Interaction-free measurement (IFM) infers the presence of an absorbing object from a photon that, in the counterfactual sense, never interacted with it, and is widely described as a route to minimally invasive sensing. We ask what it actually buys, in estimation-theoretic terms. Writing the Elitzur-Vaidman interferometer as a channel-estimation problem, its Fisher information for the transmissivity $T$ of an object in one arm is $[2T(1-T){]}^{-1}$, exactly half that of direct transmission probing, and the two schemes deliver identical Fisher information per absorbed photon, $[T(1-T{)}^2{]}^{-1}$: for measuring how transparent something is, the interferometer buys nothing. The advantage lies in discrimination, and we show that what it requires is not that the object be opaque but that the competing hypothesis be the object’s absence. Against empty space the Chernoff information per absorbed photon grows as $(8/{\pi }^2)[(1-\sqrt{T})/(1+\sqrt{T})]NlnN$ in the number of Zeno cycles $N$, even for a weakly absorbing object, while between two partial transparencies it does not grow at all. Parasitic per-cycle loss $ϵ$ caps the advantage: the conclusive interrogations per absorbed photon peak at $(0.2625/ϵ)(1-\sqrt{T})/(1+\sqrt{T})$, attained at $N_{opt}=1.5936/ϵ$ for an opaque object, which identifies the loss per cycle as the figure of merit governing how far interaction-free sensing can be pushed. Finally, the negative result is not special to the interferometer: for a single photon meeting a memoryless, non-dispersive object any number of times through arbitrary fixed optics, the accessible quantum Fisher information never exceeds that of the same incident flux spent on independent single-pass probes, and is generically far below it. This recovers the bound of Massar, Mitchison and Pironio for this class, by a short argument that also identifies when it is tight.