You set up your AI receptionist to send SMS confirmations. It works beautifully. Customers get clean, friendly messages confirming their booking or quoting a price. Then you check the first month's invoice and the SMS line is three times what you budgeted. No one warned you that a single character - an em-dash, a curly quote, a degree symbol - can double or quadruple the cost of every message your AI sends.

This is not a carrier gouge. It is a character encoding problem that most AI voice platforms never explain, and it quietly drains thousands of dollars a year from Australian businesses running automated SMS workflows.

GSM-7 versus UCS-2: the invisible switch

Every SMS travels in one of two character sets. GSM-7 is the original: 160 characters per segment, covering basic Latin letters, numbers, and a handful of punctuation marks. UCS-2 is Unicode: 70 characters per segment, supporting every emoji, accented letter, and typographic flourish your AI might generate.

The carrier does not ask which encoding you want. If your message contains even one character outside the GSM-7 table, the entire message switches to UCS-2. A 150-character confirmation that includes a single curly apostrophe suddenly costs the same as a 70-character UCS-2 message - and if it runs to 71 characters, you pay for two segments.

Most AI platforms generate natural-sounding text. Natural-sounding text, by default, includes smart quotes, em-dashes, and ellipses. Every one of those characters forces UCS-2. Your AI sends what looks like plain English, and you pay double or quadruple without noticing until the bill arrives.

The characters that trigger the switch

Here is the short list of common culprits:

• Curly quotes and apostrophes (" " ' ') instead of straight (" ')
• Em-dashes and en-dashes (- -) instead of hyphens (-)
• Ellipsis character (...) instead of three full stops (...)
• Degree symbol (°) and other extended punctuation
• Any emoji or non-Latin character

A message that reads "We'll confirm your appointment - see you Thursday!" looks identical to a human reader whether it uses straight quotes and a hyphen or smart quotes and an em-dash. The carrier sees them as completely different messages, and one costs twice as much to send.

Why AI makes the problem worse

Large language models are trained on published text: books, articles, formal writing. Published text uses typographically correct punctuation. When you prompt an AI to "write a friendly confirmation message", it defaults to the same style it learned from millions of training examples. It writes "We're" with a curly apostrophe because that is what correct English looks like in print.

The AI is not wrong. It is doing exactly what you asked. But SMS is not print, and the character set that looks polished on a web page triggers a cost multiplier in the carrier's billing system.

On a call we listened to last week, a VoxReach hybrid agent confirmed a tradie's quote and sent an SMS follow-up. The message included "We'll send the invoice - thanks for choosing us!" with smart quotes and an em-dash. Seventeen words, 54 characters. Because of two invisible Unicode substitutions, it billed as a UCS-2 message instead of fitting comfortably in a single GSM-7 segment. The customer paid $0.60 for a message that should have cost $0.30.

Scale that across hundreds of confirmations a month, and the difference is real money.

The one-line fix

VoxReach strips smart quotes, em-dashes, and ellipsis characters from every SMS before sending. The output is plain ASCII. Straight quotes, single hyphens, three full stops. It reads the same to the recipient, costs half as much to send, and fits more characters per segment.

You can apply the same filter to any AI-generated SMS workflow:

• Replace curly quotes (" " ' ') with straight equivalents (" ')
• Replace em-dashes and en-dashes with a single hyphen (-)
• Replace ellipsis (...) with three full stops (...)
• Strip emojis unless you genuinely need them and budget for UCS-2

If your platform does not offer this as a built-in toggle, write a pre-send sanitiser or ask your dev team to add one. The code is trivial. The savings compound every month.

When UCS-2 is worth paying for

Sometimes you need Unicode. If your customer base includes non-English names, if you operate in a multilingual market, or if your brand voice genuinely requires emoji, UCS-2 is the right choice. Just budget for 70 characters per segment instead of 160, and write shorter messages.

What makes no sense is paying UCS-2 rates by accident because your AI slipped in a character you never intended to send.

What to do now

Pull your last SMS invoice. Look at average message length and total cost per message. If you are paying more than $0.60 for a standard confirmation, check whether your platform is forcing UCS-2 encoding. If it is, ask your provider to strip non-GSM characters before sending, or switch to a platform that does it automatically.

VoxReach strips by default. Every SMS goes out in plain ASCII unless you explicitly configure a UCS-2 template. You see the saving in the first week.

Sign up free at app.voxreach.com.au/signup - no card required, 30 minutes of calls to test your own workflows. Or ring +61 2 5926 2202 to talk to Frank, our live AI broker, running on the same platform.