SMS Segments and Character Counting

A text message has two lengths: the characters you can see, and the segments it occupies once it is encoded for sending. The two agree only when the message uses a plain alphabet and stays under a budget that depends on that alphabet. This guide explains how the segment count is derived under GSM 03.38, the standard the site's engine implements: the two encodings and their budgets, the characters that cost double, the single character that can more than halve the budget, and why a character-limit check and an SMS segment count can disagree about the same string. Every figure was computed at build time by the engine behind the text analyzer, from short original messages written for this guide. No tool on this site currently shows a segment count: the engine computes it, and this guide and the methodology page document what it computes.

Two encodings, two budgets

GSM 03.38 defines an alphabet, GSM-7, whose messages are measured in septets: each character in a fixed basic table costs one septet. A message written entirely from that table fits 160 septets in a single segment; if it needs more than one, the message is concatenated and each segment carries 153 septets instead. The moment a message contains any character that is not in the table, the whole message is encoded as UCS-2, where a segment holds 70 UTF-16 code units on its own or 67 when concatenated, and most emoji cost two units. There is no mixing: one character outside the table changes the encoding of every character in the message.

The engine walks the text one character at a time, adding one septet for a basic-table character and two for an extended one; on meeting a character in neither table it stops, marks the message UCS-2, and takes the text's length in UTF-16 code units instead. The segment count is one if the units fit the single-segment budget, otherwise the units divided by the concatenated budget and rounded up; an empty message reports zero.

The cliff at 160 septets

Take a booking confirmation written from the basic table: "Your table for four is booked for Saturday at seven. Reply YES to confirm or NO to release it. Parking is behind the building and the side door stays open late." That is 160 characters and, since none of them is extended, 160 septets — reported as 1 GSM-7 segment with 0 septets to spare. Change the last word from "late" to "later" and the message is 161 characters and needs 2 segments. The per-segment figure fell from 160 to 153, so the two segments together hold 306 septets and the engine reports 145 remaining. That figure is the capacity of the segments the message occupies — segments multiplied by the concatenated budget — minus the units used. Because segments fill in order, it is also the free space in the last one: the second segment carries only 8 of its 153 septets. One character at the boundary does not add one unit of cost; it adds a whole segment and reduces what each segment can carry.

Characters that cost two septets

The GSM-7 extended table holds a short list of characters that cost two septets each — € [ ] { } ~ \ | ^ and the form-feed control character. They are still GSM-7, so they do not trigger UCS-2, but they quietly inflate the unit count. A short lunch note — "Lunch is 9 € today, pay at the counter." — is 39 characters yet 40 septets, because the euro sign counts twice. Far from any boundary, the extra septet is harmless.

At the boundary it is not. This pickup notice is exactly 160 characters: "Your order is ready for pickup. The balance due is 24€ and the shop is open until six today. Bring the order number from this message and a photo ID, thank you." A character counter says it fits one segment. The engine says 161 septets, 2 segments, GSM-7, with 145 septets to spare across the pair — the single euro sign pushed it over. The other extended characters behave the same way.

The switch to UCS-2

The most expensive character in a message is often one you did not notice typing. Here is a meeting reminder with two straight apostrophes: "Don't forget the team meeting moved to 3 pm in the small room. Bring last week's numbers and the draft budget so we can settle the venue question before Friday." It is 160 characters, 160 septets, 1 GSM-7 segment with 0 to spare. Replace both apostrophes with the typographic curly form and nothing visible changes: still 160 characters. But the curly apostrophe is not in the GSM-7 table, so the message is UCS-2, its 160 UTF-16 units divide into segments of 67, and it is reported as 3 segments with 41 units unused out of a 201-unit capacity.

An editor that replaces straight quotes with curly ones as you type makes that substitution invisibly, and typographic dashes and ellipses are outside the table too. Accented letters are decided the same way, by membership rather than appearance. The engine's basic table includes é, è, ù, ì, ò, à, ä, ö, ü, ñ, ß, å, æ and ø with a handful of capitals and several Greek capitals, so "café" is GSM-7 at 4 septets. A lowercase ç is not in the table (only the capital Ç is), so "façade" is UCS-2 at 6 units.

Emoji cost two units, after costing the alphabet

No emoji is in the GSM-7 table, so an emoji does two things at once: it forces UCS-2 for the whole message and then, as most emoji do, costs two UTF-16 units rather than one. "Running ten minutes late, please order me a coffee 🙂" is 52 code points but 53 UCS-2 units, 1 segment with 17 to spare. The alphabet switch is the larger effect. A birthday note without the cake — "Happy birthday! The cake is in the kitchen, help yourself before noon" — is 69 septets, 1 GSM-7 segment with 91 left. Append a space and a cake emoji and it becomes 72 UCS-2 units, more than the 70-unit single-segment budget, so the engine reports 2 segments of 67 with 62 units unused out of 134. The space and the emoji added 3 units of text; the message gained a segment, and its per-segment budget fell to less than half of what it was.

Three ways to count one string

The engine's limit check counts Unicode code points — the convention the text analyzer uses for its character totals, described in what counts as a word, where emoji are characters rather than words. The SMS segmenter counts septets for GSM-7 and UTF-16 units for UCS-2. For plain letters, digits and punctuation all three agree. A typical emoji is one code point and two UTF-16 units, and some visible symbols are assembled from several code points joined together: the engine reports the single-glyph developer emoji as 3 code points for the limit check and 5 UTF-16 units for SMS. None of the numbers is wrong; they answer different questions.

Accents, and why the same word can count twice

An accented letter can be stored two ways: as a single precomposed code point, or as a base letter followed by a combining accent mark. Both render identically. The engine's letter count normalizes text to NFC before counting, so the two forms give the same letter and character totals; the limit check and the SMS segmenter take the text as it is. Consider a menu line in precomposed form — "Crème brûlée à la carte" — and the same line with every accent stored as a combining mark. The limit check counts 23 code points for the first and 27 for the second, 4 more, one per accent. The letter count reports 19 letters and 23 characters for the first and 19 letters and 23 characters for the second, because after normalization they are the same string.

SMS is harsher. "café" with a precomposed é is GSM-7 at 4 septets and 4 code points. The decomposed spelling, e followed by a combining acute, is 5 code points, and because the combining mark is not in the GSM-7 table the word becomes UCS-2 at 5 units — while the letter count still sees 4 letters. The count depends on what the text is made of, not on how it looks; text pasted from different sources can carry the same word in different forms, and retyping a suspicious word is the simplest fix.

What the engine computes, and the common misreadings

For any message the engine computes five things: encoding, units used, segment count, the per-segment budget at that segment count, and remaining capacity. No tool on this site currently displays them; the figures in this guide come from build-time runs of the engine. Four misreadings cause most confusion. Units are not characters: extended characters and most emoji cost two, and a decomposed accent costs its own code point. Remaining capacity is the free space in the last segment the message occupies, so a message that has just crossed into a second segment shows a large remainder, not a small one. The 160-septet budget belongs to GSM-7 alone; any non-table character means measuring against 70 and then 67. And the per-segment figure changes with the segment count, which is why a message can gain a segment while its remaining capacity goes up.

What the count cannot tell you

The engine implements the GSM 03.38 length rules and nothing beyond them. It cannot know whether a particular sending service adds text of its own, substitutes characters it does not support, or splits messages by another rule, and it models no platform's message limits. As with every count on this site, when a figure is contractual the only counter that matters is the recipient's; what this guide explains is the rule behind that figure, and why it moves when it does. The guides index covers the site's other counts the same way.

Frequently asked questions

Why does a message of exactly 160 characters count as 2 segments?

Because one of those characters is an extended GSM-7 character. The pickup example contains a euro sign, which costs two septets instead of one, so 160 characters become 161 units — one past the 160-septet single-segment budget. Once a GSM-7 message needs more than one segment, each segment holds 153 septets, so the message is reported as 2 segments with 145 septets to spare.

Why did curly apostrophes turn 1 segment into 3?

The straight apostrophe is in the GSM-7 table; the curly one is not. Any character outside the table switches the whole message to UCS-2, where the budget is 67 UTF-16 units per concatenated segment rather than 160 septets in a single one. The meeting example is 160 characters either way: 1 GSM-7 segment with the straight apostrophes, 3 UCS-2 segments with the curly ones.

How much does an emoji cost in an SMS?

Two things. First it forces UCS-2, because no emoji is in the GSM-7 table. Then, like most emoji, it costs two UTF-16 units rather than one. The coffee example is 52 code points but 53 UCS-2 units; the birthday example goes from 1 GSM-7 segment at 69 septets to 2 UCS-2 segments at 72 units after one emoji is appended.

Which accented letters stay in GSM-7?

Only the ones in the GSM-7 basic table. The engine's table includes é, è, ù, ì, ò, à, ä, ö, ü, ñ, ß, å, æ, ø and a handful of capitals, so "café" is GSM-7 at 4 septets. A lowercase ç is not in the table, so "façade" is UCS-2 at 6 units. The visible accent is not the deciding factor; membership in the table is.

Why do "characters" and "units" differ for the same text?

They count different things. The character-limit check counts Unicode code points; the SMS segmenter counts septets for GSM-7 or UTF-16 units for UCS-2. For plain letters and digits all three agree. For the single visible symbol in the developer-emoji example, the limit check counts 3 code points and the SMS segmenter 5 UTF-16 units, because the symbol is assembled from pictographic code points joined by a zero-width joiner: the limit check counts every code point once, while in UTF-16 the 2 pictographic code points need two units each and the joiner needs one.

Why did remaining capacity jump from 0 to 145 when "late" became "later"?

Because the message crossed into a new segment. Remaining capacity is the total capacity of the segments the message occupies minus the units used, and since segments fill in order that is the free space in the last segment. The 160-character booking example fills its single 160-septet segment exactly, leaving 0. The 161-character version needs 2 GSM-7 segments of 153 septets, a capacity of 306; the second segment carries only 8 septets, so 145 of it are free. A single-segment message measures against 160 for GSM-7 or 70 for UCS-2.

Every figure in this guide was computed at build time by the site's analysis engine, and analysis on the tool pages happens locally in your browser — nothing you paste is transmitted or stored. The counting conventions are documented on the methodology page.