diff --git a/reasoning_gym/algorithmic/word_sorting.py b/reasoning_gym/algorithmic/word_sorting.py
index e0b40da3..51f96140 100644
--- a/reasoning_gym/algorithmic/word_sorting.py
+++ b/reasoning_gym/algorithmic/word_sorting.py
@@ -95,7 +95,7 @@ class WordSortingDataset(ProceduralDataset):
 
         return {
             "question": f"Sort these words in {direction} order: {', '.join(transformed_words)}",
-            "answer": str(answer),
+            "answer": ", ".join(answer),
             "metadata": {
                 "original_words": original_words,
                 "transformed_words": transformed_words,
diff --git a/tests/test_word_sorting.py b/tests/test_word_sorting.py
index bb673883..14f44204 100644
--- a/tests/test_word_sorting.py
+++ b/tests/test_word_sorting.py
@@ -94,7 +94,7 @@ def test_word_sorting_dataset_items():
 
         # Verify sorting
         direction = item["metadata"]["direction"]
-        sorted_words = eval(item["answer"])
+        sorted_words = item["answer"].split(", ")
         if direction == "ascending":
             assert sorted_words == sorted(sorted_words)
         else: