Prompts and Prompt Templates

What prompts are, why templates help, the main template types, worked examples including few-shot, and prompt engineering tips.

04LLM Models and Integrations 06Chains in LangChain 11LangChain with LLM Providers

5.1

What are Prompts?

A prompt is the input instruction given to the LLM.
It guides the model to generate the desired output.
Good prompts lead to better, accurate and consistent responses.
LangChain provides prompt templates to create dynamic and reusable prompts.

5.2

Why Use Prompt Templates?

Reusable and maintainable prompts.
Insert dynamic variables (e.g., user input).
Separate prompt logic from code.
Easy to manage complex prompts.
Support different prompt types (chat, system, human, etc.).
Helps in building structured and consistent LLM applications.

5.3

Types of Prompts in LangChain

Type	Description	Use Case
PromptTemplate	Simple text prompt with variables	Basic Q&A, text generation
ChatPromptTemplate	Chat-based prompt with system, human, AI messages	Chat models (ChatOpenAI, ChatAnthropic, etc.)
MessagesPlaceholder	Placeholder for dynamic message history	Chat with conversation history
FewShotPromptTemplate	Use examples in the prompt	Few-shot learning
PipelinePromptTemplate	Combine multiple templates	Complex multi-step prompts

5.4

PromptTemplate

python

from langchain_core.prompts import PromptTemplate

template = PromptTemplate(
    input_variables=["topic"],
    template="Explain {topic} in simple words."
)

prompt = template.format(topic="LangChain")
print(prompt)
# Output: Explain LangChain in simple words.

5.5

ChatPromptTemplate

python

from langchain_core.prompts import ChatPromptTemplate

template = ChatPromptTemplate.from_messages([
    ("system", "You are a helpful AI assistant."),
    ("human", "Explain {topic} in simple words.")
])

prompt = template.format_messages(topic="LangChain")

# This creates formatted chat messages
# to be used with ChatOpenAI or other chat models.

5.6

FewShotPromptTemplate

python

from langchain_core.prompts import FewShotPromptTemplate, PromptTemplate

examples = [
    {"input": "Python", "output": "A programming language."},
    {"input": "LangChain", "output": "A framework to build LLM applications."}
]

example_prompt = PromptTemplate(
    input_variables=["input", "output"],
    template="Input: {input}\nOutput: {output}"
)

few_shot = FewShotPromptTemplate(
    examples=examples,
    example_prompt=example_prompt,
    prefix="Answer the question:",
    suffix="Input: {question}\nOutput:",
    input_variables=["question"]
)

5.7

Prompt Engineering Tips

Be clear and specific in instructions.
Provide examples (few-shot) when needed.
Use a structured format (JSON, lists, etc.).
Set a role using system prompts.
Break complex tasks into smaller steps.
Use constraints (e.g., length, tone, format).
Experiment with different phrasings.

5.8

Example: Using a Prompt with an LLM

python

from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate

llm = ChatOpenAI(model="gpt-3.5-turbo", temperature=0.7)
template = ChatPromptTemplate.from_messages([
    ("human", "Explain {topic} in simple words.")
])

chain = template | llm
response = chain.invoke({"topic": "LangChain"})
print(response.content)