Prompt

A prompt can consist of several building blocks: the actual instruction, context such as documents or background information, format requirements for the desired answer, and examples. Language models generate their output directly from this input – they don't guess what was meant, they process what is actually there. That is why the same task, worded differently, produces noticeably different results. Working with this deliberately is called prompt engineering: inputs are refined step by step, tested, and made reusable.

In everyday business, prompt quality decides whether AI tools save time or create rework. A vague request like "summarize this" rarely delivers what you need; it helps to include audience, length, and purpose – for example, "summarize this proposal in five bullet points for the management team". Many teams therefore collect proven prompts as templates, so that good wording doesn't stay with individuals but gets reused across the whole team.

At Chifty, prompts are the starting point in Chifty Chat: employees phrase their task and choose between models from OpenAI, Google, Anthropic, and Mistral. When a prompt refers to internal knowledge, the answer draws exclusively on sources the asking person is allowed to see – with sources shown. The permission filter is applied at query time, directly in the search; even a cleverly worded prompt cannot bypass it.