Keyword density is the percentage of the words on a page taken up by the keyword you want to rank for. You calculate it by dividing the number of times the keyword appears by the total number of words in the text and multiplying by one hundred.
We measured that number on 29 articles from the Italian edition of this blog, counting every exact occurrence of the focus keyphrase set in Yoast. The median is 0.3%. Nineteen out of twenty-nine sit below 0.5%, the threshold the plugin flags as too low. None reaches 3%. And the variable that explains almost all the difference from one article to the next is not how they are written: it is how many words the keyword has.
How to calculate keyword density
The formula is a division: (keyword occurrences ÷ total words on the page) × 100. On a 2,000-word text, ten occurrences give you 0.5%.
It gets more complicated with multi-word keywords, because tools don't all count the same way. Some divide occurrences by total words. Others first multiply the occurrences by the number of words in the keyphrase, since a four-word phrase takes up four slots in the text, not one. Same article, same keyword, and the two methods return numbers that differ by a factor of four. For the measurements in this article we used the second method, which also matches the behaviour Yoast describes: the longer the keyphrase, the fewer occurrences you need to turn the light green.
Before measuring anything, it is worth asking whether the keyword is the right one, and that is decided during keyword research by looking at search intent, not once the text is finished.
The percentages going around were not written by Google
When someone tells you the ideal density is 1%, they are repeating a plugin threshold, one of the most persistent SEO myths. The figures that circulate in the industry have a precise, documented origin, and it is not Google's documentation.
| Source | Stated range | What happens outside the range |
|---|---|---|
| Yoast SEO (free version) | 0.5% to 3% | Green light only inside the range; counts exact matches only |
| Yoast SEO Premium and Shopify | up to 3.5% | Recognises word forms of the keyphrase (train, trained, training) |
| Rank Math | 1% to 1.5% | Full score only inside the range; over-use warning above 2.5% |
| All in One SEO | 0.5% or more | Gives 2% as the point to watch for keyword stuffing |
| Google documentation | no percentage | Keyword stuffing is defined in words, never with a numeric threshold |
The sources are the plugins' own pages: Yoast states that "in the free version of Yoast SEO for WordPress, you'll get a green traffic light if your keyphrase density lies between 0.5 and 3%", while the Rank Math knowledge base says that "a keyword density of 1-1.5% is sufficient in most cases" and scores accordingly. These are product thresholds, chosen by whoever wrote the plugin. Neither comes from anything Google has said.
What Google actually says
In Google Search's spam policies, keyword stuffing is defined as "the practice of filling a web page with keywords or numbers in an attempt to manipulate rankings". Three examples are listed: lists of phone numbers without substantial added value, blocks of text listing cities and regions, and repeating the same words or phrases "so often that it sounds unnatural". There is no percentage, and there never has been: the test is qualitative.
Search Central's page on creating helpful content includes a self-assessment question about word count that applies here too: "Are you writing to a particular word count because you've heard or read that Google has a preferred word count? (No, we don't.)" The same logic holds for density.
The bluntest answer comes from John Mueller, Search Advocate at Google. At the end of December 2021, in a Reddit thread asking whether keyword density was still an SEO factor, he replied with a flat "no". The exchange was reported by Search Engine Roundtable on 7 January 2022, which notes that Mueller had been saying the same thing for about ten years.
The test: 29 articles measured one by one
Statements only go so far. We preferred to measure. The sample is the archive of Visilay's Italian blog: every published post with a focus keyphrase filled in and at least 3,000 characters of text, 29 articles in all, averaging about 2,600 words each.
The method, so you can repeat it: content stripped of HTML tags and shortcodes, text lowercased with accents removed, tokenised on alphanumeric sequences, exact occurrences of the keyphrase counted as a contiguous sequence of words, density calculated as occurrences multiplied by the number of words in the keyphrase, divided by total words.
The aggregate results:
- median density: 0.30%
- highest value across the whole archive: 2.15% (the article on featured snippets)
- articles below 0.5%, i.e. below Yoast's minimum: 19 out of 29
- articles above 3%, i.e. in the red: none
- articles with an exact density of 0.00%: 7
A different measurement of the same archive, with a different method, is in our article on keyword stuffing: there we took 90 articles and, for each one, the most repeated term overall rather than the focus keyphrase, and the values go up to 6.62%. The two figures do not contradict each other; they measure different things. The most frequent single word in a text and the exact phrase you are trying to rank for are never the same thing.
The interesting part comes when the 29 articles are grouped by the number of words in the keyphrase. Density drops steadily with each word added, and the writing has nothing to do with it.
| Words in the keyphrase | Articles | Median density | Mean density | Maximum density |
|---|---|---|---|---|
| 1 (e.g. "serp", "geo") | 2 | 1.70% | 1.70% | 1.90% |
| 2 (e.g. "keyword stuffing") | 9 | 0.42% | 0.74% | 2.15% |
| 3 (e.g. "intento di ricerca", search intent) | 8 | 0.25% | 0.31% | 0.83% |
| 4 (e.g. "quanto costa la seo", how much SEO costs) | 10 | 0.14% | 0.26% | 1.33% |
From one-word to four-word keyphrases, median density falls twelvefold, from 1.70% to 0.14%. The articles are written by the same people, with the same method, at the same length. The only thing that changes is the keyword. Which means the percentage measures the shape of your keyword, not the quality of your text.
Why it happens: maths, not style
A four-word keyphrase is a contiguous sequence. For the counter to register it, those four words have to appear in exactly that order, and in a text written to be read that rarely happens: the writer says "how much does SEO cost" once in the title and then moves on to "the budget", "the monthly spend", "the cost of a project". That is not laziness, it is how language works.
There is a deeper reason underneath, and it is the way information retrieval systems handle term frequency. In the BM25 model, described by Stephen Robertson and Hugo Zaragoza in The Probabilistic Relevance Framework: BM25 and Beyond (Foundations and Trends in Information Retrieval, volume 3, issue 4, 2009), a term's contribution to relevance saturates: the first occurrences count, later ones add less and less, and the curve flattens. That behaviour is built into the formula, not an editorial choice by Google. Repeating a word twenty times instead of five does not multiply relevance by four; it adds a fraction that tends towards zero.
BM25 is a 2009 model and Google has used far more complex systems for years, but frequency saturation is a property later models have kept or strengthened. If you want to know which signals really matter, our guide to ranking factors gives the wider picture.
The US data, and how far it travels
There is a large-sample measurement for the US market. Rankability analysed 1,536 Google results across 32 competitive keywords and found an average density of 0.04% for positions 1 to 10, against 0.07% for positions 11 to 20 and 0.08% for positions 21 to 30. The sample is American, so the figure should not be read as a threshold to copy for the UK or anywhere else. Read it for its shape: the pages that rank best do not have higher density, and the relationship between the two is not a straight line.
Then there is a language problem. Exact-match counters lose inflected forms along the way, and the more a language inflects, the more they lose. English inflects less than Italian, but plurals and verb endings still slip through ("page" and "pages", "optimise" and "optimising"). In our Italian sample, the article with the focus keyphrase "obiettivi SEO" (SEO goals) has one exact occurrence but four once inflected variants are allowed. The featured snippets article goes from 27 to 30 occurrences. The one on "motori di ricerca alternativi" (alternative search engines) goes from zero to one. These differences move the plugin's traffic light without changing a comma of the text. The paid version of Yoast recognises word forms; the free one, which is what most sites run, does not.
When the percentage still tells you something
The metric is not useless, it is just used backwards. As a target it does nothing. As a symptom, in three cases, it does.
- Density above 3% on a short text. On a 200-word product page, six repetitions of the same phrase are noticeable when you read it. Here the number anticipates what a human editor would say anyway, and it takes you straight into black hat SEO territory without you having decided to go there.
- Exact density of zero on a page that should be about that thing. You don't need to raise it, you need to understand why: often it means the page covers a different subtopic from the one you are pushing it for, and the problem is in the on-page set-up, not the frequency.
- Identical density across many pages of the same site. This is the case we see most often in audits, and we come to it shortly.
What we look at instead of the percentage
Three checks, in this order. The first is coverage: which subtopics the ranking pages cover and which are missing from yours, which is a job of building the content, not counting. The second is prominence: where the keyword appears, i.e. whether it is in the title, the URL, the first paragraph, at least one H2 and the title tag. The third is named entities: companies, tools, people, places, numbers with a source. They are the signal that separates a text written by someone who does the job from one rewritten by someone who has read the others.
An example from our archive, measured rather than told. The Italian article on SEO for manufacturing has an exact density of 0.00% on its focus keyphrase: the full phrase never appears as a contiguous sequence. In the three months between 1 June and 28 August 2026 that page had an average position of 13.65 on Google Italy according to Search Console, one of the best on the whole blog. It contains the case of Macropix, a Milan manufacturer of custom LED screens that went from position 88 to position 2 for the main keyword in its market. The work that produced that result never had anything to do with how often a word appeared.
If the question is where to start on a real site, we cover it on our SEO services page and in the other projects we have published with the numbers included.
The only use of density that has helped us
Measured on a single page, keyword density tells you nothing. Measured across a whole archive, it tells you something no other check shows as quickly: which pages are chasing the same keyword. If a site with two hundred articles has eight with density above 1% on the same phrase, you don't have eight optimised pages, you have cannibalisation to fix before writing anything else. It is a diagnostic, comparative use, applied to the whole set, and it takes a two-minute database query. It is also why we still keep that field filled in, while we have ignored the traffic-light threshold for years. The rest of the work goes through internal links and the choice between long-tail and short-tail keywords, not through a decimal point.
Keyword density FAQs
There is no ideal value set by Google: no percentage appears anywhere in its official documentation. The ranges that circulate are plugin product thresholds (0.5-3% for free Yoast SEO, 1-1.5% for Rank Math). Across the 29 Visilay blog articles we measured, the median is 0.3%, and nineteen out of twenty-nine are below Yoast's minimum.
No. Google's John Mueller answered no to the direct question in a Reddit thread at the end of 2021, reported by Search Engine Roundtable on 7 January 2022, and he has given the same answer for more than ten years. In Google's spam policies keyword stuffing is defined qualitatively, with no numeric threshold.
Divide the keyword's occurrences by the total number of words and multiply by one hundred. With a multi-word keyphrase, tools disagree: some count occurrences, others first multiply them by the number of words in the phrase. The two methods can differ by a factor of four on the same text, so before comparing two measurements, check which formula the tool uses.
Not before checking how many words your keyphrase has. In our sample, median density goes from 1.70% for one-word keyphrases to 0.14% for four-word ones, with the same authors, method and text length. If the phrase is long, the plugin's orange light is describing the shape of the keyword, not a flaw in the content.
It depends on the tool. The free version of Yoast SEO counts exact matches only, while Yoast Premium and the Shopify version recognise word forms. In a heavily inflected language such as Italian the difference is noticeable: on one of our articles occurrences go from 1 to 4 just by allowing inflected variants of the same keyphrase.
It is not the percentage itself that gets penalised but the behaviour that usually produces it. Google acts on keyword stuffing, meaning text where keywords appear in lists, out of context or repeated to the point of sounding unnatural. If the text reads well, the number is not a problem; if the number is high because the text is unreadable, the problem is the text.
