diff --git a/llm/api.py b/llm/api.py
index 767535c..e64a750 100644
--- a/llm/api.py
+++ b/llm/api.py
@@ -49,7 +49,13 @@ def create_chat(request):
 
         openai.api_key = organization.openai_key
 
-        prompt = request.data.get("prompt").strip()
+        question = request.data.get("question").strip()
+        system_prompt = (
+            request.data.get("system_prompt", None) or organization.system_prompt
+        )
+        system_prompt = system_prompt.strip() if system_prompt else None
+        logger.info(f"Using the system prompt : {system_prompt}")
+
         gpt_model = request.data.get("gpt_model", "gpt-3.5-turbo").strip()
         session_id = (request.data.get("session_id") or generate_session_id()).strip()
 
@@ -59,7 +65,7 @@ def create_chat(request):
             messages=[
                 {
                     "role": "user",
-                    "content": f"Detect the languages in this text: {prompt}",
+                    "content": f"Detect the languages in this text: {question}",
                 }
             ],
             functions=[
@@ -111,7 +117,7 @@ def create_chat(request):
 
         # 2. Pull relevant chunks from vector database
         prompt_embeddings = openai.Embedding.create(
-            model="text-embedding-ada-002", input=prompt
+            model="text-embedding-ada-002", input=question
         )["data"][0]["embedding"]
 
         embedding_results = (
@@ -146,10 +152,11 @@ def create_chat(request):
         # 3. Fetch the chat history from our message store to send to openai and back in the response
         historical_chats = Message.objects.filter(session_id=session_id).all()
 
-        # 4. Retrievial question and answer (2nd call to OpenAI, use language from 1. to help LLM respond in same language as user question)
+        # 4. Retrieval question and answer (2nd call to OpenAI, use language from 1. to help LLM respond in same language as user question)
         response = openai.ChatCompletion.create(
             model=gpt_model,
             messages=context_prompt_messages(
+                system_prompt,
                 organization.id,
                 language_results["language"],
                 relevant_english_context,
@@ -168,7 +175,7 @@ def create_chat(request):
             evaluator_prompts = organization.evaluator_prompts  # { 'coherence': ... }
             for criteria, evaluator_prompt in evaluator_prompts.items():
                 score = evaluate_criteria_score(
-                    evaluator_prompt, prompt, prompt_response, gpt_model
+                    evaluator_prompt, question, prompt_response, gpt_model
                 )
                 evaluation_scores[criteria] = score
                 logger.info(f"Evaluated criteria: {criteria} with score: {score}")
@@ -182,7 +189,7 @@ def create_chat(request):
         Message.objects.create(
             session_id=session_id,
             role="user",
-            message=prompt,
+            message=question,
             evaluation_score=evaluation_scores,
         )
         Message.objects.create(
@@ -195,7 +202,7 @@ def create_chat(request):
 
         return JsonResponse(
             {
-                "question": prompt,
+                "question": question,
                 "answer": prompt_response.content,
                 "language_results": language_results,
                 "embedding_results_count": len(embedding_results),
diff --git a/llm/utils/prompt.py b/llm/utils/prompt.py
index eab53af..7ec47d1 100644
--- a/llm/utils/prompt.py
+++ b/llm/utils/prompt.py
@@ -9,6 +9,7 @@
 
 
 def context_prompt_messages(
+    system_prompt: str,
     organization_id: int,
     language: str,
     english_context: str,
@@ -17,10 +18,9 @@ def context_prompt_messages(
 ) -> list[dict]:
     org = Organization.objects.filter(id=organization_id).first()
 
-    org_system_prompt = org.system_prompt
     examples_text = org.examples_text
 
-    system_message_prompt = {"role": "system", "content": org_system_prompt}
+    system_message_prompt = {"role": "system", "content": system_prompt}
     human_message_prompt = {
         "role": "user",
         "content": f"""Use the following pieces of context to answer the question at the end. If you don't know the answer, just say that you don't know, don't try to make up an answer.