diff --git a/backend/migrations/20260528143710_draft_user_sync_schema.down.sql b/backend/migrations/20260528143710_draft_user_sync_schema.down.sql
index b458f05d31..2ba104a1d0 100644
--- a/backend/migrations/20260528143710_draft_user_sync_schema.down.sql
+++ b/backend/migrations/20260528143710_draft_user_sync_schema.down.sql
@@ -1,5 +1,6 @@
 ALTER TABLE draft DROP COLUMN id;
 
+DROP INDEX IF EXISTS draft_user_sync_idx;
 DROP INDEX IF EXISTS draft_pkey_legacy;
 DROP INDEX IF EXISTS draft_pkey_with_user;
 
diff --git a/backend/migrations/20260528143710_draft_user_sync_schema.up.sql b/backend/migrations/20260528143710_draft_user_sync_schema.up.sql
index ed0b389c16..96c5c6e4af 100644
--- a/backend/migrations/20260528143710_draft_user_sync_schema.up.sql
+++ b/backend/migrations/20260528143710_draft_user_sync_schema.up.sql
@@ -61,4 +61,13 @@ CREATE UNIQUE INDEX draft_pkey_legacy
     ON draft (workspace_id, path, typ)
     WHERE email IS NULL;
 
+-- Hot path for `sync_drafts`: each active editor polls
+-- `WHERE workspace_id = ? AND email = ? [AND created_at > ?]` every 2-10s.
+-- Neither partial unique index above helps (both lead with `path, typ`);
+-- this covers the initial sync (leading two columns) and the incremental
+-- missed-drafts query (range scan on `created_at`).
+CREATE INDEX draft_user_sync_idx
+    ON draft (workspace_id, email, created_at)
+    WHERE email IS NOT NULL;
+
 ALTER TABLE draft ADD COLUMN id BIGSERIAL PRIMARY KEY;