diff --git a/dask_expr/_reductions.py b/dask_expr/_reductions.py
index 357b4849..ce4a0de7 100644
--- a/dask_expr/_reductions.py
+++ b/dask_expr/_reductions.py
@@ -244,7 +244,8 @@ def _lower(self):
 
         # Reset the index if we we used it for shuffling
         if split_by_index:
-            shuffled = SetIndexBlockwise(shuffled, split_by, True, None)
+            idx = list(self._meta.index.names) if split_by != ["index"] else split_by
+            shuffled = SetIndexBlockwise(shuffled, idx, True, None)
 
         # Convert back to Series if necessary
         if self.shuffle_by_index is not False:
diff --git a/dask_expr/tests/test_groupby.py b/dask_expr/tests/test_groupby.py
index e5b45e6d..05ca7f3c 100644
--- a/dask_expr/tests/test_groupby.py
+++ b/dask_expr/tests/test_groupby.py
@@ -157,6 +157,13 @@ def test_groupby_no_numeric_only(pdf, func):
     assert_eq(agg, expect)
 
 
+def test_value_counts_split_out(pdf):
+    df = from_pandas(pdf, npartitions=10)
+    result = df.groupby("x").y.value_counts(split_out=True)
+    expected = pdf.groupby("x").y.value_counts()
+    assert_eq(result, expected)
+
+
 def test_unique(df, pdf):
     result = df.groupby("x")["y"].unique()
     expected = pdf.groupby("x")["y"].unique()