Sharpen Your Prompts: The Key Tips That Make You a Better Prompter

There is a stage in learning this that nobody warns you about. Your prompts are good now. Role, context, task, format, all four in place. The answers come back usable maybe four times out of five.

That fifth answer is the one worth talking about. It is not wrong because you forgot an ingredient. It is wrong because of something smaller, and the fixes are small too. This week is six of them.

Show One Example Instead of Describing Ten Rules

Describing the tone you want takes three sentences and still leaves room for interpretation. Pasting one thing you already wrote takes four seconds and removes all of it.

If you want your quote follow-ups to sound like you, find the best one you ever sent, paste it in, and write "match the voice and length of this." You have now communicated more about tone, sentence rhythm, sign-off style, and formality than any adjective could.

This works for anything you produce repeatedly. Invoices, job descriptions, customer updates, review responses. One good example beats a paragraph of instructions almost every time.

Say What You Do Not Want

Most people only describe the target. Naming the misses is faster, because you already know what the bad version looks like. You have read it.

"Do not open with a question." "Do not mention pricing." "Do not use the word 'excited.'" "No bullet points." Each of these is one line, and each one removes an entire category of answer you would have had to delete by hand.

That 27 percent figure is the one these tips are aimed at. Two thirds of AI output is already good enough to keep with light editing. The win available to you is not in the part that already works. It is in shrinking the quarter of answers that come back needing real surgery, and those almost always fail for reasons you could have prevented in the request.

Ask for Options, Not an Answer

"Write me a cancellation policy" gets you one policy, and you have no idea whether it is the good one.

"Give me three versions of a cancellation policy: one strict, one lenient, one in between. Two sentences each." gets you a decision instead of a draft. You read three, pick the closest, and refine from there. This takes the model the same amount of effort and saves you the round trip where you say "hmm, can you make it less harsh."

Three is usually the right number. Five is a reading assignment.

Let It Ask You Something First

Add this line to the end of a prompt for anything complicated: "Before you answer, ask me up to three questions that would change your response."

What comes back is a short list of the things you forgot to mention, which is exactly the list you needed. It is the closest thing to having a competent new hire say "quick question before I start." Answer the three, then let it run.

Use this for the big stuff: a policy document, a pricing page, an email to a client you are worried about. Skip it for a two-line task.

Put the Constraint That Matters Most at the End

Long prompts have a middle, and instructions buried in the middle get weighted less than instructions at the edges. If there is one rule you absolutely need followed, the safest place for it is the last line.

"Under eighty words" sitting in the middle of a paragraph of context will get missed sometimes. The same three words on their own line at the bottom will not.

This one sounds superstitious until you watch it work. Try the same prompt twice with the word limit in each position and count.

Name the Reader

Role tells the model who is writing. This one tells it who is reading, and readers shape writing more than writers do.

"A homeowner who has never hired a contractor before and is nervous about cost" produces different sentences than "a property manager who reviews eleven bids a week." Same job, same product, same writer. Different reader, different everything.

Most prompts skip this entirely, which leaves the model writing for a generic audience of nobody in particular.

What Is Not Worth Your Time

Some of what circulates as prompting advice does nothing measurable.

• Flattery and threats. "You are a world-class expert" and "this is very important to my career" are not doing the work people think they are doing. Specifics are.
• Elaborate formatting rituals. Wrapping everything in tags and headers helps with genuinely long prompts and does nothing for a paragraph.
• Chasing the perfect phrasing. There is no magic wording. There is only information the model has and information it does not.

Spend the effort on the six adjustments above instead. They are all just ways of handing over information you already had.

What to Do Next

Pick two of these, not six. The point is to build habits, and habits arrive one at a time.

1. Add an example to your most-used prompt. Find your best past version of that output and paste it in with "match this." Compare the result to last week's.
2. Add a "do not" line. Think of the thing you always end up deleting, and name it directly.
3. Take the AI Readiness Assessment. Ten questions, four minutes. It shows you which parts of your week are worth pointing these skills at, so the practice lands on work that actually costs you money.

Good prompting stops being about the words you use and starts being about the information you remember to include. An example instead of a description, a "do not" instead of a rewrite, a question answered before the work begins. None of it is clever, and all of it compounds. Part 4 covers what to do when one prompt is not enough and the real answer takes a conversation.