What is a PromptTemplate in LangChain, and how is it used?

What you need to know

Without templates, prompts get built with f-strings in many places. Then one change to wording means hunting through the code, and a missing variable silently sends a broken prompt.

1from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder23prompt = ChatPromptTemplate.from_messages([4    ("system", "You are a support agent for {company}. Answer in {language}. "5               "Use only the help articles below.\n\n{articles}"),6    MessagesPlaceholder("history", optional=True),7    ("human", "{question}"),8]).partial(company="PayEase")910value = prompt.invoke({"language": "English", "articles": "...",11                       "question": "How do I reset my PIN?"})12print(value.to_messages())

prompt.invoke returns a prompt value that holds a list of messages. In a chain you rarely call it yourself: prompt | llm passes that value straight to the model.

The main types

Useful features

• Validation. If a variable is missing, invoke raises a KeyError that names it, before any money is spent on a model call.
• partial. Pre-fill a variable with a value, or with a function that runs at format time, such as today's date.
• Escaping. A literal brace in the prompt, like a JSON example, must be doubled: {{"total": 0}}.
• Versioning. Prompts can be stored in LangSmith and loaded with Client().pull_prompt("name"), so they can change without a code deploy.

A real-life example

A support bot over a fintech company's help-centre docs had its prompt copied into three services. When legal asked for a new line ("Never promise a refund date"), one service was missed, and the bot kept promising "refund in 3 days" for another month. The team moved the prompt into one ChatPromptTemplate in a shared module, with {articles} and {question} as variables and company filled by partial. The next wording change was one pull request, reviewed once, and the unit test that renders the template with sample inputs caught a typo ({artciles}) before release.

Follow-up questions to expect

• "Why does my JSON example in the prompt cause a missing-variable error?" — Single braces are read as variables. Double them, {{ }}, or use template_format="mustache".
• "How do you add chat history to a prompt?" — Put a MessagesPlaceholder("history") between the system and human messages and pass a list of messages under that key.
• "String template or chat template?" — Chat template, because every current model is a chat model and system messages are handled properly.