Prompt Introduction

Open a prompt by stating what kind of document it is so the model interprets everything that follows through the right lens from the first token.

Most prompts should open with an introduction that clarifies the type of document being written and sets the model up to interpret everything that follows correctly — for example, stating "This is about recommending a book" focuses the model on book-recommendation-relevant aspects of whatever context comes next. This matters because the model has a fixed amount of computation per token and cannot pause partway through a prompt to go back and reinterpret earlier content in light of something learned later — guiding its focus early changes how it reads everything downstream, not just what it produces at the end.

Most prompts need only one introduction, for the main question. The same principle applies recursively to subsections, though: if the model needs to focus on some particular aspect of a piece of embedded context, set that aspect up at the very beginning of that piece, not after it.

The introduction sets the stage in general terms ("I'm thinking about book suggestions for X"); it is deliberately less precise than the refocus that closes the prompt, which gives operational detail instead.

---

Sources

• Prompt Engineering for LLMs (John Berryman), ch. 6