Prompting Without the Hype: Five Patterns for Useful AI Answers

Give it a role and a reader

A blank, generic question reliably gets a blank, generic answer back — that's not the model being unhelpful, it's the model correctly having nothing specific to aim at. Tell it who it is and who it's writing for instead: "You are a hiring manager at a Kolkata IT firm. Rewrite this email so a busy recruiter reads it in ten seconds and immediately understands why they should reply." The role you assign sets the vocabulary and the level of assumed knowledge; the reader you name sets the length, the tone, and what can safely be left unsaid.

This single line of setup does more useful work than any amount of clever phrasing or trick wording ever will. You're not tricking the model into performing better — you're simply removing the guesswork it would otherwise have to fill in on its own, usually with something generic, safe, and forgettable.

Try the same request with and without a role and reader specified, side by side, and the difference is immediate and obvious. Without it, you get a serviceable but flavourless answer that could have been written for anyone, about anything. With it, you get something that actually sounds like it was written for the specific situation you're in — because, in the way that matters, it genuinely was.

Show one example of "good"

Models copy patterns from an example far more reliably than they follow abstract written instructions describing what you want. If you want the output in a particular shape or style, don't try to describe that shape in words — paste one real example of it and simply say "match this format." This is faster to write and dramatically more reliable in what comes back.

Want tight, three-bullet summaries instead of rambling paragraphs? Show it one perfectly-written bullet as your model, not three paragraphs explaining what a good bullet is. Want plain, warm English that a Bengali or Hindi speaker reads comfortably on the first pass, without stumbling on idiom? Show it one sentence written exactly in that voice, and let it extend the pattern. One concrete example reliably beats a full paragraph of description, every single time we've tested it in a live session.

Keep a small personal library of these examples as you find ones that work — a good bullet format, a good tone for a client email, a good structure for a weekly report, a good way to phrase a difficult message to a colleague. Reusing them takes seconds and consistently produces better first drafts than starting from a blank instruction each time, and the library only gets more useful the longer you keep adding to it.

Ask for the thinking, then the answer

For anything that involves real reasoning — a plan with trade-offs, a comparison between two options, a decision with consequences — add one extra instruction: "think it through step by step, then give the final answer." This single addition catches a surprising amount of shaky logic before it ever reaches your boss, your client, or your own decision, because you get to see the reasoning laid out and can spot the exact point where it goes wrong.

And when the first reply back is roughly eighty percent right, resist the urge to throw it away and start over from scratch — that's the single most common mistake beginners make. Say specifically what's wrong: "good, but make it shorter and drop the jargon in the second paragraph." AI is fundamentally a conversation, not a vending machine that either gives you exactly the right thing on the first coin or nothing useful at all. In practice, the best result is usually your third message in the exchange, not your first.

This iterative habit is also the fastest way to actually get better at prompting yourself, because each correction teaches you something concrete about what was missing from your original request — which you'll then remember to include upfront next time. After a few weeks of this, most people find their first message already contains the role, the example, and the constraints, because the pattern of what usually needs fixing has become second nature.

One more habit worth building alongside this: when a plan or comparison genuinely matters, ask the model to also list what could go wrong with its own recommendation, or what a smart, sceptical critic in the room would object to first. This isn't about distrust for its own sake — it's a fast, cheap way to surface the weak points in an argument before you present it to someone else, the same way a good colleague would push back on a plan in a meeting before it ever left the room and became something you had to defend in public, under pressure, with no time left to fix it.

The five patterns, on a sticky note

Keep these five within easy reach — on a sticky note by your screen, in a pinned note on your phone, wherever you'll actually see them — and you'll rarely get a genuinely useless answer out of any AI tool again:

• Role + reader — "You are X, writing for Y" — sets the vocabulary and the tone in one short line.
• Example of good — paste one real example and say "match this" instead of describing the shape you want in words.
• Constraints — length, format, and what to specifically avoid, stated in plain, unambiguous words.
• Think-then-answer — for anything that involves genuine reasoning, ask to see the steps before the conclusion.
• Iterate — correct the eighty-percent draft with specific feedback instead of discarding it and starting again.