IVR menu prompts
The menu is the one recording every caller hears, which makes it the one recording where a mispronounced extension number costs you the most calls.
What goes wrong
Menu prompts are short, so they get treated as the easy part. They are not. A greeting contains the two things general-purpose text-to-speech is worst at: an extension or reference number that must be read digit by digit, and a pause structure that decides whether the caller can keep four options in their head long enough to choose one.
Read a menu aloud at the speed a model defaults to and the options blur together. The caller presses nothing, the timeout fires, and the call lands in the queue you built the menu to avoid.
Scripts you can use
Written to be read aloud by a machine, not by a person. The note under each one says which part a general-purpose voice model gets wrong.
Main greeting
Thank you for calling Northwind Supply. For order status, press 1. For returns and refunds, press 2. For anything else, press 0 to speak to someone.
Keep one option per sentence. The full stop is what produces the pause the caller needs to make a choice — a comma will not.
Department transfer
Connecting you to dispatch. If you are cut off, call back on 0 2 0 7 9 4 6 0 1 2 3 and ask for extension 4 4 0 2.
The number has to be read as ten separate digits, and the extension as four. A model that normalises digit strings will offer "four thousand four hundred and two", which no caller can write down.
Hold message
All of our advisers are busy. Your call is held in order and will be answered as soon as one is free. If you would rather not wait, press 1 to leave a number for a callback.
No position-in-queue claim unless your telephony stack actually supplies it. A recorded "you are third in line" that is not true is worse than silence.
Closed for the day
Thank you for calling Northwind Supply. Our office is closed. We are open Monday to Friday, 9am to 5pm. For urgent delivery problems, call 0 8 0 0 1 1 2 2 3 3.
See the after-hours page for the opening-hours formats that change per language — 9am to 5pm is not how a French or German caller hears it.
Before you record it
- Every phone number, extension and reference in the script is spaced digit by digit, and you have listened to it played back.
- One option per sentence, four options maximum before a repeat.
- The last option is a route to a person. A menu with no human exit generates complaints, not deflection.
- The closed message names actual opening hours, not "normal business hours".
Questions
How long should an IVR menu prompt be?
Under fifteen seconds for the greeting and options combined. Callers hold roughly four options in working memory; past that they press whatever came last, which routes them wrongly and costs you a transfer.
Why does the extension number sound wrong in my IVR recording?
Most text-to-speech models normalise digit strings into quantities, so 4402 becomes "four thousand four hundred and two". A phone system needs per-digit delivery: "four four zero two". That normalisation is the single most common cause of unusable IVR audio.
Generate these with IVR Voice Gen
IVR Voice Gen is text-to-speech built for phone systems: digit strings, references, dates and addresses are read the way a caller can write them down, in English, French, German, Spanish and Portuguese. It is in early access.