Word Counter

Almost every piece of writing is written against a number. An essay has a minimum, a meta description has a maximum, a conference abstract has both, and a post has a limit the platform enforces silently by truncating you. The Word Counter reports words, characters, sentences, paragraphs and lines as you type, and adds the figures that usually send you looking for a second tool: visible characters against UTF-16 units and UTF-8 bytes, unique words, average sentence length, the longest sentence quoted back at you, and roughly how many pages the piece runs to.

Set a target and the tool becomes a gauge rather than a report. Choose a platform — an X post, a Bluesky post, an Instagram caption, an SEO title, a meta description, a single SMS — and a meter fills as you write, turning amber near the limit and telling you how far over you are once you pass it. Each preset is measured in the unit that platform genuinely counts, which is not always the obvious one: Bluesky counts visible characters, so a post full of emoji fits where a naive count says it should not. Word targets work the same way, with the common essay and article lengths one click away.

Two things make the counting itself more trustworthy than the usual whitespace split. Word tokens have their wrapping punctuation trimmed and are discarded if nothing but punctuation is left, so a spaced dash or a stray bracket is not counted as a word. Sentence detection knows that a full stop inside "Dr.", "e.g.", "U.S.A.", "3.5" or "example.com" is not the end of a sentence — the naive approach reads "Dr. Smith went to the U.S.A. yesterday." as five sentences instead of one.

You can also decide what should not count. Exclude URLs and email addresses, exclude numbers, or exclude Markdown syntax and fenced code blocks while keeping the prose inside them — useful when a word limit is meant to cover argument rather than apparatus. Whenever an exclusion is active the tool states how many of the characters you typed are actually being counted, and the exports carry both numbers. Select any passage and the footer reports that selection's own count. Nothing is uploaded: the text is analysed in your browser, and it stays there.

Words are found by splitting on whitespace and on em and en dashes — the two dashes that join clauses rather than words — then trimming punctuation from each end and discarding anything with no letter or digit left in it. That is why "well-known" and "don't" each count as one word, why "word—word" counts as two, and why a dash standing alone between two spaces counts as nothing at all.

A full stop, question mark or exclamation mark only ends a sentence when whitespace or the end of the text follows it, which by itself rules out decimals and stops inside web addresses. Known abbreviations and initialisms hold the sentence open as well. The bias is deliberately toward joining rather than splitting: "He moved to the U.S.A. Then he left." reads as one sentence, because nothing short of a grammar parser can tell it apart from a continuation. Paragraphs are the blocks between blank lines, so a run of consecutive lines is one paragraph, not several.

Reading time defaults to 200 words per minute and speaking time to 130, both shown to the nearest second and both adjustable — a slide deck read aloud and a dense technical paper are nowhere near the same pace. Pages default to 250 words, which is about a double-spaced manuscript page; single-spaced is nearer 500, so that figure is adjustable too.

Character limits are less straightforward than they look, which is why the tool reports four different character figures. Most platforms count UTF-16 code units, the number a JavaScript string reports as its length. A reader counts what they can see. An emoji assembled from several joined code points is one visible character but several code units, so a caption that looks comfortably short can be rejected — and the giveaway is precisely those figures disagreeing. UTF-8 bytes matter again for anything with a storage or transmission budget, such as a database column or a QR code payload.

Platform arithmetic adds its own quirks, so the presets carry notes where it matters. A link on X consumes a flat 23 characters however long the URL is. A single non-GSM character anywhere in an SMS switches the entire message to 70-character segments rather than 160. Google truncates a title tag by pixel width rather than character count, which makes 60 a guide rather than a rule. Academic limits are their own world: references are usually excluded, the abstract often is, and institutions disagree about footnotes and tables — which is what the URL, number and Markdown exclusions are for.

Beyond hitting a number, two of these figures are worth watching while you write. Average words per sentence is a fast readability proxy — much above 20 and prose gets heavy — and the sentence spread shows whether you have any variation or just a wall of long ones. Lexical diversity, unique words as a share of total words, drops when a draft leans on the same handful of terms. For a formal grade rather than a proxy, the Readability Checker is next door.