diff --git a/src/sentry/options/defaults.py b/src/sentry/options/defaults.py
index c3a64a33142fe4..bb8d8c2a4621e6 100644
--- a/src/sentry/options/defaults.py
+++ b/src/sentry/options/defaults.py
@@ -897,6 +897,11 @@
     default={7001: 0.15},
     flags=FLAG_ALLOW_EMPTY | FLAG_AUTOMATOR_MODIFIABLE,
 )
+register(
+    "snuba.search.recommended.message-penalty-weight",
+    default=0.0,
+    flags=FLAG_AUTOMATOR_MODIFIABLE,
+)
 
 # The percentage of tagkeys that we want to cache. Set to 1.0 in order to cache everything, <=0.0 to stop caching
 register(
diff --git a/src/sentry/search/snuba/executors.py b/src/sentry/search/snuba/executors.py
index 38865a082d7a12..1a12bdcb203822 100644
--- a/src/sentry/search/snuba/executors.py
+++ b/src/sentry/search/snuba/executors.py
@@ -805,6 +805,11 @@ def _recommended_aggregation(
     # Group type boost: additive signal per issue type
     group_type_boosts = options.get("snuba.search.recommended.group-type-boost")
 
+    # Message penalty: downranks capture_message issues (no exception/stacktrace).
+    # Subtracted from the score below, and only on the events dataset -- issue-platform
+    # occurrences don't have exception_stacks.
+    message_penalty_weight = options.get("snuba.search.recommended.message-penalty-weight")
+
     # Skip zero-weighted factors: their term is always 0, so computing them in
     # ClickHouse is wasted work -- especially expensive aggregates like user
     # impact's uniq(tags[sentry:user]).
@@ -838,6 +843,11 @@ def _recommended_aggregation(
     for term in terms[1:]:
         score_expr = f"plus({score_expr}, {term})"
 
+    if type_column is None and message_penalty_weight:
+        has_exception_ratio = "divide(countIf(notEmpty(exception_stacks.type)), count())"
+        message_penalty = f"multiply({message_penalty_weight}, minus(1.0, {has_exception_ratio}))"
+        score_expr = f"minus({score_expr}, {message_penalty})"
+
     return [score_expr, ""]
 
 
diff --git a/tests/snuba/search/test_backend.py b/tests/snuba/search/test_backend.py
index b3b5026fc8b603..61ca50da34474f 100644
--- a/tests/snuba/search/test_backend.py
+++ b/tests/snuba/search/test_backend.py
@@ -4164,6 +4164,43 @@ def _recommended_scores(self, group_ids: list[int]) -> dict[int, float]:
         )[0]
         return {gid: score for gid, score in results}
 
+    def test_recommended_message_penalty(self) -> None:
+        ts = before_now(hours=1).isoformat()
+        exception_event = self.store_event(
+            data={
+                "fingerprint": ["exception-group"],
+                "event_id": "a" * 32,
+                "timestamp": ts,
+                "message": "exception-group",
+                "level": "error",
+                "exception": {
+                    "values": [
+                        {
+                            "type": "ValueError",
+                            "value": "something broke",
+                            "stacktrace": {"frames": [{"module": "app.main"}]},
+                        }
+                    ]
+                },
+            },
+            project_id=self.project.id,
+        )
+        message_event = self.store_event(
+            data={
+                "fingerprint": ["message-group"],
+                "event_id": "b" * 32,
+                "timestamp": ts,
+                "message": "message-group",
+                "level": "error",
+            },
+            project_id=self.project.id,
+        )
+
+        with self.options({"snuba.search.recommended.message-penalty-weight": 0.10}):
+            scores = self._recommended_scores([exception_event.group.id, message_event.group.id])
+
+        assert scores[exception_event.group.id] > scores[message_event.group.id]
+
     def test_recommended_zero_weight_factor_excluded(self) -> None:
         ts = before_now(hours=1).isoformat()
         for i in range(5):