Not a search query

People often treat a prompt the way they treat a search query: a few keywords, sent off, and the model figures it out.

That works sometimes. But it misses something important about how these models work.

A search engine takes your keywords and retrieves documents that contain them. A language model takes your prompt and completes it — it produces the text that would most plausibly follow from everything in the context window.

This means the prompt isn't just a question. It's the beginning of a context the model is continuing. The phrasing, the tone, the examples you include, the instructions you give — all of it shapes what "plausibly follows."

A vague prompt gets a generic response not because the model is lazy, but because a generic response is what plausibly follows a vague prompt. A specific, well-structured prompt pulls the model toward a specific, well-structured response — because that's what plausibly follows.

The model is completing a pattern. You can use that.