diff --git a/stdlib/tokenize.pyi b/stdlib/tokenize.pyi
index 5f022844cea8..2655f2f0266a 100644
--- a/stdlib/tokenize.pyi
+++ b/stdlib/tokenize.pyi
@@ -130,9 +130,8 @@ class Untokenizer:
     if sys.version_info >= (3, 12):
         def escape_brackets(self, token: str) -> str: ...
 
-# the docstring says "returns bytes" but is incorrect --
-# if the ENCODING token is missing, it skips the encode
-def untokenize(iterable: Iterable[_Token]) -> str | Any: ...  # str in most use cases
+# Returns str, unless the ENCODING token is present, in which case it returns bytes.
+def untokenize(iterable: Iterable[_Token]) -> str | Any: ...
 def detect_encoding(readline: Callable[[], bytes | bytearray]) -> tuple[str, Sequence[bytes]]: ...
 def tokenize(readline: Callable[[], bytes | bytearray]) -> Generator[TokenInfo, None, None]: ...
 def generate_tokens(readline: Callable[[], str]) -> Generator[TokenInfo, None, None]: ...