Word counter with characters, reading time and term density

Who this page is for: anyone who writes and needs to know the size of what they wrote, whether to fit an assignment limit or to pace an article, a script or a newsletter. If the text is headed for an ad and the question is whether it fits the platform field, the right page is the ad character counter, which checks each asset against the Google and Meta limits. If the question is whether a paragraph is shaped like an answer an AI can quote, use the citability checker. Here the subject is the text itself: its size, its rhythm and the words it repeats.

The number looks simple and is not

Counting words sounds like the most trivial job a computer could do, right up until you open three counters and get three answers for the same text. The gap is not a bug: it is the sum of decisions nobody documents. One counter treats "well-known" as one word, another as two. One treats "1,500.50" as a number, another as three fragments. One counts an emoji as one character, another as two, because it counted internal encoding units instead of visible characters. None of them is wrong by accident, but only one of them matches the number your editor, teacher or client will use to judge the work.

So this page states its rules instead of hiding them. The method block below says exactly what counts as a word, a character and a sentence, and the worked example reproduces, number by number, what the tool shows with the sample text already loaded. If your count has to line up with a specific system, you will know where and why the difference shows up rather than finding out at delivery.

The second thing this page does that most do not is look past the total. A 900 word text can be comfortable or unbearable depending on how those words are spread: 40 word sentences exhaust the reader, twelve line paragraphs push people away on a phone, and a term repeated 20 times makes the reading heavy long before any search engine cares. The averages and the density table exist to show that distribution, which is where editing actually happens.

How to use it

1. Paste or type the text in the box. Everything recalculates on each keystroke, with nothing to click.
2. Read the four big numbers first: words, characters, sentences and reading time. They answer the most common question in a second.
3. Adjust the reading speed if your audience is not the average reader. Technical text calls for something near 150 words per minute, light text takes 250.
4. Use the text from the inside panel to read the rhythm: average words per sentence, longest sentence and how many sentences passed the threshold you set.
5. Read the most frequent terms table asking one question only: is this what the text should be about? If the answer is no, the problem is the text, not the percentage.

How it works: the counting rules

None of the math below is hard. What matters is that it is explicit, because the definition, not the arithmetic, is what separates one counter from another:

Two of those lines deserve a note. The sentence rule undoes two false breaks: the decimal point inside a number and the period in common abbreviations such as "Mr." and "etc.". Without that care, a text full of figures shows up with twice the sentences and an absurdly low words per sentence average. The density rule divides by total words, articles and prepositions included, which is the convention nearly every SEO tool uses. Dividing only by content words would inflate every number and make it incomparable with the rest of the market.

Worked example (it reproduces the default output)

The sample text already loaded has two paragraphs and four sentences, and describes a test hypothesis for a pricing page. Checking what the tool shows:

• Basics: 77 words, 426 characters with spaces and 349 without, across 4 sentences and 2 paragraphs.
• Reading time: 77 divided by 200 is 0.385 minute, which the tool rounds and shows as 23 sec. Out loud, at 130 words per minute, it is 0.592 minute, or 36 sec.
• Rhythm: 77 words across 4 sentences give 19.3 words per sentence, and the longest sentence runs 27 words. With the threshold at 25, exactly 1 sentence is flagged as long, and it is the one worth splitting.
• Vocabulary: 54 unique words out of 77, or 70.1% distinct vocabulary. The closer to 100%, the less the text repeats itself; long texts naturally fall, because connectors keep coming back.
• Density: the top term is page, with 3 occurrences and 3.9% density. Right below it come plan, pricing and side, with 2 occurrences and 2.6% each.
• Phrases: the only one appearing twice is pricing page, also at 2.6%. That is exactly the subject of the text, which is the result you want in that table.

Now run the test that shows why the phrase table exists. Replace the second "pricing page" with "that page" and the phrase disappears from the panel, even though the text still talks about the same thing. No tool knows what your text is about: it only knows what your text repeats. When the central term shows up once, the skimming reader, the search engine and the assistant asked for a summary may all simply miss what the subject was.

Reading each number

Words and characters answer to external limits. Essays are usually capped in words, translation is quoted in characters or in standard pages, form fields and meta descriptions are measured in characters, and in that case the number that counts is the one with spaces. Check which currency your limit is written in before you start cutting.

Average words per sentence is the best single number for reading effort. Between 15 and 20 the text flows for almost any audience. Past 25 the average sentence starts demanding a second pass, and the effect is worse on a phone, where the line is short and one sentence fills the screen. The long sentence count points at the exact offenders: usually two or three giants drag the whole average, and splitting them fixes the problem without rewriting anything.

Density is a diagnosis, never a target. The right question is not "did I hit the percentage?" but "are the terms at the top actually the subject?". When the term you meant to work on is not even in the top eight, the text is talking about something else, and no percentage tweak repairs that. In the other direction, a term above 5% in a long piece almost always means lazy repetition, the kind that tires a human reader long before it bothers an algorithm.

Reading time is there to calibrate expectations and nothing else. It assumes a straight read from start to finish, which almost nobody does online: real behaviour is scanning, with stops at headings, lists and numbers. Use it to decide whether an article fits your audience's patience, or whether a script fits the video slot, not to promise anything to anyone.

Known limits

The sentence count is the roughest of them all. Abbreviations outside the short list the tool knows will still split one sentence into two, and so will an ellipsis in the middle of a quotation. On ordinary prose the drift is small; on legal text, full of abbreviations, it grows.

The term ranking does no stemming: "plan" and "plans" count as different terms, and so do "test" and "tested". That is deliberate, because grouping forms would need a dictionary per language and would bring errors of its own, but it does mean a subject written in several forms shows up diluted in the table. When reading the ranking, add the variants of the same root in your head.

Languages written without spaces between words, such as Japanese and Chinese, are not segmented correctly by any space based rule, including this one. The character count still holds there, the word count does not.

From counted text to tested text

Counting is the start. What moves results is finding out which version of the text works better, and that is not settled by eye:

• Before deciding whether the gap between two versions is real, check the traffic it would take in the sample size calculator.
• With the test running, get the verdict from the significance calculator instead of calling it on the first good day.
• If the text is a whole landing page, run the page through the landing page grader before writing the next version.
• If it needs to be found and quoted by AI assistants, start with the GEO readiness checker.

Frequently asked questions

How do I count the words in a text?

Paste the text in the box above and the count updates as you type. The rule this tool uses is the common one: a word is any run of letters or digits separated by a space or by punctuation, with apostrophes and hyphens counting inside the word. That way "well-known" and "1,500.50" each count as one word instead of two or three, which is where most counters drift by a few percent.

Why do two counters give different numbers for the same text?

Because each one settles the edge cases its own way and nobody publishes the rule. A hyphenated word can count as one or two. A decimal number can count as one or three. A character can be counted in code points or in UTF-16 units, and in the second case an emoji counts as 2 instead of 1. A sentence may or may not break at an abbreviation like "etc.". On this page all four rules are stated in the method block, and the gap against other counters usually stays under 1% on ordinary prose.

How many words fit on a page?

It depends on the format, but the working references help: a double spaced page in 12 point type holds around 250 words, a single spaced page around 500, and a blog post that means to cover a subject usually lands between 1,200 and 2,000. A one minute video script runs close to 140 words when read out loud. Check which unit your limit is written in, words or characters, before celebrating or panicking.

Is the reading time reliable?

It is an honest estimate, not a stopwatch. The math divides total words by the reading speed you choose, and the 200 words per minute default matches silent adult reading of ordinary prose. Technical text loaded with numbers and rare terms drops closer to 150. Reading out loud runs between 120 and 150, which is why the tool reports both times separately. If the text is a script you will record, look at the out loud time and ignore the other one.

What is the ideal keyword density?

There is no magic number, and chasing one is the fastest way to write badly. The density here is a diagnosis, not a target: it shows what the text is actually about. If the term you meant to work on is missing from the most frequent terms, the problem is not the percentage, it is the text. And when any single term passes 5% in a long piece, that is usually lazy repetition, the kind that tires a human reader long before it bothers any search engine.

Does the tool send my text to a server?

No. The whole analysis runs in JavaScript inside your browser with no network call at all: the text never leaves your machine, is never stored and is never logged. That matters when the material is confidential, unpublished or a client deliverable. You can confirm it by opening the network tab and typing: not a single request fires while the numbers change.

What counts as a sentence?

Any stretch closed by a period, exclamation mark, question mark or ellipsis, and also any line break, because a heading or a list item with no final period is a whole sentence to the person reading. Two false breaks are undone on purpose: the decimal point inside a number and the period in common abbreviations such as "Mr." and "etc.". An abbreviation outside that short list will still split one sentence into two, which is why the sentence count is the roughest number on this page.

Does it work in any language?

Words, characters, sentences and paragraphs work in any language written with spaces between words, because the rule is based on Unicode letters rather than a fixed list. What changes per language is the term ranking, which drops the stop words of the page language so the table does not hand you "the", "and" and "of" at the top. This page uses the English list; the Portuguese and Spanish versions use theirs.

Keep reading

To understand how the text you just counted behaves in search and inside an AI answer, continue with CRO vs SEO, read the generative engine optimization guide and learn to build quotable blocks in FAQ content for AI citation.