On-page mechanics
Keyword density is not a target — it is a warning light
There is no ideal percentage. Density is a diagnostic that tells you a page may be repetitive; the fix is almost never "remove three mentions".
7 min read
Keyword density is the share of the total words on a page that a given term accounts for. It is trivial to calculate, which is why it survived as a metric long after it stopped being a ranking factor worth optimising against.
The number itself has not been a lever for a very long time. Search engines moved to term weighting, synonyms and entity understanding, and the old advice to hit two or three percent now mostly produces text that reads like it was written for a machine — which is precisely what the helpful-content guidance targets.
That does not make the measurement useless. A high reading is a reliable symptom of something a reader will notice: repetition, thin phrasing, or a page trying to cover one narrow term instead of a topic. This article covers how to read the number, what it actually indicates, and what to fix instead.
What density is, and what the number means
Keyword density is the number of occurrences of a term divided by the total word count of the page, expressed as a percentage. A 1,500-word page mentioning a phrase fifteen times has a density of one percent.
There is no target value, and no search engine publishes one. Google has been consistent for years that repeating words does not help rankings and that writing for search engines rather than people is the failure mode to avoid. Treat any tool that reports "optimal: 2.5%" as reporting an opinion, not a specification.
| Reading | What it usually means | Action |
|---|---|---|
| 0% | The term never appears — often a mismatch between the target and the text | Decide whether the page is really about this term |
| 0.2-1.5% | Normal range for natural writing on-topic | Nothing. Do not add mentions to reach a number |
| 1.5-3% | Repetitive phrasing, or a short page with a long exact-match phrase | Read it aloud; vary the phrasing where it grates |
| Over 3% | Almost always exact-match padding or a thin page | Rewrite the sections, do not delete individual words |
Why the percentage stopped being a lever
Term weighting was never linear
Even classic information-retrieval scoring never rewarded raw repetition. TF-IDF and BM25 both apply saturation: the tenth occurrence of a term contributes far less than the second, and BM25 additionally normalises for document length. A model built on those principles cannot be gamed by doubling mentions.
Modern ranking sits several layers above that, matching meaning rather than strings, which is why a page can rank for queries whose exact wording never appears on it.
Synonyms and entities count
A page about "keyword density" that also discusses term frequency, repetition, over-optimisation and TF-IDF is more clearly on-topic than one that repeats the head phrase thirty times. Counting one string ignores everything that establishes the subject.
The visible cost is human
Over-repetition reads badly, and reading badly has measurable consequences — people leave. The helpful-content guidance explicitly asks whether text repeats phrases because it was written for search engines. A high density reading is the cheapest way to notice that before a reader does.
How to calculate it properly
One consequence worth remembering: density falls as a page grows. Adding 500 words of genuine substance lowers the percentage without touching a single mention — which is why "reduce your density" is bad advice phrased as a metric.
- Count body copy only — Exclude navigation, footer, cookie banners and boilerplate. Site-wide chrome can add dozens of matches on a short page and make the figure meaningless.
- Decide phrase or word before you count — For a two-word phrase, one occurrence is normally counted as one instance against the total word count. Counting it as two words inflates the result and is where most tool disagreements come from.
- Count variants separately — Singular, plural and reordered forms are different strings. Track them as a group to understand coverage, not as one number to optimise.
- Compare against the page, not a benchmark — The useful comparison is your own draft before and after an edit, or the same term across your own pages competing for a query.
What to fix when the reading is high
A high reading points at a writing problem. Deleting occurrences until the number falls treats the symptom and usually leaves the text worse, because the deletions land on the sentences where the term was doing real work.
| What the repetition looks like | What is actually wrong | The fix |
|---|---|---|
| Exact phrase in every heading | Headings written for a crawler, not a reader | Keep the term in two or three headings where it is natural; make the others describe their section |
| The phrase repeated in consecutive sentences | No pronouns, no synonyms — machine cadence | Alternate between the term, a natural synonym and a pronoun |
| Term stuffed into the first 100 words | Cargo-culted intro advice | One natural mention in the opening is enough; the intro is for the answer |
| High density on a 400-word page | The page is thin, not over-optimised | Add substance — examples, data, edge cases — and the ratio corrects itself |
| Alt text and captions repeating the phrase | Alt text written for SEO instead of description | Describe the image; the term belongs there only when it is in the image |
| Several pages all dense on the same term | Cannibalisation, not density | Merge or differentiate the pages — an audit decision, not an editing one |
Where the term genuinely matters
Placement carries more weight than frequency. A handful of mentions in structurally prominent positions establishes the subject better than thirty scattered ones.
- The H1 and the title tag: the page subject should be recognisable from both without a keyword being forced in.
- The first sentence or two: an answer-first opening naturally contains the subject.
- Two or three H2s where the section really is about that aspect.
- The URL slug, if it does not require an awkward string.
- The meta description, because it is a click decision, not a ranking one.
- Once in an image alt attribute, only when the image genuinely depicts the subject.
Over-optimisation: the real risk
Keyword stuffing remains a documented spam policy: filling pages with repeated terms, lists of keywords or out-of-context phrasing to manipulate rankings. Ordinary editorial repetition does not fall under it, and a page at 2% density is not spam.
The realistic risk is duller than a manual action. Text that reads as machine-written performs worse with the people who land on it, and after the helpful-content work it also competes against a system explicitly asking whether content was made for search engines first. That is the cost you are actually paying, and it does not show up as a penalty notice.
A better set of questions than "what is my density"
Paste a draft into the free optimizer if you want the mechanical view of this: it reports the per-term read-out alongside heading structure, length and keyword placement, so a high number can be read next to the reason it is high.
- Does the page cover the subject or just the string? — List the subtopics a knowledgeable reader would expect. Missing ones matter far more than the frequency of the head term.
- Can each section be quoted on its own? — Extraction cares about self-contained answers, not term counts.
- Would you read it aloud without wincing? — The read-aloud test catches every case of harmful repetition and needs no tool.
- Is one page competing with another of yours? — Duplicate targeting explains more lost positions than any density figure.
Questions people ask
What is a good keyword density for SEO?
- There is no good value, because no search engine targets one. Natural writing on a focused page usually lands under 1.5%. Treat anything higher as a prompt to read the text again, not as a number to correct.
How do you calculate keyword density?
- Divide the number of occurrences of the term in the body copy by the total word count and multiply by 100. Exclude navigation and footer text, and count a multi-word phrase as one occurrence.
Does keyword density still matter in 2026?
- As a ranking lever, no. As a diagnostic, yes — a high reading reliably indicates repetitive phrasing, a thin page, or exact-match padding, all of which are worth fixing for reasons that have nothing to do with the percentage.
Can low keyword density hurt a page?
- Only if the term is missing where it matters — the title, H1 and opening. A page that never names its own subject is ambiguous. Adding occurrences beyond that does nothing.
Is keyword stuffing still penalised?
- Yes. Keyword stuffing is part of Google's spam policies and covers repeating words and phrases to manipulate rankings. Normal editorial repetition is nowhere near that threshold.
References
- Google Search Central — Spam policies: keyword stuffing — The documented definition of keyword stuffing and what it covers.
- Google Search Central — Creating helpful, reliable, people-first content — Includes the self-assessment question about repeating phrases for search engines.
- Google Search Central — SEO Starter Guide — Baseline guidance on descriptive text and avoiding text written for engines.
- Introduction to Information Retrieval — Term weighting — Why term-frequency contributions saturate rather than scale linearly.
Keep reading
- The checklist item this replaces — The keyword group of the on-page checklist, with pass conditions instead of percentages.
- Deciding whether a thin page is worth saving — Where repetition is a symptom of a page that needs merging or pruning.
- A density read-out with no target attached — What the optimizer reports about your terms, and why it never asks you to add mentions.