Prompt Engineering Guide

What is inside

The guide is organised into sections you can dip into:

• Introduction: LLM settings (temperature, top-p and so on), prompt elements and general tips.
• Prompting techniques: zero-shot and few-shot, chain-of-thought, self-consistency, generated knowledge, prompt chaining, tree of thoughts, retrieval augmented generation, ReAct and more, each with a short explanation, an example and the source paper.
• AI agents: components of an agent and function calling.
• Applications: fine-tuning, code generation, generating synthetic data.
• Prompt Hub: example prompts grouped by task.
• Models: notes on widely used LLMs.
• Risks and misuses: adversarial prompting, prompt injection, factuality and bias.
• Research findings and a papers list for further reading.

The GitHub repository behind it had about 78,600 stars and 8,600 forks as of September 2026, making it one of the most widely used open prompting references. It is MIT-licensed; the last repository update was in March 2026.

Why the citations matter

A lot of prompting advice circulates as folklore, and some of it was never true or stopped being true as models improved. Because each technique links to its paper, you can check when it was published, which models it was tested on, and how large the effect was. Techniques that gave big gains on 2022 models sometimes add little on current ones.

How to use it

• Read the introduction and the core techniques (few-shot, chain-of-thought, prompt chaining) in one sitting of an hour or two.
• Treat the rest as a reference to look up when you meet a term or a problem.
• Try each technique on a task of your own and compare results with a plain prompt. Keep what measurably helps.
• For the specific model you are building on, follow the provider's own prompting documentation, which reflects how that model was trained and changes with new versions.

Where it is thin

It is a survey, so explanations are short. Some model pages and examples lag behind the newest releases, and it does not teach systematic evaluation in depth. Pair it with a practical course on evaluation, or Anthropic's and OpenAI's prompting guides, when you are shipping a feature.

DAIR.AI also runs paid courses through its academy (for example on prompt engineering and building AI agents). The guide itself is free.

Common questions

Is it free? Yes. The guide is free to read and the source is open.

Is it tied to one provider? No. Examples mostly use common APIs, but the techniques are general.

Is it suitable for beginners? Yes for the introduction and basic techniques. Later sections assume some familiarity with LLM APIs.

Best for: developers and students who want a free, evidence-linked overview of prompting techniques and LLM concepts.