I will present a new proof of the Ehrhard inequality. In fact I will talk about a more general result and the Ehrhard inequality will be consequence of it. The idea of the method is similar to Brascamp--Lieb's approach to Prekopa--Leindler inequality via sharp reverse Young's inequality for convolutions. Indeed, we shall rewrite essential supremum as a limit of Lp norms but with very specially chosen test functions and measures. Next rewriting Lp norm by duality as a scalar product the question boils down to an estimate of double integral of compositions of test functions by the mass of these functions. To verify the last estimate which looks like Jensen's inequality we will use a subtle inequality, a ``modified Jensen's inequality'', which in its turn boils down to the fact that a corresponding quadratic form has a definite sign, and this is the main technical part of the method. If time allows we will show that in the class of even probability measures with smooth strictly positive density Gaussian measure is the only one which satisfies the functional form of the Ehrhard inequality on the real line with their own distribution function.