A few power techniques worth knowing

Once you've got the basics (specific, context, format, iterate), a few accessible "power techniques" can level up your results on trickier tasks. None require any technical skill.

1. Ask it to work step by step (for reasoning tasks). For problems involving logic, math, or multiple steps, asking the AI to "think through it step by step" or "show your reasoning" often improves accuracy — it works through the problem rather than jumping to an answer. An important 2026 note: the newest "reasoning" models already do this internally, so you often don't need to add "think step by step" with them (and some let you dial up a "reasoning effort" instead). But for regular chat models, this step-by-step nudge still genuinely helps on multi-step problems. Treat it as a useful tool for the right situation, not a universal magic phrase.

2. Give the AI permission to say "I don't know." One of the most practical anti-error habits: tell the AI "If you're not sure or don't have the information, say so rather than guessing." This noticeably reduces made-up answers, because otherwise AI tends to produce a confident response even when it shouldn't. A simple line that makes AI more trustworthy.

3. Have it ask YOU questions. For a complex or fuzzy task, add: "Before you start, ask me any clarifying questions you need." Instead of guessing at the gaps, the AI gathers what it's missing — leading to a much better result. (Tip: it's often best to fold the answers back into a cleaner prompt rather than a long back-and-forth.)

4. Break big tasks into steps. For anything large, don't ask for it all at once — break it up. Ask for an outline first, then have the AI fill in each section. Tackling a big task in pieces (and reviewing each) produces better, more manageable results than one giant request.

Use these for the right moments. You won't need these on simple requests — but for reasoning problems, uncertain topics, complex tasks, and big projects, they're genuinely useful. Combined with the basics, they cover almost everything a regular person needs.

The mindset: a few accessible power techniques level up trickier tasks: ask the AI to work step by step on reasoning/math problems (though note the newest "reasoning" models do this internally, so you may not need to), give it permission to say "I don't know" to cut down on made-up answers, have it ask you clarifying questions on fuzzy tasks, and break big tasks into steps rather than one giant request. These aren't tricks — they're sensible moves for the right situations. With the basics plus these, you're well-equipped for almost any everyday AI task.