BUG: value_counts not preserving object dtype

doc/source/whatsnew/v2.1.4.rst

@@ -26,6 +26,7 @@ Bug fixes
 - Fixed bug in :meth:`DataFrame.__setitem__` casting :class:`Index` with object-dtype to PyArrow backed strings when ``infer_string`` option is set (:issue:`55638`)
 - Fixed bug in :meth:`Index.insert` casting object-dtype to PyArrow backed strings when ``infer_string`` option is set (:issue:`55638`)
 - Fixed bug in :meth:`Series.str.translate` losing object dtype when string option is set (:issue:`56152`)
+- Fixed bug in :meth:`Series.value_counts` not preserving object dtype when ``infer_string`` is set (:issue:`56187`)
 
 .. ---------------------------------------------------------------------------
 .. _whatsnew_214.other:

pandas/core/algorithms.py

@@ -871,6 +871,8 @@ def value_counts_internal(
         Series,
     )
 
+    input_dtype = None if not isinstance(values, Series) else values.dtype
+
     index_name = getattr(values, "name", None)
     name = "proportion" if normalize else "count"
 
@@ -929,7 +931,7 @@ def value_counts_internal(
 
             # For backwards compatibility, we let Index do its normal type
             #  inference, _except_ for if if infers from object to bool.
-            idx = Index(keys)
+            idx = Index(keys, dtype=input_dtype)
             if idx.dtype == bool and keys.dtype == object:
                 idx = idx.astype(object)
             elif idx.dtype != keys.dtype:

pandas/tests/series/methods/test_value_counts.py

@@ -269,3 +269,14 @@ def test_value_counts_masked(self):
             [2, 1, 1], index=Index([2, 1, 3], dtype=dtype), dtype=dtype, name="count"
         )
         tm.assert_series_equal(result, expected)
+
+    def test_value_counts_infer_string(self):
+        # GH#56187
+        pytest.importorskip("pyarrow")
+
+        ser = Series(["a", "b"], dtype=object)
+
+        with pd.option_context("future.infer_string", True):
+            result = ser.value_counts()
+        expected = Series([1, 1], index=Index(["a", "b"], dtype=object), name="count")
+        tm.assert_series_equal(result, expected)