Messages in ChatModel

​

Core message types

• System message — a global instruction that defines the assistant’s persona, behavior, or tone for the session (for example, “You are a dietician” or “You are a physics teacher”).
• Human message — the user’s input or prompt. Human messages are sent repeatedly as the conversation continues.
• AI message — the model’s responses generated in reply to the human messages and shaped by the system message.

​

Quick reference table

​

Example: building a chat prompt in LangChain (Python)

from langchain.chat_models import ChatOpenAI
from langchain.schema import SystemMessage, HumanMessage, AIMessage

# Initialize a chat model (e.g., LangChain's wrapper for OpenAI)
chat = ChatOpenAI(temperature=0.2)

# Build the conversation: system message first, then a human message
messages = [
    SystemMessage(content="You are a professional dietician who gives concise, evidence-based advice."),
    HumanMessage(content="What is a healthy breakfast for someone trying to lose weight?")
]

# Call the chat model with the messages list
response = chat(messages)

# The reply is available as an AI message; print its text content
print(response.content)

​

Best practices and common usage patterns

• Always place the system message at the start of the message list so it influences later replies.
• Send human messages repeatedly as the user interacts; the model will reply with AI messages for each turn.
• If you need to change the assistant’s persona mid-session, prefer replacing the system message rather than appending multiple system messages (multiple system messages can be confusing).
• To preserve a persona across separate sessions, include the system message at the start of each new session—models do not retain state between sessions.

​

Next steps

​

Links and references

• LangChain
• Introduction to OpenAI course