Zero-shot prompting

Zero-shot prompting is the practice of asking a large language model to perform a task using only an instruction, without including any example outputs.

Zero-shot is the simplest possible prompting style: "Write a tweet about X". It works for tasks the model has seen many examples of in training. It fails — or produces inconsistent output — for tasks that require specific format, tone, or constraints the model wasn’t trained on.

For brand voice work, zero-shot prompting is usually too weak. The model has no examples of how the specific brand sounds, so it defaults to its training-data average voice — which is generic by construction. Few-shot prompting (including a few example outputs in the prompt) or retrieval-augmented generation (pulling examples from the brand’s corpus) both materially improve output quality.

Marketing tools that rely on zero-shot prompting under the hood produce the same generic voice as the underlying model. Tools that include examples — whether hand-crafted or retrieved — produce noticeably more on-brand output.