What is prompt writing?

Prompt writing, in AI training, means writing the questions and instructions used to test or train a model. The goal is usually not an easy question but a hard one: a prompt the current model is likely to get wrong, so its answer reveals a weakness that can be fixed.

What the work looks like

You write prompts in your own field, often with constraints such as a required format, a specific audience or a multi-step task. Many projects then ask you to run the prompt against the model, judge its answer, and write the ideal answer yourself. Prompts that the model already answers perfectly are usually rejected, because they teach it nothing.

What makes a strong prompt

• It reflects a real task someone in your field would face.
• It has one clearly correct answer, or clear criteria for a good one.
• It targets a known weak spot, such as multi-step reasoning, recent rules or edge cases.
• It is self-contained, with everything needed to answer it.

Example

A tax specialist writes: "A freelancer moved countries mid-year and was paid by clients in both. Which income is taxable where in the year of the move?" The model mixes up residency rules, so the prompt is accepted along with a corrected answer.