diff --git a/docs/examples/emulator/robustgasp/plot_scalargasp.py b/docs/examples/emulator/robustgasp/plot_scalargasp.py
index e4143bc1a667a01aa47d4038aec3252183843bd8..710b163905f8e3558e262c2969267a96c9c84051 100644
--- a/docs/examples/emulator/robustgasp/plot_scalargasp.py
+++ b/docs/examples/emulator/robustgasp/plot_scalargasp.py
@@ -104,4 +104,15 @@ plt.fill_between(testing_input, predictions[:, 1], predictions[:, 2], alpha=0.3,
 plt.xlabel('x')
 plt.ylabel('emulator-predicted y')
 plt.xlim(testing_input[0], testing_input[-1])
-_ = plt.legend()
\ No newline at end of file
+_ = plt.legend()
+
+
+# %% md
+#
+# Above example shows how to train a GP emulator based on noise-free training data,
+# which is often the case of emulating a deterministic simulator. If you are dealing
+# with noisy training data, you can
+#
+# - set the parameter ``nugget`` to a desired value, or
+# - set ``nugget`` to :math:`0` and ``nugget_est`` to `True`, meaning that ``nugget``
+#   is estimated from the noisy training data.
\ No newline at end of file