Why three tones?
Special information tones were designed in the 1970s so that both people and machines could tell a failed call from a ringing one. Anyone hears a rising three-note phrase that means "listen to the announcement". A machine can do more: in North America each of the first two segments comes in a low or high frequency and a short (274 ms) or long (380 ms) version, and the combination names the reason. The third segment is always the same low, long 1776.7 Hz.
That made SIT useful to automatic dialers. Telemarketing and predictive-dialing systems listen for the first two segments and mark the number as bad, which is why some people used to put a SIT recording at the start of their answering-machine greeting to try to get off call lists. How well that works today is anecdotal; modern systems use many other signals, so treat it as folklore rather than a method.
The international version
Outside North America the ITU defines a simpler tone: 950, 1400, and 1800 Hz, in that order. Each lasts 330 ± 70 ms, with gaps of up to 30 ms, and about 1 s of silence follows before it repeats (ITU-T E.180 §7, which absorbed the old Q.35). It carries no reason code; the recorded announcement does that.
Sources and accuracy
The North American values follow the ANSI T1.401 / Telcordia encoding as it is publicly documented. Those standards are not free to read, so we mark these segments "approximate" and list them in our open questions. The ITU values come straight from the recommendation. We play only the tones: we never synthesize or ship the voice announcement that follows them.
Related: dial, ringback and busy tones by country · the caller-ID burst between rings.