks like if aftercare slips.

Page B got there with 8 terms. “Dental implants Montreal” appears 40 times on its own.

Identical scores. One page a periodontist would sign. The other reads like a page trying to convince a machine.

So contextual density on its own cannot separate them, and any agency quoting you a single density figure is showing you a number that hides this. Two further measurements fix it.

Unique contextual density counts each term once, no matter how often it recurs. Page A scores 3%. Page B scores 0.8%. The gap that the density figure concealed appears immediately.

The repetition index divides one by the other and answers a plain question: on average, how many times does each covered term repeat? Page A sits at 3. Page B sits at 11. A page in the 2 to 3 range varies its vocabulary as it goes. A page at 6 or higher is leaning on a handful of phrases.

The second and third formulas

Unique Contextual Density (UCD) = (U(KV) + U(RE) + U(LSI)) / (TW − SW) × 100

U(KV), U(RE) and U(LSI) count distinct terms present in the text, each once, regardless of repetition. TW and SW are identical to the CD formula. Only the numerator changes.

Repetition Index = CD ÷ UCD

UCD carries its own bias, in the opposite direction from CD. Its denominator is length, so it flatters short pages. A 400-word page placing 17 distinct terms will post a higher UCD than a 4,000-word page placing 118, even though the longer page covers roughly seven times more of the subject. Neither figure is trustworthy alone. Read together, with the repetition index between them, they describe a page fairly.

Contextual density is not keyword density

These two get confused constantly, and the confusion is expensive, because the advice that follows from each is opposite.

Keyword density measures how often one exact phrase appears, divided by the total word count. It is the metric behind two decades of advice to work your keyword in at 2% or 3%.

Contextual density measures how much of the subject’s entire vocabulary appears, divided by the page’s useful words. Different numerator, different question, different answer.

A page can post a healthy keyword density and a poor contextual density. That is Page B above: one phrase hammered, the subject barely covered. The reverse also happens, and it is what a real expert’s writing looks like on a chart.

Your site is a library, and Google reads the spines

A useful picture, and one we lean on when explaining this to clients: your website is a library.

The title tag is the lettering on a book’s spine. It’s what someone scanning a shelf reads before deciding whether to pull the book down, and it’s the shortest, most compressed statement of what’s inside. The H1 is the title on the cover, where you get a few more words to say the same thing properly. The H2s are the chapter titles, and they tell a browser flipping through whether the book actually covers what the spine promised.

Now read those three from the outside in. If the spine says dental implants, the cover says dental implants, and the chapters cover consultation, imaging, bone grafting, placement, the crown and aftercare, the book is what it claims to be. If the spine says dental implants and every chapter is titled some rearrangement of “dental implants Montreal,” a browser puts it back on the shelf.

Google works the shelf the same way, and this is where contextual density stops being an abstraction. Every element on the page is a place to say what the page is about: the spine, the cover, the chapter titles, the body text, the words you use as link anchors, the alt text on your images. The pages that rank use all of them, and use them to say the same thing about the same subject. A page that puts the subject in the title and then abandons it in the chapters is a book with a misleading spine.

Where the core keyword still matters

Some of the correlation data around exact-phrase repetition sits close to zero, and the SEO industry has drawn a lazy conclusion from it: that the core keyword no longer matters. That reading is wrong, and acting on it will cost you rankings.

Our position, from thirteen years of doing this: the exact core keyword has to be present. It belongs 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 genuinely cannot tell us is the required amount. No density threshold for the core term has ever been found, by us or by anyone publishing on it, and we do not assert one. The reason the correlation reads near zero is narrower than it appears: pages that omit the term entirely are largely absent from the ranked sample being measured in the first place, so the cost of leaving it out never shows up in the numbers.

Presence is the prerequisite. Variation is the driver. Once the core term sits where it belongs, the stronger lever is the number of distinct ways you cover the subject around it.

What the correlation data shows

We test against Cora, software that measures thousands of on-page factors across the top 100 results for a query, and we compare its exports over time rather than trusting a single reading.

The numbers, and what a correlation actually is

These figures are correlations with ranking position. They run negative because a better position is a smaller number: position 1 beats position 30. A coefficient of −0.28 therefore means more of that factor tends to accompany better positions. Nothing in search correlates overwhelmingly with rank, so what matters is which factors consistently sit at the top of the list, not the absolute size of any one figure.

One decline is worth naming rather than hiding: entities in the title tag fell from −0.25 to −0.18, the largest drop of any factor between the two readings. The title tag still matters, and the core keyword still belongs there. Stacking every entity you can fit into it has stopped paying what it used to.

Source: Cora SEO correlation data, published correlation studies, and ProStar SEO in-house analysis. Exports of March 6, 2025 and August 23, 2026.

The short version, if you skip the numbers: the factors that consistently sit at the top measure how many different relevant terms a page uses, and how much its vocabulary overlaps with the pages Google already ranks. Repeating one phrase does not appear anywhere near the top.

Breadth times density, not breadth or density

The practical target is a page that is both broad and efficient, and the two pull against each other.

Breadth without density is a page that mentions everything once, buried in forty thousand words of filler. Density without breadth is Page B, repeating six phrases until the percentage looks respectable. Both score badly on the combination, and the combination is what we measure.

The useful way to think about it: a page cannot fake density with padding, because adding filler lowers the percentage. It can fake density with repetition. The breadth measurement is what catches that, because forty occurrences of one phrase collapse to a single distinct term.

What this changes about how you write

Nothing about this asks you to write for a machine. It asks you to write the way someone who knows the subject already writes.

Cover the whole subject, not the part that is easy. If your implant page never mentions bone grafting or aftercare, a page that does covers the subject better, and both a patient and Google can tell.

Use the real names. Say Straumann, say CBCT, say osseointegration, and explain them. Vague writing scores badly because vague writing is worse.

Vary how you say things, the way you would in conversation. Say implant, tooth replacement, single-tooth implant, restoration. Nobody repeats one noun phrase forty times when they are talking to a person.

Look at what the pages already ranking have in common, then cover more than they do. Matching the shared vocabulary of the winners is the single strongest correlation in the current data, and going past it is how you overtake them. That benchmarking work is what The SEO Contextual Index sets out in full.

What we cannot tell you

Three honest limits, because a method that only reports its strengths is marketing.

Correlation is not causation. Pages rich in subject vocabulary tend to be good pages, and good pages tend to rank. We measure the pattern with confidence. We cannot prove which way the arrow points.

The sample has biases. Cora’s dataset is built from the queries its users choose to run, which skews toward the US web and English-language search. It is not a census of the internet, and no correlation tool has one.

Google moves constantly. In the last year for which it published complete figures, it reported 3,234 launched changes to Search out of 654,680 experiments, roughly nine a day, almost none of them announced. Individual coefficients drift between readings. The ordering of the categories has held across every re-measurement we have run since 2025, which is the part we rely on.

Contextual density is also one factor among many. Technical foundations come first: a page Google struggles to crawl, index or render will not be rescued by its vocabulary. Brand recognition, links and user behaviour all carry weight we cannot influence through wording alone.

Across the 600 client sites where we have applied this method, rankings improved in roughly 80% of cases where the technical base was sound. That figure is an operational observation from our own client base, not a controlled study, and we state it as such.

Frequently asked questions

What is contextual density in one sentence?

Contextual density is the percentage of a page’s useful words that come from the vocabulary of its subject, counting keyword variations, relevant entities and relevant LSI terms.

How is contextual density different from keyword density?

Keyword density counts repetitions of one exact phrase against the total word count. Contextual density counts the whole subject vocabulary against the page’s useful words. The two metrics answer different questions and lead to opposite writing advice.

What is a good contextual density score?

There is no universal number, and any agency quoting you one is guessing. The target comes from the market: measure the pages already ranking for your keyword, and cover their shared vocabulary at a comparable density with a lower repetition index. A competitive commercial query might demand a list of 180 terms. A narrow local one might take 60.

Does the exact keyword still need to be in the page?

Yes. It belongs in the title tag, the H1 and the body. The required amount is unknown and no threshold has been established, but a page missing the term from its title or H1 is very difficult to rank.

Can I just repeat my keyword more to raise the score?

No, and the repetition index is what exposes it. Repeating a term already on the page lifts contextual density while leaving unique contextual density flat, which pushes the repetition index up. A page at 6 or higher is visibly built on a narrow vocabulary.

Is contextual density a Google ranking factor?

Contextual density is our composite metric, so Google does not measure it by that name. Its components, distinct entities, distinct LSI terms and distinct keyword variations, are the factors that sit at the top of the correlation data we test against. The May 2024 Google API documentation leak also exposed internal metrics named siteFocusScore and siteRadius, which the SEO community reads as measurements of how tightly content concentrates around a topic. That reading is consistent with our approach, though the leaked material does not say how much weight those metrics carry.

How long does contextual density work take to show results?

Expect three to four months for movement and six to twelve for significant results, assuming the technical foundation is sound. Anyone promising faster is selling something else.

---

Want to know how your pages measure?

We will run the numbers on one of them, tell you the contextual density, the unique contextual density, the repetition index and the gap against the pages outranking you, and explain what we would change. No cost, no commitment, and you keep the measurements whether or not you work with us.