Prompt engineering is the practice of designing the text instructions, called prompts, that are given to a large language model so that the model returns output that is accurate, consistent and in a usable form. Common techniques include stating the role and the task plainly, supplying worked examples, asking for a fixed output format such as a table or a score, and breaking a large task into smaller steps. The same model can answer differently worded prompts very differently.
In markets, prompts are written to make tools such as ChatGPT classify headlines for sentiment analysis, summarize company filings, extract figures from text or draft research code. The output is not guaranteed to be correct: models can state false facts with confidence, and results change between versions. A prompt that feeds a trading signal therefore needs the same testing as any other model input, including backtesting on data the model has not seen.