Prompt engineering is the practice of structuring instructions to an AI model to get specific, useful outputs. It involves describing outcomes, providing context, constraining output format, including examples, setting negative constraints, and iterating on responses.

PMs write better prompts by describing desired outcomes rather than implementation steps, providing business and technical context upfront, specifying output format, including examples of what good output looks like, and adding explicit constraints about what to avoid.

A good prompt includes six elements: a clear outcome description, relevant context (technical, business, and what has been tried), output format constraints, one or more concrete examples, explicit don't rules, and a willingness to iterate rather than restart.

Yes. Including one or two examples of the desired output format significantly improves consistency. This technique is called few-shot prompting. One working example communicates more than five sentences of description.

Length depends on task complexity. Simple tasks may need two to three sentences. Complex tasks benefit from 100 to 300 words of context, constraints, and examples. A longer prompt with relevant detail almost always outperforms a short vague one.

Negative constraints are explicit 'don't' rules that prevent the AI from doing things you did not ask for. Examples include 'don't add error handling I didn't ask for' or 'don't modify files outside the scope.' They are highly effective at preventing AI scope creep.

Usually not. Targeted corrections within the same conversation are faster and produce better results because the AI retains context from previous exchanges. Only start fresh if the conversation has gone fundamentally off track after multiple correction attempts.

Few-shot prompting is including one or more concrete examples of the desired output in your prompt. The AI uses these examples as a template, matching the pattern, structure, naming conventions, and style. It is one of the most reliable techniques for consistent results.

The core principles — clarity, context, constraints, examples — work across all major AI models. Specific behaviors may vary, but a well-structured prompt produces better results on any model compared to a vague one.

If the AI's response includes phrases like 'assuming you are using...' or 'depending on your setup...', it is guessing because you did not provide enough context. Those hedging phrases are a signal to add more detail to your prompt.