The long-distance leg of a call between two ordinary mobile phones is likely to use IP already, and neither caller has any way to tell.
VoIP can mean a transport detail or a regulatory category with obligations attached.
What does the plain word cover?
Any arrangement where the audio travels as packets on an IP network.
That includes a softphone on a laptop, a desk phone plugged into an office switch, a browser tab, a carrier's own backbone between two cities, and a Session Initiation Protocol (SIP) trunk connecting a phone system to a provider. It says nothing about who you can call, what number you present, whether anyone can call you back, or what happens when you dial an emergency number.
Because it covers all of that at once, "is it VoIP" is rarely the question somebody actually has. The questions underneath it are usually about reach and obligation, and those have a sharper definition.
What is interconnected VoIP?
A defined category with a four-part test, and the fourth part is where the weight sits.
The first three parts describe most things that carry voice over the internet. They require two-way voice, broadband and internet protocol-compatible customer premises equipment (CPE). The fourth is the discriminator: can the users reach, and be reached from, the public switched telephone network? A service whose users can only call other users of the same service fails it however good the audio is, and a service that hands calls to and from the telephone network meets it however small it is.
In the United States 47 CFR 9.3 carries that four-part definition:
An interconnected Voice over Internet Protocol (VoIP) service is a service that:
(i) Enables real-time, two-way voice communications; (ii) Requires a broadband connection from the user's location; (iii) Requires internet protocol-compatible customer premises equipment (CPE); and (iv) Permits users generally to receive calls that originate on the public switched telephone network and to terminate calls to the public switched telephone network.
That regulation adds an asymmetry worth reading twice, because it is the kind of thing that surprises people building outbound-only products:
Notwithstanding the foregoing, solely for purposes of compliance with the Commission's 911 obligations, an interconnected VoIP service includes a service that fulfills each of paragraphs (1)(i) through (iii) of this definition and permits users generally to terminate calls to the public switched telephone network.
In other words, for emergency-calling purposes the test drops the inbound half. Being able to dial out is enough.
What does VoIP change about a call?
It moves several guarantees from the network onto you.
A circuit-switched call reserved a path and gave you a fixed amount of capacity for its whole duration. Packets do not work that way, so three properties that used to be the network's problem become properties of your own path: how much audio a codec spends per second, how consistently packets arrive, and how many go missing. Those are the levers behind almost every complaint about call quality on an IP path, and none of them is visible in a call record that reads as answered.
The second change is architectural, and what is SIP covers it in full: setting a call up and carrying its audio become separate protocols on separate paths. That is what makes a connected call with no sound a routine failure rather than a contradiction.
What about emergency calls?
This is where a number stops implying a place, and it is the sharpest practical consequence of the whole category.
On the older network, the line reaching a building was the location, so a call to an emergency number could be routed from the circuit it arrived on. A number reached over the internet carries no such fact. The equipment can be anywhere, it can move between calls, and nothing in the signalling says where it is. Emergency routing therefore depends on a location the provider holds rather than on anything the call carries, which is why it is an obligation providers take on explicitly rather than a capability that comes free.
You need a separate route for emergency calls because Bird Voice does not carry calls to emergency services. A call to 112, 911, 999 or any other emergency number is refused on every trunk in every country, and that does not depend on your plan, balance or destinations. The refusal comes back as a 503 so a phone system with another route in its dial plan can try it. Placing calls covers what to configure instead. Give your phone system a separate path for those numbers and test it.
So what should I call what I am building?
Describe the reach, not the transport.
If your users can dial ordinary phone numbers and be dialled from them, that is the fact that matters to a regulator, a procurement questionnaire and a customer, and the answer to almost any question that starts "is this VoIP" is a sentence about which of the four parts above your service meets. If they can only reach each other, say that instead, because it is the more useful description and it is the one that keeps the obligations straight.
What is a SIP trunk covers the connection that puts a phone system on the telephone network, and what is a voice API covers the same reach from application code.
In short
VoIP describes a transport, not a product.
It means the audio travels as IP packets rather than over a circuit. Almost every call is partly VoIP now, which makes the plain word close to useless as a category.
The useful category is interconnected VoIP, and it has a four-part test.
Two-way voice, a broadband connection, IP-compatible equipment, and the ability to reach and be reached from the public telephone network.
The fourth part is what separates a phone service from an app.
A service whose users can only call other users of the same service does not meet it, whatever the audio quality.
Reaching the network is not the same as being locatable on it.
A number no longer implies a place, which is why emergency calling is the obligation that separates carriers from platforms.