Prompt

A prompt is everything the model sees in one turn: your question, the conversation before it, any system instructions, and any documents you attached. In practice, “prompt” = the full input.

Most systems have at least two layers:

1. System prompt — invisible instructions set by the app (“you are a helpful assistant, always answer in French…”). You don’t write this; the product does.
2. Your prompt — the message you type. Plus whatever context the app re-injects (history, files, search results).

Prompting is less “magic words” and more communication. What actually works:

• Be concrete. “Fix this function” → “This function returns null on empty input; make it return [] instead.”
• Show examples. One good example beats three paragraphs of description.
• Give it the material. If the answer needs data, put the data in the prompt — the model can’t mind-read.
• Iterate. Treat the first answer as a draft, then refine the prompt.