Concept Intermediate · 3 min read

What is ReAct prompting

Q: What is ReAct prompting

ReAct prompting is a prompt engineering technique that integrates reasoning and acting steps within a single prompt to enable language models to think through problems and take actions iteratively. It improves complex task performance by allowing the model to generate intermediate reasoning traces and execute commands or queries in context.

Quick answer

ReAct prompting is a prompt engineering technique that integrates reasoning and acting steps within a single prompt to enable language models to think through problems and take actions iteratively. It improves complex task performance by allowing the model to generate intermediate reasoning traces and execute commands or queries in context.

ReAct prompting is a prompt engineering method that combines reasoning and acting steps to enable language models to think and interact with environments for better task solving.

How it works

ReAct prompting works by interleaving reasoning steps (thoughts, deductions) with acting steps (actions, commands) in the model's output. This mimics human problem-solving where you think about the problem, then take an action, then reflect again. The model generates a chain of thought and decides when to perform an action like querying a database or searching the web, then uses the result to continue reasoning.

Think of it as a conversation where the AI alternates between "thinking aloud" and "doing something" to gather information or update its knowledge before answering.

Concrete example

Here is a simplified ReAct prompt example for a question-answering task that requires web search:

python

from openai import OpenAI
import os

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

prompt = '''
Question: Who won the Best Actor Oscar in 2023?

Thought: I need to find the latest Oscar winners.
Action: Search["2023 Best Actor Oscar winner"]
Observation: Brendan Fraser won the Best Actor Oscar in 2023.
Thought: Now I know the answer.
Answer: Brendan Fraser
'''

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": prompt}]
)

print(response.choices[0].message.content)

output

Brendan Fraser

When to use it

Use ReAct prompting when tasks require both complex reasoning and interaction with external tools or environments, such as web search, database queries, or API calls. It is ideal for multi-step problem solving where intermediate information gathering is needed.

Do not use it for simple, direct questions or tasks that do not require external actions, as the overhead of reasoning and acting steps may reduce efficiency.

Key terms

Term	Definition
ReAct prompting	A prompting technique combining reasoning and acting steps in language model outputs.
Reasoning	The process of generating intermediate thoughts or chains of thought.
Acting	Performing an action such as querying a tool or environment based on reasoning.
Chain-of-thought	A sequence of reasoning steps generated by the model.
Observation	The result or feedback from an action taken by the model.

✅

Key Takeaways

Use ReAct prompting to combine reasoning and actions for complex, interactive tasks.
Interleave thoughts and actions in prompts to enable iterative problem solving.
Ideal for tasks requiring external tool use like search or API calls.
Avoid for simple queries where direct answers suffice.
Include clear action and observation steps in the prompt for best results.

Verified 2026-04 · gpt-4o

Verify ↗