Advanced prompting: pro techniques

You know the context, format and role. Here are the four techniques professionals actually use, ranked by power.

1. Show examples (few-shot)

Instead of describing what you want, show it. "Rewrite my notes as a report. Here's an example: [raw notes] → [finished report]. Now do the same with: …". One or two examples beat ten lines of explanation: the model picks up the format, tone and level of detail in one go.

2. Make it think before answering

"Think step by step before giving your conclusion." This simple phrase noticeably improves answers on logic, calculation or analysis problems. Recent so-called "reasoning" models do this internally, but spelling it out is still useful, because you can read the reasoning and spot where it goes off the rails.

3. Break down complex tasks

Don't ask for "write me a business plan" in one go. Chain it: first the structure, then each section, then a critical review of the whole. Each step is better because the model focuses on one thing, and you can adjust course between steps.

4. AI critiquing AI

Once you have the answer: "Re-read your answer like a demanding proofreader: errors, omissions, weaknesses?" The model often finds its own mistakes. A powerful variant: ask for two opposing approaches, then a synthesis.