Formally, Rice's Theorem can be stated as follows.
Let C be a set of languages. Let the language L defined as L = {⟨M⟩ | L(M) ∈ C } .
Then either L is empty, or it contains the descriptions of al l Turing machines, or it is undecidable.
Informally, it says that a decision problem where we are given a Turing machine and we are asked to determine a property of the language recognized by that machine, this decision problem is always undecidable. The only exceptions will be the trivial properties that are always true or always false.
In other words, we can say Any nontrivial property about the language recognized by a Turing machine is undecidable. A property about Turing machines can be represented as the language of all Turing machines, encoded as strings, that satisfy that property. The property P is about the language recognized by Turing machines if whenever L(M)=L(N) then P contains (the encoding of) M iff it contains (the encoding of) N. The property is non-trivial if there is at least one Turing machine that has the property, and at least one that hasn't. (reference)