Keyword Extractor — Extract Keywords from Any Text or URL
Find the terms and phrases any text is really about. Paste an article, upload a file, or fetch a page by URL and get a ranked keyword list built with the RAKE statistical algorithm — every keyword with its relevance score, exact count and density, plurals and possessives merged. Check a target keyword's placement, click any keyword to see it highlighted in context, compare two texts for a content-gap list, and export to CSV. No AI claims, no upload, no daily limit — extraction runs entirely in your browser, in English, বাংলা and more.
Extraction runs 100% in your browser with the RAKE algorithm — your text is never uploaded, and there is no daily limit.
How to Use
- 1Paste your text, upload a .txt, .md or .html file, load a sample, or switch to From URL and fetch any public page — yours or a competitor's.
- 2Set the stopword language (Auto-detect handles English and বাংলা), the maximum phrase length, minimum occurrences and any words to exclude.
- 3Click Extract keywords. Read the ranked table, filter by phrase length, and click any keyword to see every occurrence highlighted in your text.
- 4Type a target keyword to check its count, density and placement, then copy the list, copy hashtags, or download CSV/TXT. For competitor work, use Compare to get the gap list.
Key Features
- ✓
Ranked keywords, not a bag of words
Most free extractors strip out common words and hand you the rest in a pile. This one ranks every term and phrase with the RAKE algorithm — the classic statistical method that scores a word by how strongly it co-occurs with other content words — blended with how often it actually appears. Each keyword shows its relevance score, its exact count and its density, so you can see why it ranked and decide for yourself whether it matters.
- ✓
Phrases up to four words, filterable
Nobody searches for a single word like “machine” — they search for “espresso machine cleaning”. The extractor builds one-, two-, three- and four-word keyphrases and lets you filter the table by phrase length, search inside the results, and re-sort by relevance, count, density or alphabet. Long-tail phrases stop hiding inside the unigram noise.
- ✓
Click a keyword, see it in context
Click any row and every occurrence of that keyword lights up inside your source text — including its plural and possessive forms, because matching is done on normalised words. You see at a glance whether a term is spread naturally through the piece or crammed into one paragraph, which is the difference between topical depth and stuffing.
- ✓
Target keyword check built in
Type the phrase you actually want to rank for and get its verdict: occurrences, density with a plain-English band, where it first appears, and whether it made the first and last hundred words. Fetch the text from a URL and the same check extends to the page itself — is the target in the title tag, the H1, the H2s? It is the focus-keyword report writers usually need a WordPress plugin for, on any text at all.
- ✓
Compare two texts, get the gap list
Paste your draft in one box and a competitor's page in the other. The compare view splits the keywords three ways — shared ground, only yours, and the gap: terms they cover that you don't. That gap list is the working core of content-gap analysis, produced in one click and downloadable as CSV, instead of the usual export-two-files-and-diff-them-in-a-spreadsheet ritual.
- ✓
Bengali and multilingual extraction that actually works
Tokenisation here is Unicode-aware down to combining marks, so Bengali words stay whole instead of shattering at every vowel sign — and Bengali gets its own stopword list, with the danda (।) treated as the sentence end it is. English, Spanish, French and German stopword sets are built in too, plurals and possessives are folded together for Latin scripts, and everything runs in your browser: no upload, no account, no ten-extractions-a-day cap.
Frequently Asked Questions
Is this keyword extractor really free?
How does it extract keywords — is it AI?
Can I extract keywords from a URL or a competitor's website?
Which languages does it work in?
What do count, density and relevance score mean?
How is this different from the keyword density checker?
Can I compare my page against a competitor's?
Is my text uploaded or stored?
What keyword extraction is — and what it is not
Keyword extraction answers a narrow, useful question: what is this text actually about, in the text's own words? You give it an existing piece of writing — a draft, a competitor's article, a product description, a job post — and it returns the terms and phrases that carry the topic, ranked by importance. That is different from keyword research, which starts from a seed word and pulls candidates from a search engine's database along with volumes and difficulty scores. Extraction needs no database because it reads the document in front of it. It is also different from keyword density checking: a density tool measures one phrase you already chose, while an extractor discovers the candidates you might choose. The two sit side by side — extract first to find your terms, then use the keyword density checker to audit the winner in depth.
How the ranking works: RAKE, honestly explained
This tool uses RAKE (Rapid Automatic Keyword Extraction), a statistical algorithm published by Rose, Engel, Cramer and Cowley in 2010 and implemented in libraries across every major language. It works in three steps. First, the text is split into candidate phrases wherever a stopword or sentence boundary interrupts the run of content words — so “keyword research is the foundation of search” yields the candidates keyword research and foundation and search. Second, every word earns a score from its social life inside those phrases: degree counts how often it appears alongside other words (words in longer phrases have more company), frequency counts how often it appears at all, and the word's score is degree divided by frequency. A word that always travels in meaningful phrases outranks one that merely shows up a lot. Third, a phrase scores the sum of its words' scores. Two adjustments keep the ranking honest for web content: the raw sum is averaged per word (otherwise any long one-off phrase would crown itself), and the result is boosted by the square root of the phrase's occurrence count, because a topic phrase used three times is stronger evidence than one used once. Before any of that, light normalisation merges “detector”, “detectors” and “detector's” into one keyword. There is no neural network and no language model — and no need for one. This is arithmetic on your text: instant, identical on every run, and explainable down to the formula.
Reading the results like an editor
Start with the two- and three-word phrases; that is where search intent lives. A healthy article usually shows one clear primary phrase near the top, a family of supporting phrases beneath it, and a long tail of one-offs. Then check the numbers with judgement. Count tells you the text's commitment to a term. Density — the share of the whole text the phrase accounts for — tells you its weight: for multi-word phrases, somewhere around half a percent to two percent reads naturally in most articles, while a single word pushing past three percent deserves a second look (read the passages, don't obey a number — a 110-word how-to will legitimately repeat its subject often). Relevance is the ranking signal, useful precisely because it disagrees with raw count: a phrase can be frequent but generic, or rare but structurally central, and the score surfaces both cases. Clicking a row to see its occurrences in context is the fastest sanity check of all — keywords that appear in introductions, headings and conclusions are doing more work than the same count buried in one section. If the top of the list surprises you, the text may genuinely be about something other than what you intended — the single most valuable thing an extractor can tell a writer.
The workflows this replaces
Competitor analysis. Fetch the page that outranks you, read its top phrases, then run the compare view against your own draft: the gap list is your editing brief. Content briefs. Extract from the top three ranking pages on a topic and merge the lists — the shared phrases are the vocabulary search engines expect on the subject, and your brief writes itself. Tagging and metadata. The same ranked list, copied as comma-separated tags or hashtags, seeds YouTube tags, marketplace keywords and blog taxonomies far better than brainstorming, because it comes from the content itself. Audits. Run old articles through the extractor next to the keyword you meant to target; drift between the two is content decay made visible. Non-SEO reading. The top keywords of a 5,000-word report, a contract or a research paper are a fair two-minute summary of its subject — students and analysts use extraction as a reading tool, not a ranking tool. In every case the extracted list feeds naturally into the word counter for length and readability work.
Bengali text, and why tokenisation is the hard part
Most online extractors are quietly English-only in a way that breaks other scripts. Bengali words are built from letters plus combining vowel signs — the া in ঢাকা is a separate Unicode mark, not a letter — so a naive “letters only” splitter shatters every Bengali word into fragments and the “keywords” that come out are meaningless shards. This extractor's tokeniser includes combining marks, treats the danda (।) as a sentence terminator, ships a Bengali stopword list (এবং, থেকে, জন্য, করে and their kin are filtered exactly as the and of are in English), and offers a combined English + বাংলা mode for the mixed-script documents Bangladesh produces every day. Spanish, French and German stopword sets cover the other common cases, and the Auto-detect setting picks the Bengali-aware mode on its own when the text calls for it. The plural-folding normalisation applies to Latin scripts only — Bengali words are never “corrected” by English rules.
The honest limits
Three, in plain terms. No search volume. Extraction measures your text; it cannot tell you how many people search a phrase, and this page will not invent that number — validate finalists in Google Keyword Planner or a research suite. No semantics. RAKE knows “espresso machine” and “coffee grinder” are central because they co-occur and repeat; it does not know they are related concepts, and it will treat synonyms as separate keywords where a human editor would group them. Text in, text out. The URL mode reads the HTML the server sends; pages that render everything with JavaScript may return little text, and the tool says so rather than analysing an empty shell. What you get in return for those limits is the thing the AI-writer funnels don't offer: your text never leaves your device, there is no account, no credit balance, no ten-per-day allowance — and every number on the results table can be traced back to the formula above.
Related Tools You Might Like
Handy utilities in Text & AI Writing Tools
Keyword Density Checker
Free keyword density checker for text, files & URLs. Analyze 1–4 word phrases, catch keyword stuffing, check placement and export a CSV report. No sign-up.
Word Counter
Count words, characters, sentences and reading time live as you type — with goals, platform limits and keyword density. 100% private.
SERP Preview Tool
Preview how your title tags and meta descriptions appear in Google desktop and mobile search results. Avoid truncation with our free SERP simulator.