diff --git a/src/harmony/parsing/text_parser.py b/src/harmony/parsing/text_parser.py
index 5e7663c..ef71824 100644
--- a/src/harmony/parsing/text_parser.py
+++ b/src/harmony/parsing/text_parser.py
@@ -75,6 +75,7 @@ def convert_text_to_instruments(file: RawFile) -> List[Instrument]:
         elif "," in first_line:
             csv_sep = ","
 
+    if file.file_type == FileType.csv and csv_sep is not None:
         string_io = StringIO(page_text)
         df = pd.read_csv(string_io, sep=csv_sep)
         df.fillna("", inplace=True)
diff --git a/tests/test_convert_text.py b/tests/test_convert_text.py
index 02776ef..160709f 100644
--- a/tests/test_convert_text.py
+++ b/tests/test_convert_text.py
@@ -135,6 +135,23 @@ def test_remove_both_ends_digits_from_csv(self):
         self.assertEqual("How are you today", questions[0].question_text)
         self.assertEqual("Are you feeling better", questions[1].question_text)
 
+    def test_csv_without_detected_separator_falls_back_to_line_parsing(self):
+        # Regression: pandas >=2.3 dropped support for pd.read_csv(sep=None).
+        # A CSV whose first line contains no tab or comma must parse via the
+        # txt branch (one question per line), not via pd.read_csv(sep=None).
+        no_sep_csv = RawFile.model_validate({
+            "file_id": "no_sep_csv",
+            "file_name": "no_separator.csv",
+            "file_type": "csv",
+            "content": "I feel anxious\nI feel restless"
+        })
+        instruments = convert_text_to_instruments(no_sep_csv)
+        self.assertEqual(1, len(instruments))
+        questions = instruments[0].questions
+        self.assertEqual(2, len(questions))
+        self.assertEqual("I feel anxious", questions[0].question_text)
+        self.assertEqual("I feel restless", questions[1].question_text)
+
 
 if __name__ == '__main__':
     unittest.main()