Character counter

What one character means once emoji and accents are involved

There are three defensible answers to the question "how many characters is this", and they disagree. A grapheme cluster is what a reader would point at and call a character. A code point is one entry in Unicode. A UTF-16 code unit is one of the sixteen-bit slots a browser stores the text in. For unaccented English the three numbers are identical, which is why the distinction stays invisible until it suddenly costs you something.

Accents are where it stops being theoretical. The letter "é" exists twice in Unicode: as a single precomposed entry, and as a plain "e" followed by a combining acute accent. Both render identically on screen and neither is wrong. A counter working in code points reports one for the first spelling and two for the second, so two strings that look the same produce different totals and the reader has no way to see why.

Text typed on a keyboard is usually precomposed. Text pasted out of a PDF, exported from some Mac applications, or produced by an older publishing system is often decomposed instead. Counting grapheme clusters removes the problem: a base letter and everything stacked on it form one cluster, so both spellings of "é" come back as one.

Why two counters give two answers

Any two tools measuring the same paragraph can disagree. The causes are mundane rather than mysterious, and knowing which one is in play tells you how much margin a given ceiling needs.

None of the causes below makes one count wrong. They make the destination the arbiter: write against this page, then check the finished text in the field that will accept or reject it, particularly if you are within the last five characters of a hard ceiling.

• Normalisation. Some systems convert text to a canonical form before measuring it and some do not, which moves the total for any string containing combining marks.
• Trimming. A destination may strip leading and trailing whitespace before it counts, so a stray newline at the end of a paste costs a character in one place and nothing in another.
• Weighting. Some services charge more than one unit for characters outside the Latin alphabet, so a sentence in Japanese consumes more of the budget than a Latin-script sentence of the same length.
• Substitution. Links are frequently measured at a fixed length whatever the real URL is, because the service intends to rewrite them on the way out.
• Invisible characters. Zero-width joiners, soft hyphens and variation selectors occupy no width on screen and still take a slot in almost every counting scheme.

Where a character limit actually bites

Character budgets come in two kinds. Some are enforced: the field refuses one more keystroke, or the message splits and you pay twice. Others are cosmetic: nothing rejects your text, it simply gets cut off in front of the audience you wrote it for. The second kind does more damage, because nothing tells you it happened.

SMS: how one emoji turns 160 characters into 70

SMS is the clearest case of a ceiling that is not really about characters at all. The message is encoded before it is sent, and the encoding sets the ceiling.

The default encoding is GSM-7, a compact alphabet of roughly 128 symbols plus a small extension set. It covers unaccented Latin letters, digits, ordinary punctuation and a specific short list of accented letters — é and à are in it, á, ã, õ and ê are not. A message that stays inside that alphabet fits 160 characters into one segment.

The moment a single character falls outside it, the message switches to UCS-2, which spends two bytes on everything. Not on the offending character: on the whole message. The ceiling drops to 70.

With spaces, without spaces, and who asks for which

Both totals are shown because different people ask for different ones and are rarely explicit about which.

With spaces is the correct default for anything about fit. A space occupies width on a line, in a field and inside an SMS segment, so any limit describing how much text will physically go somewhere counts it. Meta descriptions, form fields, bios and messages are all with-spaces ceilings.

Without spaces turns up where the number stands in for how much content there is rather than how much room it needs. Some translation and typesetting conventions price that way, and a few course rubrics specify it. If a brief says "characters" and stops there, it means with spaces — but the gap between the two totals on ordinary prose is about a sixth of the text, which is large enough to be worth one question before you write to the wrong one.

Three things inflate a with-spaces count without looking like they should. A double space after a full stop, kept out of habit or inherited from an old document. A non-breaking space, which a word processor inserts to stop a line breaking awkwardly and which no keystroke distinguishes from an ordinary one. And a soft hyphen, an invisible marker for where a word may be split, which arrives in quantity when you paste from a PDF. All three are counted. Only the first can be seen.

Why there is no table of platform limits here

It would be the easiest section on the page to write and the fastest to rot. Ceilings on social networks, ad platforms and content systems change on their owners' schedule, and nobody notifies the pages that quoted the old number. A stale limit is worse than no limit, because a reader who finds a figure here has no reason to doubt it and no way to check it without leaving.

The counter takes its ceilings from you instead. Name a target, set its number, and it is tracked against whatever is in the box. Several can run at once, which is the normal case — one piece of text often has to serve as a headline in one place, a summary in another and a caption in a third.

Getting a ceiling you can rely on takes about a minute:

• Open the field the text is destined for: the real composer, admin panel or submission form, not a screenshot of it.
• Read the counter it shows you, or paste in filler and watch where it stops accepting input. That refusal point is the only ceiling that is definitely real.
• Enter that number here as a named target.
• Subtract a small margin if the destination trims whitespace or rewrites links, so that a two-character surprise on submission day does not cost you the last clause.