The short version
- Keyword density has never been a ranking factor. There is no target percentage, and there never was one.
- Semantic coverage — genuinely addressing the concepts a topic entails — is real and is what density was a crude proxy for.
- Writing to hit a density figure reliably makes content worse, which is the opposite of the intended effect.
- The practical test: could an expert read your page and find nothing important missing? That is coverage.
Keyword density is the most persistent myth in SEO. It survives because it is quantifiable, because plugins display it, and because in the very early days of search it correlated loosely with relevance. It has not been a useful target for well over a decade, and following it actively damages content.
What density was crudely gesturing at is real: a page about a topic should contain the language and concepts that topic involves. The difference is that coverage is about completeness of ideas, not repetition of a phrase.
The short answer
Fit for your workflow
Ignore density entirely. Aim for coverage — completeness, not repetition.
Write the page a knowledgeable person would write for someone with the question. The relevant terminology appears naturally because you cannot explain the topic without it. If you find yourself inserting a phrase to hit a number, you have stopped writing for the reader and started writing for a widget that does not influence rankings anyway.
- Writing new content
- Write it properly first. Check coverage after, never before.
- A page that will not rank
- Check for missing subtopics, not missing keyword instances.
- Using an optimisation tool
- Treat its term list as a gap checklist, not a quota.
The difference, concretely
The fifth row deserves care, because it is where optimisation tools overclaim. Semantic coverage is not itself a ranking factor with a dial attached. It is a characteristic of pages that genuinely answer a question well — which is what actually gets rewarded. Covering more terms does not cause ranking; being genuinely comprehensive tends to produce both.
How to use optimisation tools without being damaged by them
Surfer, Clearscope and similar tools analyse what currently ranks and suggest terms your page lacks. That is genuinely useful as a gap analysis — if every ranking page discusses a subtopic you have not mentioned, that is worth knowing.
- Write the page first, without the tool open. Anything else produces content shaped by the scoring model rather than the subject.
- Run the analysis afterwards and read the missing terms as questions: is there a real subtopic here I have not addressed?
- Add the ones that are genuinely missing. Ignore the rest.
- Never optimise to a target score. The last ten points are almost always achieved by making the page worse.
- Remember the corpus is the current top ten, which means the tool is optimising you toward the average of your competitors — not toward being better than them.
The clearest sign you have the balance right is that the relevant terminology appears without your having thought about it. If you are writing about canonical tags, the words "canonical", "duplicate" and "indexing" will occur because the subject requires them. If you have to check whether they occur often enough, you have stopped writing about canonical tags and started writing about a term list — and readers can tell the difference immediately, even when they could not articulate why.
The regression-to-the-mean problem
Optimisation scores are computed against pages that already rank. Score highly and you have produced something statistically similar to the existing top ten. That is a floor, not a differentiator — and it is precisely the kind of content a generative summary replaces most easily. See what changed in the AI era.