Encoding & pricing

Pages, not messages

SMS is billed per page (or “segment”), per recipient. How many pages your message costs depends entirely on which characters are in it.

Encoding	Chars per page	Used when
GSM-7	160	Every character is in the GSM alphabet
UCS-2	70	Any character isn’t

That second row is where the money goes.

One emoji re-encodes the entire message

Encoding is chosen for the whole message, not per character.

A single 👋, curly quote, or accented letter forces the entire message into UCS-2 — dropping your page size from 160 characters to 70. A 150-character message that cost one page now costs three.

That’s a 3× bill increase from one character, and it’s the most common billing surprise in SMS.

What’s in GSM-7

Safe: A–Z, a–z, 0–9, space, and @ £ $ ¥ è é ù ì ò Ç Ø ø Å å Δ _ Φ Γ Λ Ω Π Ψ Σ Θ Ξ Æ æ ß É ! " # ¤ % & ' ( ) * + , - . / : ; < = > ? ¡ Ä Ö Ñ Ü § ¿ ä ö ñ ü à

These count as two characters each: ^ { } \ [ ] ~ | €

Everything else — emoji, “smart quotes”, – en-dashes, … ellipses, most non-Latin scripts — forces UCS-2.

The invisible culprits

The characters that catch people out aren’t emoji, which are at least visible. They’re the ones a word processor or a designer’s copy-deck silently inserts:

Looks like	Actually is	Safe?
'	' curly apostrophe (U+2019)	❌ UCS-2
'	' straight apostrophe	✅ GSM-7
"…"	"…" curly quotes	❌ UCS-2
-	– en-dash (U+2013)	❌ UCS-2
...	… ellipsis (U+2026)	❌ UCS-2

Copy a message out of Word, Google Docs, or Notion and it will almost certainly contain curly quotes. Your 158-character message just became three pages, and nothing about it looks different.

Normalise before you send:

const toGSM = (s) =>
  s.replace(/[‘’]/g, "'")   // curly single → straight
   .replace(/[“”]/g, '"')   // curly double → straight
   .replace(/[–—]/g, '-')   // en/em dash  → hyphen
   .replace(/…/g, '...');        // ellipsis    → three dots

toGSM('We'll see you at 5–6pm…'); // → "We'll see you at 5-6pm..."

That one function can cut a campaign’s cost by two thirds.

Multi-page messages

Longer messages are split and reassembled on the handset — the recipient sees one message. But each page carries a header, so the usable space per page shrinks:

Encoding	Single page	Per page when concatenated
GSM-7	160	153
UCS-2	70	67

So a 320-character GSM-7 message is three pages (153 + 153 + 14), not two.

Message	Encoding	Chars	Pages
Your code is 481920.	GSM-7	20	1
160 chars, plain ASCII	GSM-7	160	1
161 chars, plain ASCII	GSM-7	161	2
Your code is 481920. 👋	UCS-2	22	1
71 chars with one emoji	UCS-2	71	2
150 chars with one curly quote	UCS-2	150	3

Check the cost before you commit

Every accepted batch returns a cost_estimate. pages is the page count for one message — multiply by recipients for the batch:

"cost_estimate": {
  "total_cost": 19.2,
  "currency": "NGN",
  "pages": 3,
  "accepted_recipients": 2,
  "sufficient": true
}

pages: 3 on a message you thought was short is your signal that something re-encoded it to UCS-2.

const batch = await sendSMS({ to, content });

if (batch.cost_estimate.pages > expectedPages) {
  logger.warn(
    { reference_id: batch.reference_id, pages: batch.cost_estimate.pages },
    'message segmented more than expected — check for non-GSM characters',
  );
}

Catch it in CI, not on the invoice

Assert on encoding in your test suite. A template that renders to one page today can silently become three when someone pastes in a curly quote next sprint.

test('OTP template stays GSM-7 and fits one page', () => {
  const msg = renderOTP({ code: '481920' });
  expect(isGSM7(msg)).toBe(true);
  expect(msg.length).toBeLessThanOrEqual(160);
});

Practical advice

• Keep OTPs under 160 GSM-7 characters. They’re your highest-volume, most cost-sensitive traffic. No emoji, ever.
• Run marketing copy through a normaliser. Marketing copy comes from documents; documents contain curly quotes.
• Emoji are a deliberate choice. They lift engagement — sometimes worth 3× the cost. Just make it a decision, not an accident.
• Watch pages in your logs. It’s the cheapest early warning you have.