diff --git a/backend/go/qdrant-store/store.go b/backend/go/qdrant-store/store.go
index 8769fdd34..361bb8f48 100644
--- a/backend/go/qdrant-store/store.go
+++ b/backend/go/qdrant-store/store.go
@@ -17,6 +17,7 @@ import (
 	"github.com/mudler/LocalAI/pkg/grpc/base"
 	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
 	"github.com/mudler/LocalAI/pkg/store"
+	"github.com/mudler/xlog"
 	"github.com/qdrant/go-client/qdrant"
 	"google.golang.org/grpc"
 	"google.golang.org/grpc/codes"
@@ -84,6 +85,10 @@ func (s *QdrantStore) Load(opts *pb.ModelOptions) error {
 		return fmt.Errorf("qdrant-store: connect to %s: %w", cfg.Addr(), err)
 	}
 	s.client, s.cfg, s.collection = client, cfg, collectionName(cfg, opts.GetModel())
+	if cfg.Distance != qdrant.Distance_Cosine {
+		xlog.Warn("qdrant-store: non-COSINE distance_metric returns raw distances or dot products, not similarities; keep COSINE for router, face and voice stores",
+			"distance_metric", cfg.Distance.String(), "collection", s.collection)
+	}
 
 	ctx, cancel := s.ctx()
 	defer cancel()
diff --git a/docs/content/features/stores.md b/docs/content/features/stores.md
index a28b44a39..9350ae9f5 100644
--- a/docs/content/features/stores.md
+++ b/docs/content/features/stores.md
@@ -221,6 +221,14 @@ results come back nearest-first. If a collection already exists with a different
 `distance_metric`, loading the store (or, if the collection appeared later, the first `find`) fails
 rather than returning scores with the wrong meaning.
 
+{{% notice warning %}}
+Only `COSINE` returns similarity-like scores. With `DOT`, `EUCLID` or `MANHATTAN` the backend
+returns Qdrant's raw dot products or distances, which are not bounded to `[-1, 1]` and, for
+`EUCLID` and `MANHATTAN`, run in the opposite direction. Router stores, face recognition and voice
+recognition compare scores against cosine thresholds, so keep them on `COSINE`. The backend logs a
+warning at load time when a store is configured with another metric.
+{{% /notice %}}
+
 {{% notice note %}}
 The backend creates collections with Qdrant's default settings. For custom settings (sharding and
 replication on a cluster, HNSW parameters, on-disk vectors, quantization), create the collection