Prompting Concepts
Slow-moving principles behind the model guides. These pages explain the mechanics that show up across ChatGPT, Claude, Gemini, open-weight models, and API products.
Chain of thought and reasoning models
Why visible step-by-step scaffolding is not the same as better reasoning.
Last reviewed: August 25, 2026Context window management
How to keep long conversations and large documents useful when the model can only see so much at once.
Last reviewed: August 25, 2026Few-shot prompting
Examples still help, but they work best when they define a hard boundary or voice target.
Last reviewed: August 10, 2026Prompt iteration
A practical way to debug a prompt by changing one thing at a time.
Last reviewed: August 10, 2026Structured output prompting
How JSON mode, schemas, and prose instructions fit together across current model APIs.
Last reviewed: August 10, 2026System prompts
What a system prompt is, where it belongs, and how to write one without turning it into policy soup.
Last reviewed: August 25, 2026Tokens and context windows
Why token counts vary by model, language, tools, files, and conversation length.
Last reviewed: August 10, 2026Tool use and function calling
The prompting basics for models that can call tools, fetch data, or hand work to application functions.
Last reviewed: August 25, 2026XML tags in prompts
When XML-style tags help a model parse context, and when they are just decoration.
Last reviewed: August 10, 2026