why do we prefer ? to ?? operator in c#?

The null coalesce operator is much clearer when checking for null, that is its main purpose. It can also be chained.

object a = null;
object b = null;
object c = new object();
object d = a ?? b ?? c; //d == c.

While that operator is limited to null checking, the ternary operator is not. For example

bool isQuestion = true;
string question = isQuestion ? "Yes" : "No";

I think people just aren’t aware of the null coalesce operator so they use the ternary operator instead. Ternary existed before C# in most C style languages so if you don’t know C# inside and out and/or you programmed in another language, ternary is a natural choice. If you are checking for null though, use the null coalesce operator, it is designed for that, and the IL is slightly optimized (compare ?? to an if then else).

Here is an example comparing the use of each

object a = null;
object b = null;
object c = null;

object nullCoalesce = a ?? b ?? c;

object ternary = a != null ? a : b != null ? b : c;

object ifThenElse;

if (a != null)
    ifThenElse = a;
else if (b != null)
    ifThenElse = b;
else if (c != null)
    ifThenElse = c;

First, just look at the syntax for null coalesce, it is way clearer. Ternary is really confusing. Now lets look at the IL

Null Coalesce Only

.entrypoint
.maxstack 2
.locals init (
    [0] object a,
    [1] object b,
    [2] object c,
    [3] object nullCoalesce)
L_0000: ldnull 
L_0001: stloc.0 
L_0002: ldnull 
L_0003: stloc.1 
L_0004: newobj instance void [mscorlib]System.Object::.ctor()
L_0009: stloc.2 
L_000a: ldloc.0 
L_000b: dup 
L_000c: brtrue.s L_0015
L_000e: pop 
L_000f: ldloc.1 
L_0010: dup 
L_0011: brtrue.s L_0015
L_0013: pop 
L_0014: ldloc.2 
L_0015: stloc.3 
L_0016: ldloc.3 
L_0017: call void [mscorlib]System.Console::WriteLine(object)
L_001c: ret 

Ternary Only

.entrypoint
.maxstack 2
.locals init (
    [0] object a,
    [1] object b,
    [2] object c,
    [3] object ternary)
L_0000: ldnull 
L_0001: stloc.0 
L_0002: ldnull 
L_0003: stloc.1 
L_0004: newobj instance void [mscorlib]System.Object::.ctor()
L_0009: stloc.2 
L_000a: ldloc.0 
L_000b: brtrue.s L_0016
L_000d: ldloc.1 
L_000e: brtrue.s L_0013
L_0010: ldloc.2 
L_0011: br.s L_0017
L_0013: ldloc.1 
L_0014: br.s L_0017
L_0016: ldloc.0 
L_0017: stloc.3 
L_0018: ldloc.3 
L_0019: call void [mscorlib]System.Console::WriteLine(object)
L_001e: ret 

If Then Else Only

.entrypoint
.maxstack 1
.locals init (
    [0] object a,
    [1] object b,
    [2] object c,
    [3] object ifThenElse)
L_0000: ldnull 
L_0001: stloc.0 
L_0002: ldnull 
L_0003: stloc.1 
L_0004: newobj instance void [mscorlib]System.Object::.ctor()
L_0009: stloc.2 
L_000a: ldloc.0 
L_000b: brfalse.s L_0011
L_000d: ldloc.0 
L_000e: stloc.3 
L_000f: br.s L_001a
L_0011: ldloc.1 
L_0012: brfalse.s L_0018
L_0014: ldloc.1 
L_0015: stloc.3 
L_0016: br.s L_001a
L_0018: ldloc.2 
L_0019: stloc.3 
L_001a: ldloc.3 
L_001b: call void [mscorlib]System.Console::WriteLine(object)
L_0020: ret 

IL isn’t one of my strong points, so maybe someone can edit my answer and expand on it. I was going to explain my theory, but I’d rather not confuse myself and others. The number of LOC is similar for all three, but not all IL operators take the same length of time to execute.