Wikipedia tends to answer your question by this (emphasis mine):
In F-ATPases, there are three copies each of the alpha and beta subunits that form the catalytic core of the F1 complex, while the remaining F1 subunits (gamma, delta, epsilon) form part of the stalks. There is a substrate-binding site on each of the alpha and beta subunits, those on the beta subunits being catalytic, while those on the alpha subunits are regulatory. The alpha and beta subunits form a cylinder that is attached to the central stalk. The alpha/beta subunits undergo a sequence of conformational changes leading to the formation of ATP from ADP, which are induced by the rotation of the gamma subunit, itself is driven by the movement of protons through the F0 complex C subunit.
For above information, wikipedia cites this. I have put some info from it below.
There are three regions to which ATP is hydrogen bound. First, the so-called Walker or P-loop (residues β-Gly-159-β-Val-164) at the beginning of helix B. Second, the beginning region of helix C, namely β-Arg-189. And third, the residue α-Arg-373 from the α-subunit (see Fig. 1, c–e) (Abrahams et al., 1994). Sequential formation of these 15–20 hydrogen bonds ensures nearly constant force generation throughout the whole duration of the binding transition. This “binding zipper” sequence would lead to the smooth closing motion of the pocket and continuous conformational changes throughout the β-subunit (Oster and Wang, 2000a; Elston et al., 1998; Oster and Wang, 2000b).
Fig. 4, a–d provides four representative snapshots of the equilibrated closed and open binding pockets according to our simulations. These show that, in the closed pocket, the ATP molecule is surrounded by all three hydrogen binding regions. By contrast, in the open pocket a space has formed between the P-loop on one side and the helix C and the pocket's α-subunit region on the other side, with a gap between the ATP molecule and the P-loop so that ATP stays close to the α-subunit and helix C regions. An equilibrated intermediate structure where the pocket is half open is shown in Fig. 4 b. In that state the α-phosphate oxygen of ATP is still close to the P-loop, but the distance is already increasing between the β/γ-phosphate oxygens and the P-loop. The phosphate axis has rotated ∼30° and ATP is bridging the pocket. At the end of our simulations (Fig. 4 d), the ATP molecule is located between the two subunits as expected after its primary movement into the binding pocket. In that weak binding state, a newly docked ATP is expected to have contacted the ATPase, but not yet have induced conformational changes. Our simulations are consistent with this expectation. Further, we see that contacts are formed mainly between ATP and Mg2+, and between Mg2+ and the binding pocket.
The importance of β-Lys-162, β-Arg-189, and α-Arg-373, for example, was connected to their formation of hydrogen bonds with ATP. In our simulations, these residues form the strongest and the last to break hydrogen bonds with the γ-phosphate oxygens. In addition, α-Arg-373 forms a hydrogen bond with one of the α-phosphate oxygens, acting as a restraining force for ATP during the movement of the P-loop and therefore facilitating the migration of the α-phosphate oxygens along the P-loop.
You can find further information here or here.