Text Analyzer

The full workup, live on every keystroke: counts and structure, Flesch readability with a plain-English band, your top keywords by density, and how long the piece takes to read or deliver. Free, no sign-up — analysis runs entirely in your browser.

0 Words
0 Sentences
— Words/sentence
0 Unique words
0 sec Reading time
0 sec Speaking time

Type at least one full sentence to score readability.

Keywords appear once you type.

What the analyzer measures

One engine runs over your text on every keystroke, and its output lands in three places. The six cards hold the counts. Words and sentences come straight from the tokenizer described in the next section. Words/sentence is the first divided by the second, shown to one decimal, or a dash until there is a sentence to divide by. Unique words is the number of distinct spellings once case is ignored, so "The" and "the" are one entry while "run" and "runs" are two. Reading time and speaking time are derived from the word count alone, using the two averages explained below. The readability panel holds the Flesch Reading Ease score, its plain-English band — a conventional label for a stretch of the scale rather than a measurement of any reader — the Flesch–Kincaid grade, the number of syllables the engine counted, and the words-per-sentence figure again, because that average is one of the two inputs to both formulas. The keyword table lists the ten most frequent words once common function words are removed, each with its count and its density. All of it comes from the same word list, so the panels can never disagree with each other about how many words you wrote; they can only disagree with another tool, for the reasons set out below and in full on the methodology page.

How words, sentences, and paragraphs are found

The tokenizer has one rule for words: a word is any unbroken stretch of letters or digits, in any script, and a hyphen or apostrophe inside that stretch keeps it whole. Most of what the cards show follows from that. "State-of-the-art" is a single word and so is "it's"; an accented or non-Latin letter is still a letter; and a digit is a word character, which makes "2026" a word. Counters diverge at the rule's edges. A comma is not a word character, so "2,500" breaks into two words at the comma. A pasted address like https://example.com breaks at its punctuation into three (https, example, com). The two letters of "e.g." each become a one-letter word. Emoji never become words; they are tallied among the characters, one per code point. Sentences come from a second pass over the raw text, which is cut wherever a run of ending punctuation (. ! ? …) is followed by whitespace or by the end of the text; each piece counts as a sentence only if it holds at least one letter or digit. That cut is blind to abbreviations, so "Mr. Smith arrived." comes out as two sentences, while "3.14" stays whole because its period is followed by a digit, not whitespace. A heading or list item with no closing punctuation is not cut off from what comes next: its words are folded into the sentence below it. A paragraph is a block of text set off by blank lines, so a single line break never opens a new one. The guide What counts as a word takes each of these conventions in turn, with examples, and explains why two counters that differ by a few words are both behaving normally.

Worked example 1: a short paragraph

Paste this paragraph — "Strong coffee starts with fresh beans. Grind them just before brewing, because ground coffee loses aroma within minutes. A burr grinder gives a far more even grind than a blade grinder, and evenness is what makes the cup taste clean." — and the cards read 40 words, 3 sentences, 13.3 words per sentence, and 35 unique words. Reading time shows 10 sec (40 ÷ 238 words per minute, rounded to the nearest second) and speaking time 16 sec (40 ÷ 150). The readability panel reports a Reading Ease of 74.9 — fairly easy — about 7th-grade level — and a Flesch–Kincaid grade of 6.1, computed from 56 counted syllables. The keyword table opens with "coffee" (2, 5%), "grind" (2, 5%), "grinder" (2, 5%); every other ranked word appears once. Notice that "grind" and "grinder" rank separately. The engine does not stem, so word forms are never merged, and "Grind" at the start of a sentence joins "grind" later in the text only because ranking is case-insensitive. This is the profile of a short paragraph that stays on one subject.

Worked example 2: where the edges show

Now paste a sentence built to hit the tokenization edges: "Our new grinder lists at 2,500 dollars, i.e. roughly twice last year's price, and the spec sheet (see https://example.com) calls it state-of-the-art." The cards say 26 words and 2 sentences. The word count is higher than a reader's because "2,500" splits into two words at the comma, "i.e." contributes two one-letter words, and the URL contributes three. The sentence count is 2 rather than the single sentence a reader sees, because the period after "i.e." is followed by a space and the splitter treats it as an ending. Both quirks feed the derived metrics. The words-per-sentence average is 13 instead of the 26 it would be with one sentence, and that average goes straight into both Flesch formulas: the panel reports a Reading Ease of 92.8 — very easy — about 5th-grade level — a label that says more about the short words and the false split than about how easy a pricing sentence is to follow. The keyword table is close to useless here, because no word repeats: every entry has a count of 1 and the same density, 3.85%, so the ranking falls back to alphabetical order and the first row is "500" — the tail of "2,500", which survives the single-character filter that removed "2", "i", and "e". A density table only tells you something when there is repetition to rank.

Worked example 3: dense versus plain

Readability is where the numbers move furthest. Score this sentence: "The aforementioned organizational restructuring necessitates comprehensive stakeholder consultation before implementation." It is 10 words and 1 sentence, but the heuristic counts 41 syllables — 4.1 per word — and the panel returns a Reading Ease of −150.2 with a grade of 36.7: very difficult — graduate level. Convention reads the scale as running from 0 to 100, but nothing in the arithmetic enforces either end, and a sentence this dense goes straight through the floor; the analyzer reports the raw value rather than clamping it. Now score a plain rewrite of the same instruction: "We must ask the staff before we change how the company is run." That is 13 words and 16 syllables — 1.2 per word — for a Reading Ease of 89.5 (easy — about 6th-grade level) and a grade of 4. The rewrite is the longer sentence of the two, so sentence length is not what moved the score; the syllables-per-word term did, by about 240 points. That is the general lesson of the Flesch coefficients: adding one word to the average sentence takes roughly one point off Reading Ease, whereas adding 0.01 to the syllables-per-word average takes off roughly 0.85, so word length is the lever that matters. The rewrite also shows the unique-word rule at work: 13 words but 11 unique, because "we" and "the" each appear twice.

Worked example 4: two ordinary paragraphs

A longer sample shows what the full workup looks like in everyday use. Paste these two paragraphs, blank line included:

A community garden needs three things before the first seed goes in: water, sunlight, and a plan for who does the weeding. Water is the easiest to check, because a tap either exists or it does not. Sunlight takes a week of watching, since a plot shaded by a wall at noon may look bright at nine in the morning.

The weeding plan is the hard part. A garden with twenty members and no rota grows weeds faster than vegetables, and the members who do turn up soon resent the ones who do not. Write the rota down, post it by the gate, and review it every month. The garden will look after itself once the people do.

The cards read 118 words in 7 sentences (16.9 words per sentence), 77 of them unique; the engine also sees 2 paragraphs because of the blank line. Reading time is 30 sec and speaking time 47 sec. The readability panel gives a Reading Ease of 77.9 — fairly easy — about 7th-grade level — and a grade of 6.6 from 156 syllables. The keyword table is now informative. "garden" leads with 3 occurrences, 2.54% of all words, followed by "look" (2, 1.69%), "members" (2, 1.69%), "plan" (2, 1.69%), "rota" (2, 1.69%), "sunlight" (2, 1.69%), "water" (2, 1.69%), "weeding" (2, 1.69%). Two things are worth noticing. "look" makes the table on the strength of "look bright" and "look after": the engine counts spellings, not meanings, so a word with two senses is one keyword. And the densities are small because the base is all 118 words, function words included, which is why the word that leads the table can still hold only a small share of the text.

Reading and speaking time: what the two averages mean

Both timing cards are a division: the word count over a fixed rate. The reading rate is 238 words per minute, the adult average for silent reading found by Brysbaert's 2019 meta-analysis of reading-rate studies. The speaking rate is 150 words per minute, a pace typical of prepared speech. Each figure is an average across many readers or speakers and many kinds of text, so the card is a planning estimate, not a target and not a stopwatch: someone skimming will beat the reading figure, a technical passage will run past it, and a script full of figures will run past the speaking figure, because a number the tokenizer counts as two words — "2,500" — takes four words to say aloud. The arithmetic is linear, so a draft of 1,000 words reads in about 4 min 12 sec and takes about 6 min 40 sec to deliver aloud; at these two averages, speaking takes roughly 1.6 times as long as silent reading, whatever the length. The display rounds to the nearest second, reports anything shorter than five seconds as "under 5 sec", and drops the seconds when they come to zero, which is why a text of 476 words shows "2 min" rather than a figure ending in seconds. How to plan with the reading estimate, and where it stretches, is the subject of Reading time and words per minute; the delivery side, and how far a speaker drifts from the average, is covered in Speaking time and speech length.

Readability in one paragraph

The two scores share their inputs and differ only in their weights. Reading Ease starts at 206.835 and subtracts two weighted averages: 1.015 times the mean sentence length (words ÷ sentences) and 84.6 times the mean word length in syllables (syllables ÷ words). The grade figure takes the same two averages with positive weights — 0.39 times words per sentence, plus 11.8 times syllables per word, minus 15.59 — so it climbs where Reading Ease falls, and because the two weightings differ it is a second formula rather than a relabelling of the first. The weights are the published ones: the Reading Ease coefficients are Rudolf Flesch's, from 1948, and the grade coefficients come from the recalibration produced for the US Navy in 1975; the engine applies both exactly as published. Nothing else enters either formula — vocabulary difficulty, sentence structure, and meaning are all invisible to it. The word and sentence figures are the tokenizer's, so every edge described above reaches the score. Syllables are not looked up in a dictionary; they are estimated by a rule. The rule counts the groups of consecutive vowels in a word (a, e, i, o, u, and y), takes one group off when the word ends in a silent "e" — unless the ending is "-le", which keeps its syllable — and reports at least one syllable for any word that has letters. A token with no letters, such as a number, is given no syllables at all, yet it still sits in the word total that the ratio divides by. Checked against a dictionary, the rule errs in both directions: it returns 2 for "created", one short of the dictionary's 3, and 3 for "business", one more than the dictionary's 2, while "beautiful" comes back at 3, which is right. A score is an average over every word in the text, so such miscounts partly offset one another and the total usually lands within a few points of a hand count; read the score as a strong indication, not a laboratory measurement. Because the rule discards every character outside a–z before counting, the readability panel is built for English text. The band label beside the score is a fixed lookup on the Reading Ease value — the cut points are 90, 80, 70, 60, 50, and 30 — and nothing more: it does not consider the grade figure, the audience, or the subject. The full derivation, with more examples of what moves a score, is in Flesch Reading Ease, explained; the grade formula is covered separately in Flesch–Kincaid grade level, and the heuristic is taken apart word by word in How syllables are counted.

Keywords and vocabulary

The unique-word card and the keyword table look at the same word list from two angles. Unique words counts distinct spellings after lowercasing: capitalisation is ignored, but inflection is not, so "garden" and "gardens" are two entries, as are "it's" and "its". The pangram "The quick brown fox jumps over the lazy dog." is the smallest demonstration: 9 words, 8 unique, because "The" and "the" collapse into one. Set beside the word count, the unique-word figure shows how much of a draft is repetition. The keyword table ranks words by count, again ignoring case, after dropping two kinds of token: 74 common English function words held on a fixed list (the, and, of, that…) and anything one character long. A word's density is its count as a percentage of every word in the text — the dropped function words stay in that base — which is why "coffee", with 2 occurrences in the 40-word paragraph above, shows 5%; a tool that removes function words before dividing would report a larger share for the same text. Word forms are never merged: "run", "runs", and "running" are three separate rows, on purpose, because folding them together would conceal the very repetition the table is there to expose. Ties are broken alphabetically, which is why a text with no repeated words produces an alphabetical list rather than a ranking. The table describes the draft; it does not recommend a density, and no figure in it is a target. The conventions are set out in full in Keyword density, explained.

How to read the output, and how not to

The band label is not a verdict on quality. It is a lookup on two averages. The pricing sentence above scored "very easy — about 5th-grade level" because its words are short and a false split halved its sentence length; the dense sentence scored "very difficult — graduate level" because its words are long. Neither label knows what the sentence means, and a text made of short words can be conceptually hard while scoring as easy. Short texts swing. With one sentence, a single long word shifts the syllables-per-word average by a large fraction, and a single abbreviation halves the sentence length. Score running prose in paragraphs, not headlines, and expect the figures to settle as the text grows. Structured text skews sentence length. Headings, bullets, and captions without ending punctuation merge into neighbouring sentences, raising the words-per-sentence average and lowering the ease score; score the prose, not the outline. Timing is an estimate, not a stopwatch. Both constants are population averages, and the reader or speaker in front of you is not the population. Density describes; it does not prescribe. The table shows where repetition sits in the draft and nothing about where it should sit. Counts are conventions. If a figure has to be exact because a limit is enforced — a strict submission ceiling, say — the count that decides it is the one the recipient runs, and the job of this one is to tell you whether you are near enough to check.

Guides

The guides go deeper than a tool page can, each answering the questions the analyzer's cards raise:

Every guide is listed on the guides index.

Focused tools

Each analysis also has a dedicated page: the Readability Checker for score-by-score interpretation, the Keyword Density Checker for the ranked table and its conventions, the Word Frequency Counter for hunting overused words, and the Speaking Time Calculator for talks, videos, and ad reads.

Frequently asked questions

Why does my word count differ from my word processor or another counter?

Counters agree on ordinary prose and disagree at the margins: hyphenated compounds, contractions, numbers with thousands separators, URLs, and stray punctuation. Here a word is any unbroken run of letters or digits, and a hyphen or apostrophe between them keeps the run together, so "state-of-the-art" is one word and "2,500" is two. Differences of a few words on a long text are normal. The full rule and its edges are in the guide on what counts as a word.

Why is the sentence count higher than I expected?

A sentence ends where a run of . ! ? or … is followed by whitespace or by the end of the text, and every piece that results counts as a sentence if it has a letter or digit in it. An abbreviation followed by a space therefore adds a sentence — "Mr. Smith arrived." registers as two — and a heading without ending punctuation merges into the sentence that follows it. A decimal such as "3.14" stays whole, since its period is followed by a digit rather than a space.

What do the reading and speaking times assume?

The reading card divides your word count by 238 words per minute and the speaking card by 150; both rates are population averages used as planning estimates, not targets, and the section on reading and speaking time above explains where each comes from and where it stretches.

Why is my Reading Ease score negative, or above 100?

Convention reads the scale as 0 to 100, but the formula has no floor or ceiling. Long words push the syllables-per-word term far enough to take the score below zero, and very short words in short sentences push it above 100. The analyzer reports the raw value rather than clamping it; the dense sentence in the worked example on this page scores −150.2.

Why is a word I used often missing from the keyword table?

Three reasons. The ranking skips a fixed set of 74 common English function words, skips single-character tokens such as "a" or a lone digit, and shows only the ten highest entries, with ties broken alphabetically. A word that is present but outranked is not missing. Density is the word's count as a share of every word in the text, with the skipped function words still in the base.

Why does the readability panel ask for a full sentence?

Both Flesch formulas divide by the sentence count and by the word count, so the engine scores only when it has found at least one word and one sentence. Any fragment containing a letter or digit counts as a sentence even without ending punctuation, so a single unpunctuated line will score; an empty box, or one holding only punctuation or emoji, will not.

Does the analyzer handle text that is not in English?

The word, sentence, and unique-word counts use Unicode-aware rules, so accented and non-Latin letters are recognised as parts of words. The syllable heuristic behind the two Flesch scores is an English heuristic that looks only at the letters a–z, so the readability panel is built for English text and should not be read as a measure of anything else.

Is anything I paste sent to a server?

No. The analysis is JavaScript running in your own browser. Your text is never transmitted, logged, or stored, and analysing it makes no network request at all — your browser's network tab will show none. The input cap is 2,000,000 characters.

Analysis happens locally in your browser — nothing you paste is transmitted or stored. See the methodology page.