Keyword Density Checker
Check keyword density in any article or webpage content. Find over-optimised and under-used terms with single words, bigrams, and trigrams analysis. Free SEO tool.
Keyword Density Checker Tool
Read the numbers knowing what they are. Density is occurrences divided by the total word count, and the colour bands — green 0.5%-2.5%, orange to 4%, red above — are RECATOOLS' own editorial thresholds. Google's spam policies define keyword stuffing by how a page reads, not by any percentage. Four limits worth knowing before you act on a figure: counting is English-only, so Chinese, Japanese, Thai and Korean text yields nothing; punctuation is stripped, so "e-commerce" is counted as "commerce"; a target keyword of four words or more is looked up in the three-word table and always returns zero; and a phrase's density uses the same word-count denominator as a single word, so a two-word phrase filling 5% of your text reads as 2.5%.
How to Use the Keyword Density Checker
Paste your content
Paste your article or webpage content into the text area. The tool accepts plain text or HTML — any HTML tags will be stripped automatically before analysis.
Enter a target keyword (optional)
If you have a specific keyword you are optimising for, enter it in the Target keyword field. The tool will highlight it in the results table and show its occurrence count and density at the top.
Click Analyse
Keyword frequency tables appear instantly. Review the Single Words, Bigrams, and Trigrams tabs. Colour coding shows at a glance which keywords are in the good range (green), borderline (orange), or over-optimised (red).
Adjust your content if needed
If any keywords show in the orange or red zones, consider replacing some instances with synonyms or related phrases. Use the Export as CSV button to save your keyword data for reporting or further analysis.
Keyword Density in 2026 — What Still Matters for Google SEO
Keyword Density in 2026: Does It Still Matter for Google SEO?
The history of keyword density in SEO is a story of excess and correction. In the late 1990s and early 2000s, web pages stuffed with identical phrases — "buy cheap flights buy cheap flights buy cheap flights" — could rank at the top of Google with ease. Search algorithms at the time relied heavily on exact keyword frequency as a proxy for page relevance. The result was a web full of unreadable, manipulative content.
Google's Panda update in February 2011 changed everything. It targeted low-quality, content-farm pages with unnatural keyword repetition and thin value — causing millions of pages to lose rankings overnight. Keyword stuffing as a strategy died. Then in 2013, the Hummingbird algorithm introduced semantic search: Google began understanding the intent behind queries rather than just matching exact words. A search for "how to fix a leaky tap" could now surface pages that never used the phrase "leaky tap" — as long as they comprehensively addressed the underlying plumbing problem.
BERT (2019) and MUM (2021) pushed this further, applying deep language models to understand context, synonyms, and nuance at scale. Google's John Mueller has confirmed on multiple occasions that keyword density is not a ranking signal — the algorithm does not count how many times a phrase appears and reward pages for higher numbers. What it does evaluate is topical authority, comprehensiveness, and whether a page satisfies user intent across its full content.
So why does a keyword density checker still matter in 2026? Because over-optimisation remains a real risk. Unnatural repetition signals low quality to both algorithms and human readers. A page where the phrase "best accounting software Singapore" appears 18 times in 800 words reads as spam to a visitor — and Google's classifiers have become very good at detecting that kind of manipulation. Density tools are also useful for catching under-use: if you are targeting a specific phrase but it never appears in your content, that is equally problematic for relevance. The goal is natural, varied language that uses your target keyword and its synonyms in appropriate proportion. A density table helps with that by showing what you actually wrote about rather than what you meant to write about: a page on coffee whose top terms are "roast," "brew," "espresso" and "caffeine" is covering a topic, while one whose top three are all the same phrase is repeating it. Note that "LSI keywords" — a phrase you will meet constantly in SEO advice — describes nothing Google does. Latent semantic indexing is a 1988 information-retrieval patent, and Google has said publicly that it does not use it. The useful idea underneath the label is topical breadth, which is what the table above shows you.
Where the 0.5%-2.5% band comes from (it is ours)
Be clear about the colours in the table above: the thresholds are RECATOOLS' own editorial choice and nothing more. Below 0.5% is grey, 0.5% to 2.5% green, 2.5% to 4% orange, above 4% red. No published study sets those numbers, no search engine endorses them, and you will find other tools using different ones with equal confidence. Google's spam policies define keyword stuffing as "filling a web page with keywords or numbers in an attempt to manipulate rankings in Google Search results" and give examples — lists of phone numbers with no added value, blocks of city and region names, the same phrase repeated unnaturally through a paragraph. There is no percentage anywhere in that definition, because the thing being described is a manner of writing, not a ratio. Treat red as "read this paragraph again out loud", not as a penalty threshold.
Keyword stuffing is listed in Google Search Essentials among the spam policies — the documents that replaced the old Webmaster Guidelines in 2022 — and the consequences described there are algorithmic ranking effects or a manual action taken by a human reviewer, which can remove a page or a site from results until it is fixed and reconsidered.
The more sophisticated modern approach is TF-IDF: Term Frequency-Inverse Document Frequency. Invented in 1972 by Karen Spärck Jones, TF-IDF does not just measure how often a word appears in your document — it weighs that frequency against how common the word is across all documents in a corpus. A word that appears frequently in your article but rarely across the web is a strong, distinctive signal of your topic. That weighting is the part this tool does not do: it has no corpus to compare against, so it cannot tell you that "the" appearing often is unremarkable while "photobioreactor" appearing often is the whole point. What you get here is raw frequency and share of text — a diagnostic you read, not a target you hit.
A density figure cannot tell you whether a sentence reads badly. It can tell you which sentence to go and read.
ASEAN Content Marketing: Writing for English and Local Language SEO
ASEAN's digital markets present unique keyword density challenges because most content teams operate across at least two languages — and search behaviour differs significantly between them.
Singapore's SEO landscape is English-first, but Singlish phrases and Singaporean colloquialisms ("hawker centre," "HDB flat," "kopitiam") rank extremely well for local queries. A page targeting Singaporean audiences may use formal English for general coverage while naturally incorporating local terms that signal geographic relevance. This creates an interesting density situation: your "English" keywords and your "local" keywords need to be balanced across the same page. Note what that does to the numbers below: a local term used twice in a long page will sit in the grey band, which is a statement about its share of the text and not about whether it belongs there.
Malaysia's content strategy is more explicitly dual-language. Bahasa Malaysia (BM) and English pages often exist as separate URLs, but some brands produce bilingual content on a single page — which creates keyword density profiles this tool will read badly. The morphological structure of Bahasa Indonesia and Malay is the reason: root words take prefixes and suffixes, so "jual," "menjual," "penjual" and "penjualan" all derive from one root but register here as four separate tokens, each with a quarter of the density. The tool has no stemmer, so it cannot put them back together. Add the four figures yourself before drawing a conclusion.
For ASEAN brands targeting both local and expatriate audiences — common in Singapore and Kuala Lumpur — the content challenge is managing two distinct keyword ecosystems on one page. Multilingual pages that mix Chinese and English (common for Singapore Chinese-language content) face density analysis limitations with English-only tools, since Chinese text is not space-delimited in the same way. This tool is optimised for English text; for multilingual content analysis, pair it with a native-language keyword tool for each additional language.
What this tool counts, what it cannot count, and where the thresholds came from
Google's spam policies define keyword stuffing without a single number in it: "filling a web page with keywords or numbers in an attempt to manipulate rankings". The percentages here are ours, not Google's.
Density is counted as occurrences divided by total words, including stop words. A phrase is measured the same way, so a two-word phrase filling 5% of your text is reported as 2.5% — read the phrase tables on their own scale.
Stop words are stripped before the single-word table is built but kept for phrases, which is why "of the" can appear as a bigram while "the" never appears as a word.
Punctuation is removed before counting, so "e-commerce" is split into "e" and "commerce" and the one-character fragment is dropped. The table will show commerce; it is the same text, counted differently than you might expect.
The tokeniser keeps a-z and 0-9 only, so Chinese, Japanese, Thai, Khmer and Korean text produces no tokens at all and the tool reports nothing. It is an English-language instrument.
A target keyword of four words or more is looked up in the three-word table and will always come back as zero occurrences. Keep the target to three words or fewer, or read the phrase tables directly.
Google's Panda update (24 February 2011) was the point at which repetition stopped paying; Hummingbird (2013), BERT (October 2019) and MUM (2021) each moved ranking further from word matching toward meaning.
TF-IDF — the weighting that asks not just how often a term appears but how rare it is elsewhere — comes from Karen Spärck Jones, Journal of Documentation, 1972. This tool does the term-frequency half and has no corpus for the other half.
"LSI keywords" are not a Google feature. Latent semantic indexing is a 1988 retrieval patent Google has said it does not use; the idea worth keeping from the label is topical breadth, which the table above shows directly.
Nothing you paste here leaves the page: the text is tokenised, counted and rendered in JavaScript, with no network call and no storage write. The CSV export is built in memory and handed to the browser.
Frequently Asked Questions
-
Keyword density is the percentage of times a specific word or phrase appears in your content relative to the total word count. The formula is: (keyword count ÷ total words) × 100. For example, if your 1,000-word article contains "SEO" 15 times, the keyword density for "SEO" is 1.5%. Stop words (common words like "the," "and," "is") are excluded from single-keyword analysis to focus on meaningful content terms.
-
There isn't one, and anybody quoting you a figure is quoting a convention. The 0.5%-2.5% green band on this page is RECATOOLS' own editorial choice, chosen to be wide enough that normal prose falls inside it; other tools use other numbers with the same confidence and none of us has a search engine behind us. Google's documentation defines keyword stuffing by how the text reads — keywords or numbers loaded onto a page to manipulate rankings — and states no percentage at all. Write for a reader, then use the table to check whether anything got away from you.
-
Yes. Keyword stuffing — the practice of loading a page with keywords in an attempt to manipulate rankings — violates Google's Webmaster Guidelines. It can result in algorithmic demotion (Panda/core updates) or a manual penalty applied by a human quality rater. Since 2011, keyword stuffing has consistently hurt rather than helped SEO performance. Use synonyms and natural language variations instead of repeating the exact same phrase.
-
Single keywords (unigrams) are individual words — e.g. "SEO," "content," "marketing." Bigrams are consecutive two-word phrases — e.g. "content marketing," "keyword density," "search engine." Trigrams are three-word phrases — e.g. "keyword density checker," "search engine optimisation." Bigrams and trigrams often represent more specific, long-tail search queries and are typically less competitive while having higher conversion intent.
-
Stop words are common function words (articles, prepositions, conjunctions) like "the," "and," "of," "in," "is," "it." They carry no meaningful SEO signal because they appear in virtually every piece of English text. Including them would skew the density table — "the" would top every list with 3–5% density. Stop words are excluded from single-keyword analysis but are retained in bigram and trigram calculations to allow natural phrase detection (e.g. "best of breed" or "in the cloud").
-
TF-IDF (Term Frequency-Inverse Document Frequency) is a more sophisticated metric than raw keyword density. It measures how often a word appears in your document (TF) weighted against how commonly it appears across a large corpus of documents (IDF). A word that appears often in your article but rarely across the web is a strong topical signal. Simple keyword density cannot make this distinction. This tool computes only the term-frequency half: it has no corpus to compare against, so it cannot weigh a word by how rare it is elsewhere. Read its figures as a check on repetition, not as a measure of topical relevance.
-
Always write for humans first. Google's explicit guidance ("Write for users, not search engines") has been consistent for over a decade and is backed by how its algorithms actually work — they evaluate signals that correlate with user satisfaction (time on page, return visits, click-through rate, absence of Pogo-sticking) not purely for keyword presence. Well-written human-first content naturally includes keyword variations, covers the topic comprehensively, and earns links — all of which also improve search rankings.
-
Not a thing, despite how often you will read the phrase. Latent semantic indexing is a real technique from a 1988 information-retrieval patent, and Google has stated publicly that it does not use it; "LSI keywords" is a label the SEO industry attached to an idea that has nothing to do with the algorithm it is named after. The idea itself is sound and much simpler than the acronym suggests: writing properly about coffee will bring in espresso, roast, brew and caffeine without anyone planning it, and a page that never mentions them is probably thin. That is something the table above shows you directly — look at what your top twenty terms are, not at whether you hit a list.
-
As many times as the writing needs and no more — there is no correct number and this tool does not know one. The green band exists so that you have something to react to, not something to hit: a page written naturally about its subject usually lands inside it without being steered, and a page that had to be steered into it generally reads like it was. If you want a rule that is actually about search, Google's is that the keyword should be present where it is genuinely descriptive — the title, the first paragraph, a heading that really covers that ground — and that repetition beyond natural usage is what the spam policies name. Writing to a percentage is the behaviour those policies describe.
-
100% free, forever. No account required, no word limit, no hidden paywalls. The analysis runs entirely in your browser — your text is never sent to our servers. RECATOOLS is free to use, with no subscription — we plan to fund it with contextual advertising, shown only with your consent. All tools work with or without ad consent enabled.
Method & sources
How it computes
Counts term frequency in the browser: strips HTML, lowercases, keeps a-z and 0-9 only, drops one-character tokens, removes stop words for the single-word table and keeps them for the two- and three-word tables, then reports each term's count and its share of the total word count as (count ÷ total words) × 100. The colour bands are RECATOOLS' own editorial thresholds; Google's spam policies define keyword stuffing by how a page reads and state no percentage.
What this tool implements
- Colour bands as shipped: grey under 0.5%, green 0.5-2.5%, orange 2.5-4%, red above 4%.
- Term frequency only. The inverse-document-frequency half of TF-IDF (Spärck Jones, 1972) is not computed, because the tool has no reference corpus — stated on the page so a reader does not mistake a raw count for a weighted one.
- The colour thresholds are declared as ours in three places: under the results, in the about section and in the two FAQs that used to attribute them to industry analysis.
- Google's definition of keyword stuffing is quoted rather than paraphrased into a number, and the page names Search Essentials and the spam policies rather than the Webmaster Guidelines, which were superseded in 2022.
- Density for a multi-word phrase uses the same total-word denominator as a single word, so an n-word phrase reports roughly 1/n of its true share of the text. The page states this rather than silently presenting both on one colour scale.
- English only: the tokeniser's a-z0-9 character class means CJK and Thai input produces no tokens at all, and there is no stemmer, so inflected Malay and Indonesian forms of one root count separately. Both are stated.
Sources
- Google Search Central. Spam policies for Google web search — 'Keyword stuffing refers to the practice of filling a web page with keywords or numbers in an attempt to manipulate rankings in Google Search results', with examples (phone-number lists without added value; blocks of cities and regions; the same phrase repeated unnaturally). No density percentage appears anywhere in the document. Page last updated 2026-08-28 UTC; fetched 2026-09-13. https://developers.google.com/search/docs/essentials/spam-policies
- Spärck Jones K. A statistical interpretation of term specificity and its application in retrieval. Journal of Documentation 1972;28(1):11-21. — the origin of inverse document frequency, the weighting this tool deliberately does not compute.
- Google. Finding more high-quality sites in search (the Panda update), Google Search Central Blog, 24 February 2011. https://developers.google.com/search/blog/2011/02/finding-more-hi…
- Google. Understanding searches better than ever before (BERT in Search), 25 October 2019. https://blog.google/products/search/search-language-understanding…
What can make this go out of date
- Google's spam policies are revised without notice (this edition last updated 2026-08-28); if a future revision ever quantified keyword stuffing, the page's central claim that no percentage exists would need restating.
Pick up where you left off
Stored only in this browser — never sent to our servers.