diff --git a/plugins/teams_pipeline/meetings.py b/plugins/teams_pipeline/meetings.py
index f0772c423d..994894ac64 100644
--- a/plugins/teams_pipeline/meetings.py
+++ b/plugins/teams_pipeline/meetings.py
@@ -2,6 +2,8 @@
 
 from __future__ import annotations
 
+import base64
+import binascii
 import re
 import tempfile
 from pathlib import Path
@@ -88,7 +90,32 @@ def looks_like_transcript_id(value: str, *, odata_type: str | None = None) -> bo
 
     if "calltranscript" in str(odata_type or "").lower():
         return True
-    return "transcript" in str(value or "").lower()
+    text = str(value or "")
+    if "transcript" in text.lower():
+        return True
+    return "transcript" in _decoded_id_hint(text)
+
+
+def _decoded_id_hint(value: str) -> str:
+    """Best-effort base64 decode of a Graph id for artifact-marker sniffing.
+
+    Graph transcript ids are base64url blobs whose *decoded* payload carries a
+    ``...-TranscriptV2`` suffix while the encoded form contains no readable
+    marker (this is exactly the id shape from getAllTranscripts
+    ``resourceData.id``). Returns lowercase decoded text, or "" when the value
+    does not decode.
+    """
+
+    stripped = value.strip()
+    if len(stripped) < 16:
+        return ""
+    padded = stripped + "=" * (-len(stripped) % 4)
+    for decoder in (base64.urlsafe_b64decode, base64.b64decode):
+        try:
+            return decoder(padded).decode("utf-8", "ignore").lower()
+        except (binascii.Error, ValueError):
+            continue
+    return ""
 
 
 def _meeting_path(meeting_ref: TeamsMeetingRef | str) -> str:
diff --git a/tests/plugins/test_teams_pipeline_plugin.py b/tests/plugins/test_teams_pipeline_plugin.py
index 6723bd66e9..42b298ccda 100644
--- a/tests/plugins/test_teams_pipeline_plugin.py
+++ b/tests/plugins/test_teams_pipeline_plugin.py
@@ -461,6 +461,24 @@ def test_looks_like_transcript_id_detects_graph_call_transcripts():
     assert not looks_like_transcript_id("MSo-meeting")
 
 
+def test_looks_like_transcript_id_detects_base64_encoded_marker():
+    # Real-world getAllTranscripts resourceData.id shape: base64url blob whose
+    # DECODED payload ends in "-TranscriptV2" while the encoded form has no
+    # readable marker (from a field report, Aug 2026).
+    import base64
+
+    encoded = base64.urlsafe_b64encode(
+        b"...meeting_ABC@thread.v2#55d6e479-2db7-4297-8590-98f6011a9613-1787004872-TranscriptV2"
+    ).decode().rstrip("=") + "="
+    assert "transcript" not in encoded.lower()
+    assert looks_like_transcript_id(encoded)
+    # Meeting ids (base64 of "0#...#0**19:meeting_...@thread.v2") must NOT match.
+    meeting_encoded = base64.b64encode(
+        b"0#af8b854e-e592-4d5a-b04c-6e63a3b502e7#0**19:meeting_ABC@thread.v2"
+    ).decode()
+    assert not looks_like_transcript_id(meeting_encoded)
+
+
 def test_create_job_from_get_all_transcripts_uses_meeting_id_not_transcript_id(tmp_path):
     store = TeamsPipelineStore(tmp_path / "teams-store.json")
     pipeline = TeamsMeetingPipeline(graph_client=FakeGraphClient(), store=store)