diff --git a/internal/handler/search.go b/internal/handler/search.go
index 8fdbeee..a944167 100644
--- a/internal/handler/search.go
+++ b/internal/handler/search.go
@@ -70,6 +70,20 @@ var collectionQueryByWeights = map[string]string{
 	"other_documents": "10,10,2,2",
 }
 
+// collectionTextMatchType lets specific collections opt-in to `sum_score`
+// instead of the Typesense default `max_score`.
+//
+// `sum_score` boosts documents that match in MULTIPLE fields, so e.g. a letter
+// whose `recipient_display` AND `content` both contain the query ranks above
+// a letter that only matches in `content`. Useful for collections that have a
+// strong metadata field (recipient, title) tied to the user's intent.
+//
+// `max_score` (default elsewhere) preserves single-field dominance — important
+// for verses where the canonical verse should beat commentaries that quote it.
+var collectionTextMatchType = map[string]string{
+	"letters": "sum_score",
+}
+
 // searchResultItem represents a single search result in the API response.
 type searchResultItem struct {
 	Collection string                 `json:"collection"`
@@ -206,6 +220,9 @@ func (h *SearchHandler) multiSearch(ctx context.Context, collections []string, q
 		if weights != "" {
 			params.QueryByWeights = stringPtr(weights)
 		}
+		if tmt := collectionTextMatchType[col]; tmt != "" {
+			params.TextMatchType = stringPtr(tmt)
+		}
 		searches = append(searches, params)
 	}
 
@@ -278,6 +295,9 @@ func (h *SearchHandler) singleSearch(ctx context.Context, collection, query stri
 	if weights := collectionQueryByWeights[collection]; weights != "" {
 		params.QueryByWeights = stringPtr(weights)
 	}
+	if tmt := collectionTextMatchType[collection]; tmt != "" {
+		params.TextMatchType = stringPtr(tmt)
+	}
 
 	sr, err := h.ts.Client.Collection(collection).Documents().Search(ctx, params)
 	if err != nil {