diff --git a/sphinx/transforms/i18n.py b/sphinx/transforms/i18n.py
index 570154185e9..b87894dbb7d 100644
--- a/sphinx/transforms/i18n.py
+++ b/sphinx/transforms/i18n.py
@@ -135,10 +135,9 @@ def compare_references(
         warning_msg: str,
     ) -> None:
         """Warn about mismatches between references in original and translated content."""
-        # FIXME: could use a smarter strategy than len(old_refs) == len(new_refs)
-        if not self.noqa and len(old_refs) != len(new_refs):
-            old_ref_rawsources = [ref.rawsource for ref in old_refs]
-            new_ref_rawsources = [ref.rawsource for ref in new_refs]
+        old_ref_rawsources = [ref.rawsource for ref in old_refs]
+        new_ref_rawsources = [ref.rawsource for ref in new_refs]
+        if not self.noqa and old_ref_rawsources != new_ref_rawsources:
             logger.warning(
                 warning_msg.format(old_ref_rawsources, new_ref_rawsources),
                 location=self.node,
diff --git a/tests/test_intl/test_intl.py b/tests/test_intl/test_intl.py
index 2375c1a6687..e6a2737746f 100644
--- a/tests/test_intl/test_intl.py
+++ b/tests/test_intl/test_intl.py
@@ -192,6 +192,12 @@ def test_text_inconsistency_warnings(app):
             'translated': '\\[\\]',
         }
         + warning_fmt
+        % {
+            'reftype': 'citation references',
+            'original': "\\['\\[ref2\\]_'\\]",
+            'translated': "\\['\\[ref3\\]_'\\]",
+        }
+        + warning_fmt
         % {
             'reftype': 'references',
             'original': "\\['reference_'\\]",