See what a page actually spends its words on: phrase density at four lengths, variants counted as one term, and where each phrase sits.
Nothing you paste is stored
Word count not what you expected? Here is what we count, and what we leave out.
A frequency table on its own tells you which words are common, which is rarely the question. These are the things that turn it into an answer about one page.
Four tables, built in one pass and switched without another fetch. Single words tell you the subject; the longer phrases tell you what you actually claim to be about.
Search engine and search engines count as one term, not two half-strength ones. Every folded form is listed underneath with its own count, so you can see what was merged.
A tick or a dash for the title, meta description, H1, subheadings, image alt text and anchor text. Frequency without placement tells you half the story.
Switch to body only and every anchor is removed from the count. A term that looks well used until its link text is taken away was never really in your writing.
One sentence: the density, where the term ranks among phrases of its own length, and which term outweighs it if one does.
Words counted, words on the page, the split between body and link text, and how much of it is stop words. Every percentage on the page divides by a number you can see.
Paste any public page, or paste text you have not published yet. Drafts take up to 100,000 characters and need at least 50 words to be worth measuring.
The CSV carries all eight tables, both scopes at all four lengths, not just the tab on screen. A copy button puts the current view on the clipboard for a spreadsheet.
Four steps, none of which need an account. The first two take about ten seconds and the last two are where the answer usually is.
Any public page works, yours or a competitor’s. A draft that has not shipped works just as well, and nothing you paste is stored.
Optional, and it is what turns a word-frequency table into an answer. Without it you get the tables; with it you get a verdict on one term.
A phrase used twelve times that appears in no heading, no alt text and no title is decoration. One used four times across the title, H1 and a subheading is a subject.
Switch the scope to body only. If your term collapses, its presence was coming from navigation and anchors rather than from the writing.
Density is occurrences divided by words counted, times 100. A term used 12 times on a page of 800 counted words sits at 1.5 per cent. The arithmetic is not the interesting part. The denominator is, because it is the number tools disagree about, and a percentage is meaningless until you know what it divided by.
The words it counted in the scope you are looking at, and it prints that number above the table. Switch from all content to body only and the denominator changes with it, because the anchor text has left the page along with the words it contributed. Every percentage in the table can be reproduced by hand from two numbers on screen.
Twelve uses buried in body copy and twelve uses spread across the title, the H1 and two subheadings are the same number and a different page. The six placement columns are there so the count never has to be read on its own. To see the same signals scored as part of a whole page, run the URL through the on-page SEO checker.
Phrases that start or end on a stop word, bare numbers, dates, single letters and anything appearing only once. The occurrence floor is applied after variants are grouped, so a term that only reaches two by combining its forms is kept rather than dropped.
The tool bands every term against the page’s own length rather than against a universal figure, because no universal figure exists. These are its bands, and they are heuristics for spotting outliers rather than targets to write towards.
Under 0.5%
Present, but barely. On a page written for that term this usually means the writing never committed to it, or that it survives only in the navigation.
0.5% to 3%
A wide band on purpose. Most pages that read naturally land somewhere in it, and moving around inside it is not worth your afternoon.
3% to 5%
Worth a look rather than an alarm. Glossaries, specification pages and anything with a repeated product name land here legitimately.
5% and above
Repetition a reader will notice. Search engines have discounted keyword stuffing for two decades, so the cost now falls on the person reading rather than on the ranking.
Nothing is banded on a page under 300 words. Below that a single extra use swings the percentage far enough that the band would describe the page’s length rather than its writing.
Deleting every common word before counting would change your denominator and inflate every percentage on the page. The 142-word stop word list is used to reject phrases that begin or end on one, so of the and in the never reach the table while black Friday and sunday roast do.
A standard English list misses words that dominate real pages. May ranked fifth on a Wikipedia article purely from the phrase Retrieved May, and no and not sat in the top seven on an MDN reference page. Both are in the list here for that reason.
May is in the list, for the reason above. Adding the rest would be tidier and would quietly delete real keywords, so dates are caught by a separate rule that needs a date word and a numeral together. That removes 22 Feb 2022 while leaving Cyber Monday and Sunday roast alone.
The word budget reports what proportion of the counted words were stop words. It is the quickest way to tell a page that is genuinely thin from one whose word count is mostly connective tissue.
We count the words a reader actually sees in the main body, not every word in the raw HTML. Navigation, footers and sidebars are excluded, script and style content never counts, and citation markers on a page thick with them are stripped before counting so a reference list does not pad the total.
Both of those steps are guarded rather than trusted. Main-content extraction is only used if it keeps at least half the page’s words, and the citation strip is thrown away entirely if it would remove more than half. A page whose whole body sits inside a container named for references keeps its words, which is the failure mode the guard exists for.
So a tool that counts the full raw page, or one that strips citations without a guard, will report a different total. Neither number is wrong. They answer different questions, and density is only comparable when both sides divide by the same thing.
The tool takes any public URL, which makes it as useful pointed at a competitor as at your own draft. Three things are worth looking for.
Without a keyword the verdict names the terms the page leans on. It is often not the term the page is trying to rank for, which tells you something about why it does or does not.
Single words are noisy across any two pages in the same field. The three and four-word tables are where the difference between two pages on the same subject actually shows up.
The placement columns read as well on a page you did not write. A competitor ranking with a lower count and better placement is a more useful finding than one with a higher percentage.
Density describes one page. To see how a title and description will read in the result itself, use the SERP preview tool, or write a fresh set with the meta tag generator.
Density tells you what a page emphasises. It cannot tell you whether the emphasis worked. RankCaddy tracks Google organic and local rankings daily, blends your Analytics and Search Console data into one view, and flags the pages quietly competing with each other for the same term.
Start your 14-day free trialFrom $24 a month for 250 daily keywords, with unlimited projects and users on every plan. No card needed for the trial.
Keyword density is the share of a page’s words made up of a given word or phrase, written as a percentage. It is a rough measure of what a page spends its words on, which makes it useful for spotting a term that is barely present or one that has taken over. It is not a ranking factor in its own right.
Count how often the term appears, divide by the number of words counted, and multiply by 100. The denominator is the part that varies between tools. This one divides by the words it counted in the scope you are looking at, and shows you that number on the page, so you can reproduce any percentage in the table by hand.
No. Google has never published a target percentage and chasing one is a reliable way to make writing sound odd. This tool bands a term as thin, healthy, high or stuffed against the page’s own length rather than against an invented universal figure, and it does not band anything at all on a page under 300 words, where a percentage is mostly noise.
The honest answer is that the range is wide and the edges matter more than the middle. This tool treats under 0.5 per cent as thin, 0.5 to 3 per cent as healthy, 3 to 5 per cent as high and 5 per cent or more as stuffed. Those are the tool’s own bands, chosen to flag outliers rather than to name a target, and Google endorses none of them.
Not as a number to hit. It still matters as a diagnostic: a page that never uses the term it is written for has a real problem, and so does one that uses it in every other sentence. Density is how you notice either without reading the page five times.
Search engines treat close variants, plurals and verb forms as the same underlying term. Counting rank and ranking separately splits your real usage across two rows and makes both look weaker than they are. The grouped row shows every form it folded in, with counts, so nothing is hidden.
Repeating a term far past the point where it reads naturally, in the hope of ranking for it. It is one of the oldest spam tactics and search engines have been discounting it for two decades. The practical damage now is to the reader rather than to the algorithm: stuffed copy is worse to read, so it converts worse.
Yes. Paste the draft into the text tab. Placement columns and the link-text split need real HTML, so those are not shown for pasted text, but the phrase tables, the bands and the verdict all work.
Yes, with no signup and no credit card. There is no usage cap to unlock, no report to email yourself, and the export is the full run rather than a sample.