diff --git a/dnscms/dnscms/wagtail_hooks.py b/dnscms/dnscms/wagtail_hooks.py
index bef3e4e..e7d3262 100644
--- a/dnscms/dnscms/wagtail_hooks.py
+++ b/dnscms/dnscms/wagtail_hooks.py
@@ -1,10 +1,7 @@
-from django.apps import apps as django_apps
 from django.templatetags.static import static
 from django.utils.html import format_html
 from grapple.registry import registry as grapple_registry
 from wagtail import hooks
-from wagtail.documents import get_document_model
-from wagtail.images import get_image_model
 from wagtail.models import Page
 from wagtail.search.backends import get_search_backend
 
@@ -15,33 +12,37 @@ def enable_additional_rich_text_features(features):
 
 
 @hooks.register("register_schema_query")
-def filter_search_to_live_pages(query_mixins):
+def override_search_resolver(query_mixins):
     """
-    Grapple's default `search` resolver hits every page regardless of publish
-    state, exposing drafts on the public API. Prepend a mixin so MRO picks our
-    `resolve_search`, which restricts Page subclasses to live + public.
+    Override Grapple's `search` resolver. Two fixes vs. the upstream version:
+    1. Restrict pages to live + public so drafts and access-restricted pages
+       don't leak via the public API.
+    2. Run a single search across all `Page` subclasses (instead of iterating
+       per-model) so results are ranked by relevance across types rather than
+       grouped by content type. Specific instances are fetched in a second
+       bulk query and reordered to match the search ranking.
+
+    Documents and images are intentionally not searched. The upstream resolver
+    includes them, but the frontend search page only renders Page types and
+    discards everything else, so iterating those indexes is wasted work.
     """
     if not grapple_registry.class_models:
         return
 
-    class SearchLivePublicMixin:
+    class SearchOverrideMixin:
         def resolve_search(self, info, **kwargs):
             query = kwargs.get("query")
             if not query:
                 return None
             s = get_search_backend()
-            results = []
-            models = [get_document_model(), get_image_model()]
-            for app in grapple_registry.apps:
-                models += django_apps.all_models[app].values()
-            for model in models:
-                if issubclass(model, Page):
-                    results += s.search(query, model.objects.live().public())
-                else:
-                    results += s.search(query, model)
-            return results
+            ranked = list(s.search(query, Page.objects.live().public()))
+            if not ranked:
+                return []
+            ids = [p.id for p in ranked]
+            specific_map = {p.id: p for p in Page.objects.filter(id__in=ids).specific()}
+            return [specific_map[i] for i in ids if i in specific_map]
 
-    query_mixins.insert(0, SearchLivePublicMixin)
+    query_mixins.insert(0, SearchOverrideMixin)
 
 
 @hooks.register("construct_page_action_menu")
diff --git a/dnscms/tests/test_search.py b/dnscms/tests/test_search.py
index 6a183af..936de88 100644
--- a/dnscms/tests/test_search.py
+++ b/dnscms/tests/test_search.py
@@ -86,3 +86,46 @@ def test_search_excludes_draft_event_page(home_page, event_index, graphql_post):
     assert response.status_code == 200
     assert "errors" not in body, body
     assert "DraftEventSearchToken" not in _titles_for(body, "EventPage")
+
+
+def test_search_results_not_grouped_by_type(home_page, event_index, graphql_post):
+    # Two pages of different types matching the query equally, plus a third
+    # page of one of those types that should rank highest. Under the
+    # per-model-iteration resolver, all Generic results come before all Event
+    # results (or vice versa) — type-grouped — so the highest-relevance Event
+    # ends up after a less-relevant Generic. Cross-type relevance ordering
+    # should put the strongest match first regardless of type.
+    weak_generic = GenericPageFactory(
+        parent=home_page,
+        title="Klatremus klatremus klatremus",
+        slug="weak-generic",
+    )
+    weak_event = EventPageFactory(
+        parent=event_index,
+        title="Klatremus klatremus klatremus",
+        slug="weak-event",
+    )
+    strong_event = EventPageFactory(
+        parent=event_index,
+        title="Klatremus klatremus klatremus klatremus klatremus klatremus",
+        slug="strong-event",
+    )
+    _index(weak_generic)
+    _index(weak_event)
+    _index(strong_event)
+
+    response, body = graphql_post(SEARCH_QUERY, {"query": "klatremus"})
+
+    assert response.status_code == 200
+    assert "errors" not in body, body
+    order = [
+        (r["__typename"], r["title"])
+        for r in body["data"]["results"]
+        if r["__typename"] in ("GenericPage", "EventPage")
+    ]
+    assert len(order) == 3, order
+    # Per-type grouping would put all results of one type consecutively
+    # before the other type. Cross-type relevance ordering should interleave.
+    types_seen = [t for t, _ in order]
+    assert types_seen != ["GenericPage", "EventPage", "EventPage"], order
+    assert types_seen != ["EventPage", "EventPage", "GenericPage"], order