diff --git a/.github/workflows/handle_potential_conflicts.py b/.github/workflows/handle_potential_conflicts.py
index f2c460d00bd3..c564231e8ffd 100755
--- a/.github/workflows/handle_potential_conflicts.py
+++ b/.github/workflows/handle_potential_conflicts.py
@@ -154,7 +154,10 @@ def normalize_state(state: Any) -> dict[str, Any]:
 
 
 def extract_state(body: str) -> dict[str, Any]:
-    marker_index = body.find(COMMENT_START)
+    # The state block is rendered last, but comments written before that change
+    # start with it. Anchor on the last marker unless the body opens with one,
+    # so marker-like text in advisory content cannot shadow the real block.
+    marker_index = 0 if body.startswith(COMMENT_START) else body.rfind(COMMENT_START)
     if marker_index == -1:
         return normalize_state({})
 
@@ -218,9 +221,6 @@ def render_comment_body(state: dict[str, Any]) -> str:
     state_json = json.dumps(state, sort_keys=True, separators=(",", ":"))
     state_encoded = base64.b64encode(state_json.encode("utf8")).decode("ascii")
     lines = [
-        COMMENT_START,
-        state_encoded,
-        COMMENT_END,
         "## Potential PR merge conflicts",
         "",
         "This is advisory only. It does not block CI, but it marks PRs that will likely need a rebase depending on merge order.",
@@ -247,6 +247,14 @@ def render_comment_body(state: dict[str, Any]) -> str:
         inbound_items = sorted(state["inbound"].values(), key=lambda item: int(item["number"]))
         lines.extend(format_pr_line(item) for item in inbound_items)
 
+    # The state block goes last so notification previews start with the advisory text.
+    lines.extend([
+        "",
+        COMMENT_START,
+        state_encoded,
+        COMMENT_END,
+    ])
+
     return "\n".join(lines).rstrip() + "\n"
 
 
diff --git a/.github/workflows/test_handle_potential_conflicts.py b/.github/workflows/test_handle_potential_conflicts.py
index 29f12ff7f7ea..55898d7e9666 100755
--- a/.github/workflows/test_handle_potential_conflicts.py
+++ b/.github/workflows/test_handle_potential_conflicts.py
@@ -91,7 +91,7 @@ def test_rendered_comment_state_round_trips(self):
             "outbound": [
                 {
                     "number": 12,
-                    "title": "title with --> marker-like text",
+                    "title": f"title with {handle_potential_conflicts.COMMENT_START} --> marker-like text",
                     "url": "https://github.com/dashpay/dash/pull/12",
                     "files": ["src/net.cpp"],
                 }
@@ -103,7 +103,22 @@ def test_rendered_comment_state_round_trips(self):
         extracted = handle_potential_conflicts.extract_state(body)
 
         self.assertEqual(state, extracted)
-        self.assertNotIn("title with --> marker-like text", body.split("-->")[0])
+        self.assertTrue(body.startswith("## Potential PR merge conflicts"))
+        self.assertTrue(body.rstrip("\n").endswith(handle_potential_conflicts.COMMENT_END))
+
+    def test_extract_state_from_legacy_marker_first_comment(self):
+        state = {
+            "outbound": [
+                {"number": 2, "title": "two", "url": "https://example.test/2", "files": []},
+            ],
+            "inbound": {},
+        }
+        body = handle_potential_conflicts.render_comment_body(state)
+        marker_index = body.rfind(handle_potential_conflicts.COMMENT_START)
+        legacy_body = body[marker_index:] + body[:marker_index]
+
+        self.assertTrue(legacy_body.startswith(handle_potential_conflicts.COMMENT_START))
+        self.assertEqual(state, handle_potential_conflicts.extract_state(legacy_body))
 
     def test_rendered_comment_escapes_markdown_content(self):
         body = handle_potential_conflicts.render_comment_body({