diff --git a/packages/markitdown/src/markitdown/converters/_bing_serp_converter.py b/packages/markitdown/src/markitdown/converters/_bing_serp_converter.py
index fd00a70ab..c58e35949 100644
--- a/packages/markitdown/src/markitdown/converters/_bing_serp_converter.py
+++ b/packages/markitdown/src/markitdown/converters/_bing_serp_converter.py
@@ -1,7 +1,7 @@
import re
import base64
import binascii
-from urllib.parse import parse_qs, urlparse
+from urllib.parse import ParseResult, parse_qs, urlparse
from typing import Any, BinaryIO
from bs4 import BeautifulSoup
@@ -20,6 +20,18 @@
]
+def _is_bing_redirect(parsed_href: ParseResult) -> bool:
+ """Return True only for Bing's own `/ck/a` redirect links.
+
+ A normal result may use `u=` for its own purpose. So the base64 payload is
+ decoded only when the link is one that Bing wrapped.
+ """
+ host = (parsed_href.netloc or "").lower()
+ if host != "bing.com" and not host.endswith(".bing.com"):
+ return False
+ return parsed_href.path == "/ck/a"
+
+
class BingSerpConverter(DocumentConverter):
"""
Handle Bing results pages (only the organic search results).
@@ -87,6 +99,12 @@ def convert(
# Rewrite redirect urls
for a in result.find_all("a", href=True):
parsed_href = urlparse(a["href"])
+
+ # Keep normal result URLs as they are. Only Bing's redirect
+ # links carry the base64 `u` payload.
+ if not _is_bing_redirect(parsed_href):
+ continue
+
qs = parse_qs(parsed_href.query)
# The destination is contained in the u parameter,
diff --git a/packages/markitdown/tests/test_bing_serp_converter.py b/packages/markitdown/tests/test_bing_serp_converter.py
new file mode 100644
index 000000000..9d80589bc
--- /dev/null
+++ b/packages/markitdown/tests/test_bing_serp_converter.py
@@ -0,0 +1,89 @@
+#!/usr/bin/env python3 -m pytest
+"""A result URL that uses `u=` for its own purpose must survive conversion.
+
+Bing puts each search result behind a `/ck/a` redirect link. The real
+destination sits in a base64 `u` query parameter. The converter decodes that
+parameter, but it did so for every link that has `u=`. A normal result that
+uses `u=` for its own purpose was therefore rewritten into garbage.
+"""
+
+import io
+
+from markitdown import MarkItDown, StreamInfo
+
+SERP_URL = "https://www.bing.com/search?q=python"
+
+REDIRECT_TARGET = "https://docs.python.org/3/"
+REDIRECT_HREF = (
+ "https://www.bing.com/ck/a?!&&p=xyz"
+ "&u=a1aHR0cHM6Ly9kb2NzLnB5dGhvbi5vcmcvMy8="
+)
+
+NORMAL_HREF = "https://example.com/profile?u=abcdef"
+
+LOOKALIKE_HREF = "https://not-bing.com/ck/a?u=a1aHR0cHM6Ly9kb2NzLnB5dGhvbi5vcmcvMy8="
+
+LOOKALIKE_PATH_HREF = (
+ "https://www.bing.com/ck/abc?u=a1aHR0cHM6Ly9kb2NzLnB5dGhvbi5vcmcvMy8="
+)
+
+
+def _serp(*hrefs: str) -> io.BytesIO:
+ items = "".join(
+ 'Snippet %d.
'
+ % (href, i, i)
+ for i, href in enumerate(hrefs, start=1)
+ )
+ html = (
+ "python - Search"
+ '%s
' % items
+ )
+ return io.BytesIO(html.encode("utf-8"))
+
+
+def _convert(*hrefs: str) -> str:
+ return (
+ MarkItDown()
+ .convert_stream(
+ _serp(*hrefs),
+ stream_info=StreamInfo(
+ url=SERP_URL,
+ mimetype="text/html",
+ extension=".html",
+ charset="utf-8",
+ ),
+ )
+ .markdown
+ )
+
+
+def test_bing_redirect_is_decoded_to_its_destination() -> None:
+ markdown = _convert(REDIRECT_HREF)
+
+ assert "](%s)" % REDIRECT_TARGET in markdown
+ assert "ck/a" not in markdown
+
+
+def test_normal_url_keeps_its_own_u_parameter() -> None:
+ markdown = _convert(NORMAL_HREF)
+
+ assert "](%s)" % NORMAL_HREF in markdown
+
+
+def test_normal_url_survives_next_to_a_redirect() -> None:
+ markdown = _convert(REDIRECT_HREF, NORMAL_HREF)
+
+ assert "](%s)" % REDIRECT_TARGET in markdown
+ assert "](%s)" % NORMAL_HREF in markdown
+
+
+def test_lookalike_host_is_not_treated_as_a_bing_redirect() -> None:
+ markdown = _convert(LOOKALIKE_HREF)
+
+ assert "](%s)" % LOOKALIKE_HREF in markdown
+
+
+def test_lookalike_path_is_not_treated_as_a_bing_redirect() -> None:
+ markdown = _convert(LOOKALIKE_PATH_HREF)
+
+ assert "](%s)" % LOOKALIKE_PATH_HREF in markdown