diff --git a/lib/banzai/filter/issuable_state_filter.rb b/lib/banzai/filter/issuable_state_filter.rb
index d7fe012883d..8e2358694d4 100644
--- a/lib/banzai/filter/issuable_state_filter.rb
+++ b/lib/banzai/filter/issuable_state_filter.rb
@@ -18,7 +18,7 @@ module Banzai
         issuables = extractor.extract([doc])
 
         issuables.each do |node, issuable|
-          next if !can_read_cross_project? && issuable.project != project
+          next if !can_read_cross_project? && cross_reference?(issuable)
 
           if VISIBLE_STATES.include?(issuable.state) && issuable_reference?(node.inner_html, issuable)
             node.content += " (#{issuable.state})"
@@ -31,7 +31,14 @@ module Banzai
       private
 
       def issuable_reference?(text, issuable)
-        text == issuable.reference_link_text(project || group)
+        CGI.unescapeHTML(text) == issuable.reference_link_text(project || group)
+      end
+
+      def cross_reference?(issuable)
+        return true if issuable.project != project
+        return true if issuable.respond_to?(:group) && issuable.group != group
+
+        false
       end
 
       def can_read_cross_project?
diff --git a/lib/banzai/issuable_extractor.rb b/lib/banzai/issuable_extractor.rb
index 0a05d46db4c..341dbb74fe0 100644
--- a/lib/banzai/issuable_extractor.rb
+++ b/lib/banzai/issuable_extractor.rb
@@ -9,13 +9,11 @@ module Banzai
   # so we can avoid N+1 queries problem
 
   class IssuableExtractor
-    QUERY = %q(
-      descendant-or-self::a[contains(concat(" ", @class, " "), " gfm ")]
-      [@data-reference-type="issue" or @data-reference-type="merge_request"]
-    ).freeze
-
     attr_reader :context
 
+    ISSUE_REFERENCE_TYPE = '@data-reference-type="issue"'.freeze
+    MERGE_REQUEST_REFERENCE_TYPE = '@data-reference-type="merge_request"'.freeze
+
     # context - An instance of Banzai::RenderContext.
     def initialize(context)
       @context = context
@@ -24,21 +22,38 @@ module Banzai
     # Returns Hash in the form { node => issuable_instance }
     def extract(documents)
       nodes = documents.flat_map do |document|
-        document.xpath(QUERY)
+        document.xpath(query)
       end
 
-      issue_parser = Banzai::ReferenceParser::IssueParser.new(context)
-
-      merge_request_parser =
-        Banzai::ReferenceParser::MergeRequestParser.new(context)
-
-      issuables_for_nodes = issue_parser.records_for_nodes(nodes).merge(
-        merge_request_parser.records_for_nodes(nodes)
-      )
-
-      # The project for the issue/MR might be pending for deletion!
+      # The project or group for the issuable might be pending for deletion!
       # Filter them out because we don't care about them.
-      issuables_for_nodes.select { |node, issuable| issuable.project }
+      issuables_for_nodes(nodes).select { |node, issuable| issuable.project || issuable.group }
+    end
+
+    private
+
+    def issuables_for_nodes(nodes)
+      parsers.each_with_object({}) do |parser, result|
+        result.merge!(parser.records_for_nodes(nodes))
+      end
+    end
+
+    def parsers
+      [
+        Banzai::ReferenceParser::IssueParser.new(context),
+        Banzai::ReferenceParser::MergeRequestParser.new(context)
+      ]
+    end
+
+    def query
+      %Q(
+        descendant-or-self::a[contains(concat(" ", @class, " "), " gfm ")]
+        [#{reference_types.join(' or ')}]
+      )
+    end
+
+    def reference_types
+      [ISSUE_REFERENCE_TYPE, MERGE_REQUEST_REFERENCE_TYPE]
     end
   end
 end