Stop Copy-Pasting the Answer
You asked for a blog post. You got a blog post. And then you spent twenty minutes cutting filler, fixing the tone, moving the conclusion to the top, removing every sentence that began with "In today's fast-paced world", and replacing the word "leverage" four times.
The output was not wrong. It was not usable either. And tomorrow you will do exactly the same thing, because you keep asking for a finished product and receiving a first draft in costume.
The problem is not the model. The problem is that you asked for the answer instead of asking for something you could work with — and those are different requests.
Four prompts.
1. Tell it what you are going to do with the output
I need a first draft of [X]. I will edit it before publishing.
Knowing that, do not polish. Do not write transitions between sections. Do not add an introduction that summarises the piece — I will write my own. Do not hedge with "it is worth noting" or "it is important to remember".
Give me the raw thinking, section by section, and I will shape it.
The single most expensive word in a prompt is "write me". It tells the model to produce a finished artefact, and finished means polished — which means filler, hedging, and every connective phrase it has ever seen in published text.
When you say "I will edit this", the contract changes. You get material instead of performance. And material is what you actually needed.
2. Specify the shape, not the quality
Structure: [heading + 2-3 paragraphs per section, no bullet lists, no sub-headings inside sections].
Sentence length: max 20 words average. If a sentence needs a semicolon, it is two sentences.
Open each section with the claim, not with context. The reader already knows the topic — they clicked on it.
"Make it concise" is not a shape. "No paragraph over four sentences" is a shape. "Make it engaging" is not a shape. "Open with the claim, not the setup" is a shape.
Quality adjectives describe a destination without drawing the route. Structural constraints draw the route, and the destination takes care of itself.
3. Name what you will delete
Do not include: - Any sentence that could appear in a different article on a different topic without changing a word. - Any paragraph that restates what the previous paragraph already said. - The phrase "In this article, we will explore".
If you are unsure whether a sentence is specific enough, remove it.
This prompt works because deletion is cheaper than generation — for the model and for you.
Every sentence you name here is one you have deleted before. You already have this list. You have just never given it to the model, so it keeps generating the things you keep removing, and you keep paying the tax.
4. Ask for the version before the version
Give me two options for each section: A — the safe version (conventional structure, broad appeal). B — the version you would write if the only goal were to keep someone reading.
I will pick from each. Do not blend them.
This is the prompt that stops you from accepting mediocre output out of fatigue.
When you get one version, you either accept it or rewrite it, and rewriting is work. When you get two, you compare — and comparison is easy. The B version is often too aggressive to use as-is, but it shows you where the A version was playing it safe, and that is the edit you were going to make anyway.
Why editing should be choosing, not rewriting
The fastest workflow is not the one that produces a perfect first draft. It is the one where every edit is a decision between options rather than a blank-page rewrite.
A perfect first draft is a fantasy. A first draft you can scan in thirty seconds and fix in five minutes is a system. These four prompts build the system by making the model give you material that is honest about being material, structured the way you think, clean of the things you always remove, and paired with the version it was too careful to offer first.
Stop asking for the answer. Ask for the draft you can actually use.
The four prompts are in the carousel on Instagram.
Want the calm version of AI news like this, once a week? Subscribe to the Sharp AI Hub newsletter →