diff --git a/doc/release/upcoming_changes/20924.compatibility.rst b/doc/release/upcoming_changes/20924.compatibility.rst
index 0446b714fe9..02f3e9fdbac 100644
--- a/doc/release/upcoming_changes/20924.compatibility.rst
+++ b/doc/release/upcoming_changes/20924.compatibility.rst
@@ -8,4 +8,6 @@ now due to the fact that the logic is aligned with item assignment::
     # is now identical to:
     arr[0] = scalar
 
-Previously casting may have produced slightly different answers when using values that could not be represented in the target `dtype` or when using `object`s.
+Previously casting may have produced slightly different answers when using
+values that could not be represented in the target ``dtype`` or when the
+target had ``object`` dtype.
diff --git a/numpy/core/_add_newdocs.py b/numpy/core/_add_newdocs.py
index fb9c30d9308..8ee48d24645 100644
--- a/numpy/core/_add_newdocs.py
+++ b/numpy/core/_add_newdocs.py
@@ -3437,6 +3437,24 @@
     >>> a
     array([1.,  1.])
 
+    Fill expects a scalar value and always behaves the same as assigning
+    to a single array element.  The following is a rare example where this
+    distinction is important:
+
+    >>> a = np.array([None, None], dtype=object)
+    >>> a[0] = np.array(3)
+    >>> a
+    array([array(3), None], dtype=object)
+    >>> a.fill(np.array(3))
+    >>> a
+    array([array(3), array(3)], dtype=object)
+
+    Where other forms of assignments will array being assigned:
+
+    >>> a[...] = np.array(3)
+    >>> a
+    array([3, 3], dtype=object)
+
     """))