SMS pitfalls: GSM-7 versus UCS-2 encoding and message segments

What it is

A single SMS carries 140 bytes of user data. With the GSM 7-bit default alphabet (GSM-7) that is 160 characters; with UCS-2, two bytes per character, it is 70. GSM-7 covers 128 basic characters (Latin letters, digits, some accented letters, common symbols) plus an extension table for characters such as {, [, € and \, which need an escape and count as two characters. One character outside GSM-7 (a typographic apostrophe, an em dash, an emoji, most non-Latin scripts) switches the entire message to UCS-2. Longer texts are split into segments carrying a User Data Header for reassembly, which leaves 153 GSM-7 or 67 UCS-2 characters per segment; the handset reassembles them and providers typically bill per segment. The cited documentation gives the numbers and notes that a 161-character GSM-7 message is transmitted as two segments.

Why it matters

A 150-character notification that an editor "improved" with curly quotes becomes a three-segment UCS-2 message at three times the segment count. A login code that wraps into a second segment shows the code after the part the user reads first.

How to apply

• Count before sending: decide the encoding (every character in the GSM-7 basic or extension table, otherwise UCS-2), count extension characters twice, and divide by 160 or 70 for one segment, 153 or 67 for several.
• Normalise punctuation on the GSM-7 path: straight quotes, hyphen-minus, three dots instead of an ellipsis character. Some providers offer this as a "smart encoding" option; doing it in the template is more predictable.
• Put the essential content (the code, the link) within the first 67 characters so it survives any encoding and any segmentation.
• Use a short domain and no tracking parameters in links; a long URL alone can push a message into a second segment.
• Log the encoding and segment count the provider reports per message and alert when the average creeps up after a template change.
• For scripts that need UCS-2 (Cyrillic, Arabic, CJK), budget 70 characters from the start and do not mix scripts in one template.

Pitfalls

Counting UTF-8 bytes or Unicode code points instead of GSM-7 or UCS-2 units. Assuming an emoji costs one character when many need two 16-bit units (a surrogate pair). Provider caps on concatenated length (the cited table lists 1,600 GSM-7 or 700 UCS-2 characters). Sender-ID rules and regional regulations, which vary by country and are not covered here.