diff --git a/ee/lib/gitlab/llm/chain/tools/summarize_comments/executor.rb b/ee/lib/gitlab/llm/chain/tools/summarize_comments/executor.rb
index 3293807e5172896c2895694e51152da3c4716480..13d4d977b2b0fb1f85e690423d128c25458dbe1a 100644
--- a/ee/lib/gitlab/llm/chain/tools/summarize_comments/executor.rb
+++ b/ee/lib/gitlab/llm/chain/tools/summarize_comments/executor.rb
@@ -30,16 +30,16 @@ class Executor < Tool
               Utils::Prompt.as_system(
                 <<~PROMPT
                   You are an assistant that extracts the most important information from the comments in maximum 10 bullet points.
-                  Comments are between two identical sets of 3-digit numbers surrounded by < > sign.
+                  Each comment is wrapped in a <comment> tag.
 
-                  <%<num>s>
                   %<notes_content>s
-                  <%<num>s>
 
                   Desired markdown format:
                   **<summary_title>**
-                  <bullet_points>
-                  """
+                  - <bullet_point>
+                  - <bullet_point>
+                  - <bullet_point>
+                  - ...
 
                   Focus on extracting information related to one another and that are the majority of the content.
                   Ignore phrases that are not connected to others.
@@ -57,7 +57,6 @@ def perform(&block)
               content = if notes.exists?
                           notes_content = notes_to_summarize(notes) # rubocop: disable CodeReuse/ActiveRecord
                           options[:notes_content] = notes_content
-                          options[:num] = Random.rand(100..999)
 
                           if options[:raw_ai_response]
                             request(&block)
@@ -87,7 +86,7 @@ def notes_to_summarize(notes)
 
                   break notes_content if notes_content.size + note[1].size >= input_content_limit
 
-                  notes_content << note[1]
+                  notes_content << (format("<comment>%<note>s</comment>", note: note[1]))
                 end
               end
 
diff --git a/ee/spec/lib/gitlab/llm/chain/tools/summarize_comments/prompts/anthropic_spec.rb b/ee/spec/lib/gitlab/llm/chain/tools/summarize_comments/prompts/anthropic_spec.rb
index 2613fe8b35986416b278b35b99724485d12c143c..092d4f8d139bb5bc9d5a3dff60bccce6efb7ec8b 100644
--- a/ee/spec/lib/gitlab/llm/chain/tools/summarize_comments/prompts/anthropic_spec.rb
+++ b/ee/spec/lib/gitlab/llm/chain/tools/summarize_comments/prompts/anthropic_spec.rb
@@ -5,7 +5,7 @@
 RSpec.describe Gitlab::Llm::Chain::Tools::SummarizeComments::Prompts::Anthropic, feature_category: :duo_chat do
   describe '.prompt' do
     it 'returns prompt' do
-      prompt = described_class.prompt({ notes_content: 'foo', num: 123 })[:prompt]
+      prompt = described_class.prompt({ notes_content: '<comment>foo</comment>' })[:prompt]
 
       expect(prompt).to include('Human:')
       expect(prompt).to include('Assistant:')
@@ -13,16 +13,16 @@
       expect(prompt).to include(
         <<~PROMPT
           You are an assistant that extracts the most important information from the comments in maximum 10 bullet points.
-          Comments are between two identical sets of 3-digit numbers surrounded by < > sign.
+          Each comment is wrapped in a <comment> tag.
 
-          <123>
-          foo
-          <123>
+          <comment>foo</comment>
 
           Desired markdown format:
           **<summary_title>**
-          <bullet_points>
-          """
+          - <bullet_point>
+          - <bullet_point>
+          - <bullet_point>
+          - ...
 
           Focus on extracting information related to one another and that are the majority of the content.
           Ignore phrases that are not connected to others.
diff --git a/ee/spec/lib/gitlab/llm/chain/tools/summarize_comments/prompts/vertex_ai_spec.rb b/ee/spec/lib/gitlab/llm/chain/tools/summarize_comments/prompts/vertex_ai_spec.rb
index 32bef186e4bf5a9826e7d62c3a3cc4949bdfcf55..1033e17a1c1c99f5bd5df6f70b9c51793de5cf6e 100644
--- a/ee/spec/lib/gitlab/llm/chain/tools/summarize_comments/prompts/vertex_ai_spec.rb
+++ b/ee/spec/lib/gitlab/llm/chain/tools/summarize_comments/prompts/vertex_ai_spec.rb
@@ -5,21 +5,21 @@
 RSpec.describe Gitlab::Llm::Chain::Tools::SummarizeComments::Prompts::VertexAi, feature_category: :duo_chat do
   describe '.prompt' do
     it 'returns prompt' do
-      prompt = described_class.prompt({ notes_content: 'foo', num: 123 })[:prompt]
+      prompt = described_class.prompt({ notes_content: '<comment>foo</comment>' })[:prompt]
 
       expect(prompt).to include(
         <<~PROMPT
           You are an assistant that extracts the most important information from the comments in maximum 10 bullet points.
-          Comments are between two identical sets of 3-digit numbers surrounded by < > sign.
+          Each comment is wrapped in a <comment> tag.
 
-          <123>
-          foo
-          <123>
+          <comment>foo</comment>
 
           Desired markdown format:
           **<summary_title>**
-          <bullet_points>
-          """
+          - <bullet_point>
+          - <bullet_point>
+          - <bullet_point>
+          - ...
 
           Focus on extracting information related to one another and that are the majority of the content.
           Ignore phrases that are not connected to others.