The practice of structuring text inputs to a generative model so that its outputs more reliably meet a desired specification, including format, tone, factual grounding, and refusal behaviour.
In practicePrompt engineering covers instruction phrasing, role framing, exemplars (few-shot prompting), chain-of-thought scaffolds, retrieval grounding, and output schemas. It is a software-engineering discipline in its own right: prompts are versioned, regression-tested against evaluation sets, and shipped behind feature flags. In regulated settings, prompts form part of the model documentation and are reviewed alongside training data and guardrails, because a prompt change can materially alter system behaviour without any change to the underlying weights.
A claims-triage assistant uses a prompt that fixes the persona, lists the only permissible output fields as JSON, supplies three worked examples of edge cases, and instructs the model to return a refusal token when the input lacks a policy number.
This definition is maintained by Moweb partners and used in live client engagements. For how Prompt engineering applies to your estate, or to challenge a working definition, speak to a partner.