diff --git a/haystack/components/converters/csv.py b/haystack/components/converters/csv.py
index 65e59242ed..b9de23a04c 100644
--- a/haystack/components/converters/csv.py
+++ b/haystack/components/converters/csv.py
@@ -77,6 +77,8 @@ def __init__(
         self.quotechar = quotechar
 
         # Basic validation
+        if conversion_mode not in ("file", "row"):
+            raise ValueError(f"CSVToDocument: unsupported conversion_mode {conversion_mode!r}. Choose 'file' or 'row'.")
         if len(self.delimiter) != 1:
             raise ValueError("CSVToDocument: delimiter must be a single character.")
         if len(self.quotechar) != 1:
diff --git a/releasenotes/notes/csv-conversion-mode-validation-8f4c2b1d7e9a3056.yaml b/releasenotes/notes/csv-conversion-mode-validation-8f4c2b1d7e9a3056.yaml
new file mode 100644
index 0000000000..c5bade60d6
--- /dev/null
+++ b/releasenotes/notes/csv-conversion-mode-validation-8f4c2b1d7e9a3056.yaml
@@ -0,0 +1,9 @@
+---
+fixes:
+  - |
+    ``CSVToDocument`` now rejects an unsupported ``conversion_mode`` in its constructor.
+    Previously the value was only compared against ``"file"`` in ``run()``, so any other
+    string, including a typo such as ``"Rows"`` or a value with a trailing space, silently
+    selected the row mode and produced one ``Document`` per row instead of one per file.
+    When ``content_column`` was not passed, the raised error also reported
+    ``conversion_mode='row'`` for a component the caller had never configured that way.
diff --git a/test/components/converters/test_csv_to_document.py b/test/components/converters/test_csv_to_document.py
index 9717c6265d..663674c6f9 100644
--- a/test/components/converters/test_csv_to_document.py
+++ b/test/components/converters/test_csv_to_document.py
@@ -201,6 +201,20 @@ def test_init_validates_delimiter_and_quotechar(self):
         with pytest.raises(ValueError):
             CSVToDocument(quotechar='""')
 
+    def test_init_rejects_an_unknown_conversion_mode(self):
+        with pytest.raises(ValueError, match="unsupported conversion_mode 'Rows'"):
+            CSVToDocument(conversion_mode="Rows")  # type: ignore[arg-type]
+
+    def test_init_rejects_a_conversion_mode_that_only_differs_in_whitespace_or_case(self):
+        with pytest.raises(ValueError, match="unsupported conversion_mode 'file '"):
+            CSVToDocument(conversion_mode="file ")  # type: ignore[arg-type]
+        with pytest.raises(ValueError, match="unsupported conversion_mode ''"):
+            CSVToDocument(conversion_mode="")  # type: ignore[arg-type]
+
+    def test_init_accepts_both_documented_conversion_modes(self):
+        assert CSVToDocument(conversion_mode="file").conversion_mode == "file"
+        assert CSVToDocument(conversion_mode="row").conversion_mode == "row"
+
     def test_row_mode_large_file_warns(self, caplog, monkeypatch):
         # Make the threshold tiny so the warning always triggers.
         import haystack.components.converters.csv as csv_mod