How does Riemann hypothesis implies estimates? In Iwaniec, Luo and Sarnak article (precisely (4.23)), it is said that GRH for $L(s, \mathrm{sym}^2(f))$, for a holomorphic cusp newform $f$ of level $N$ and weight $k$, implies
$$\sum_{p \nmid N} \frac{\lambda_f(p^2) \log p}{p} \ll \log\log kN$$
Why is that true?
It is not the first time that I use this "blackbox" of GRH giving bounds, is there a general intuition to have behind it, or a systematic formal justification (like using the explicit formula to relate zeroes and eigenvalues)?
 A: We have that
\[-\frac{L'}{L}(s,\mathrm{sym}^2 f) = \sum_{n = 1}^{\infty} \frac{\Lambda_{\mathrm{sym}^2 f}(n)}{n^s},\]
where $\Lambda_{\mathrm{sym}^2 f}(n)$ is equal to $\lambda_f(p^2) \log p$ if $n = p$ with $p \nmid N$, is essentially a bounded multiple of $\log p$ if $n = p^k$, and vanishes otherwise. (There are some minor issues at $p \mid N$ that are no big deal). In particular, this shows that the value at $1$ of this sum is basically equal to the desired sum up to a negligible error term.
Now use Theorem 5.17 of Iwaniec and Kowalski, which states that for $s = \sigma + it$ with $1/2 < \sigma \leq 5/4$ and assuming RH for $L(s,\mathrm{sym}^2 f)$ (as well as the Ramanujan conjecture, which is known via the work of Deligne),
\[-\frac{L'}{L}(s,\mathrm{sym}^2 f) = \frac{r}{s - 1} + O\left(\frac{1}{2\sigma - 1} (\log \mathfrak{q}(\mathrm{sym}^2 f, s))^{2 - 2\sigma} + \log \log \mathfrak{q}(\mathrm{sym}^2 f, s)\right),\]
where $r$ is the order of the pole of $L(s,\mathrm{sym}^2 f)$ at $s = 1$ and $\mathfrak{q}(\mathrm{sym}^2 f, s)$ is the analytic conductor. Note that $N$ is squarefree and $f$ has trivial nebentypus, so that $r = 0$. Taking $s = \sigma = 1$ and noting that $\log \mathfrak{q}(\mathrm{sym}^2 f, s) \ll \log kN$ yields the result.
