diff --git a/services/extractor/main.py b/services/extractor/main.py
index a7e25c7..1ebe0b0 100644
--- a/services/extractor/main.py
+++ b/services/extractor/main.py
@@ -61,6 +61,25 @@ async def main() -> None:
             ticker = job.get("ticker", "")
             text = job.get("text", "") or job.get("normalized_text", "")
 
+            # If no text in job, try to fetch from MinIO via the document's normalized_storage_ref
+            if not text:
+                ref_row = await pool.fetchrow(
+                    "SELECT normalized_storage_ref FROM documents WHERE id = $1::uuid",
+                    document_id,
+                )
+                if ref_row and ref_row["normalized_storage_ref"]:
+                    try:
+                        ref = ref_row["normalized_storage_ref"]
+                        # ref format: s3://bucket/path
+                        parts = ref.replace("s3://", "").split("/", 1)
+                        if len(parts) == 2:
+                            obj = minio_client.get_object(parts[0], parts[1])
+                            text = obj.read().decode("utf-8")
+                            obj.close()
+                            obj.release_conn()
+                    except Exception as e:
+                        logger.warning("Could not fetch normalized text for doc %s: %s", document_id, e)
+
             logger.info("Processing extraction job for doc %s / %s", document_id, ticker)
 
             # Refresh company map every 100 jobs