You can use Laplce mecanism in the same way you would on the arithmetic mean.
And, as for the arithmetic mean, for it to be $\varepsilon$-differentially private, you need to suppose that your data is bounded in absolute value by a positive number $M$.
Let $m$ be a laplacian mechanism on the geometric mean query.
The pdf of the Laplace distribution is $f(x|\lambda) = \frac{1}{2\lambda}\exp^{\frac{-|x|}{\lambda}}$.
Then $(x_1, \dots, x_n) \mapsto m(x_1, \dots, x_n) =\prod{x_i}^{\frac{1}{n}} + \eta $ where $\eta \sim Laplace(\lambda = \frac{\varepsilon}{2M})$ is an $\varepsilon$-differentially private mechanism.
You can easily show it, by considering two databases $x_1, ..., x_n$ and $x_1, ..., x'_n$ with only one different datapoint (chosen to be the $n$-th one without loss of generality) :
$$
\begin{array}[ccc]
\;\frac{P(m(x_1, \dots, x_n) = t)}{P(m(x_1, \dots, x'_n) = t)} &= &\frac{\exp(\lambda\mid t - ( x_1\times\dots \times x_n)^{\frac{1}{n}}\mid)}{\exp(\lambda\mid t - (x_1\times\dots \times x'_n)^{\frac{1}{n}}\mid)}\\
& =& \exp(\lambda \mid t - ( x_1\times\dots \times x_n)^{\frac{1}{n}}\mid - \lambda\mid t - ( x_1\times\dots \times x'_n)^{\frac{1}{n}}\mid)\\
& \leq&\exp(\lambda \mid (x_1\times\dots \times x_n)^{\frac{1}{n}} - (x_1\times\dots \times x'_n)^{\frac{1}{n}} \mid)\\
& \leq & \exp(\lambda \times 2M)\\
& = &\exp(\varepsilon)
\end{array}
$$
But maybe this isn't the best way to have an $\varepsilon$-differentially private geometric mean. Indeed the mechanism above could output a negative value (!). A better way could be to compute the arithmetic mean of the logarithm of the data, make it $\varepsilon$-differentially private with Laplace mechanism, and then take the exponential of the result. This would make it impossible to have a negative output of the mechanism, and this would still be $\varepsilon$-differentially private (since this is preserved by deterministic transformation).
You cannot excape from the requirement of boundedness of data. In practice extreme data are trimmed to a given value (chosen before looking at the data!).
I hope this helped.