The way to determine the constant $A$ is as follows. Let $D$ denote a disk of radius $R$, centred at the origin, and let $\partial D$ denote the circular boundary of this disk. By the two-dimensional divergence theorem,
$$ \int_{\partial D} (\nabla G ). \hat n dl = \int_{D} (\nabla^2 G) \ d^2S,$$
where $\hat n$ denotes the unit normal at each point of the circular boundary.
Now $\nabla G = (A/r )\hat{\vec r}$ (where $\hat{\vec r}$ denotes the radial unit vector), and $\hat n = \hat r$ at each point of the circular boundary, so $(\nabla G) . \hat n = (A / R) (\hat {\vec r} . \hat {\vec r}) = A / R$ at every point of the circular boundary. Therefore,
$$ \int_{\partial D} (\nabla G ). \hat n dl = \frac A R \int_{\partial D} dl = \frac A R \times (2\pi R) = 2\pi A.$$
Meanwhile,
$$ \int_{D} (\nabla^2 G )\ d^2S = \int_D (-\delta^2 (\vec r)) \ d^2 S= - 1 ,$$
and equating the two results gives $A = -1/2\pi.$
I'm not sure how to interpret the hint about "using the geometry of the problem". I suppose the fact that we chose to consider integrals on a circular disk was motivated by the geometry. But you could also argue that the geometry motivated you to seek a solution independent of $\phi$ in the first place...