[{"data":1,"prerenderedAt":19},["ShallowReactive",2],{"post-en-bessere-ki-prompts":3},{"slug":4,"title":5,"description":6,"date":7,"dateFmt":8,"minutes":9,"tags":10,"related":14,"image":15,"imageOg":16,"imageAlt":17,"html":18},"bessere-ki-prompts","Writing better AI prompts: how to get usable results out of ChatGPT & co.","Writing better AI prompts — why most of them lead to mediocre results, the four building blocks of a good prompt, and concrete before-and-after examples.","2026-06-21","June 21, 2026",3,[11,12,13],"prompts","chatgpt","productivity","ki-fuer-marketing","\u002Fblog\u002Fbessere-ki-prompts-cover.webp","\u002Fblog\u002Fbessere-ki-prompts-cover.jpg","Stacked building blocks become a beam of light flowing into an AI core; on the right, an orderly grid of results emerges.","\u003Cp>"The AI only gives mediocre results." We hear this a lot — and almost always it's not the model's fault, it's the prompt's. Good AI output isn't luck and isn't a secret formula; it's a question of input. Here's the honest, practical guide — no "10 magic prompts" to memorize.\u003C\u002Fp>\n\u003Ch2>Why most prompts produce mediocre results\u003C\u002Fh2>\n\u003Cp>The typical prompt reads: "Write me a text about our new product." The result is then just as arbitrary as the instruction. The AI isn't a mind reader — it fills every gap you leave open with average. The more you leave to it, the more generic the answer gets. So the lever isn't a better model, it's a clearer brief.\u003C\u002Fp>\n\u003Ch2>The four building blocks of a good prompt\u003C\u002Fh2>\n\u003Cp>\u003Cimg src=\"\u002Fblog\u002Fbessere-ki-prompts-anatomie.webp\" alt=\"Five glowing layers float one above another and assemble into a structured whole — the building blocks of a good prompt.\">\u003C\u002Fp>\n\u003Cp>A good prompt answers four questions before the AI starts:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>Role\u003C\u002Fstrong> — from whose perspective? "You are an experienced sales manager in mechanical engineering." That sets tone and technical depth.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Task\u003C\u002Fstrong> — what exactly should be produced? Not "something about X", but "a follow-up email to a customer who hasn't replied in two weeks".\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Context\u003C\u002Fstrong> — what does the AI need to know? Audience, background, relevant facts. Without context, it guesses.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Format\u003C\u002Fstrong> — what should the result look like? "Max 100 words, friendly, with a clear call to action, no small talk."\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>These four — role, task, context, format — are the scaffold. For 90% of cases, you need nothing more.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Put together\u003C\u002Fstrong>, instead of a one-liner you get a brief the AI can actually work with:\u003C\u002Fp>\n\u003Cblockquote>\n\u003Cp>"You are an experienced sales manager in mechanical engineering. Write a follow-up email to a customer who hasn't replied to our quote in two weeks. They had previously signalled interest in delivery by Q3. Max 100 words, friendly, with one concrete question at the end, no small talk."\u003C\u002Fp>\n\u003C\u002Fblockquote>\n\u003Cp>The difference in the result isn't gradual — it's the difference between "I can't send that" and "small tweak, done". And the best part: you don't have to polish every word. Once the scaffold is there, the AI reliably delivers something usable.\u003C\u002Fp>\n\u003Ch2>Four levers that work immediately\u003C\u002Fh2>\n\u003Cp>\u003Cstrong>1. Concrete instead of vague.\u003C\u002Fstrong>\nBefore: "Write a product description."\nAfter: "Write a product description for an ergonomic office chair, audience: home-office users, three sentences, emphasize back health, no advertising superlatives."\u003C\u002Fp>\n\u003Cp>\u003Cstrong>2. Give examples.\u003C\u002Fstrong> If you want a specific style, show it. A single example ("this is how it should sound: …") beats three adjectives.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>3. Think in steps.\u003C\u002Fstrong> For complex tasks: let the AI outline first, then write. "First create a bullet-point outline, wait for my OK, then write the text." That way you correct early instead of discarding everything at the end.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>4. Say what you don't want.\u003C\u002Fstrong> Negative constraints often work better than positive ones: "no clichés, no bullet lists, not in the first-person plural" narrows the result faster than any lengthy style description. The AI tends toward a few typical patterns — if those bother you, just name them.\u003C\u002Fp>\n\u003Ch2>What the best prompt doesn't solve\u003C\u002Fh2>\n\u003Cp>Two honest limits. \u003Cstrong>First:\u003C\u002Fstrong> no prompt replaces missing knowledge. If the AI doesn't know your products, prices and processes, it guesses — no matter how good the prompt. That calls for \u003Ca href=\"\u002Fen\u002Fblog\u002Fchatgpt-im-unternehmen\">AI with your knowledge\u003C\u002Fa>, not better wording. \u003Cstrong>Second:\u003C\u002Fstrong> for tasks you repeat daily, a perfect one-off prompt isn't worth it — a saved template the whole team uses is. Prompt-writing is a tool, not an end in itself.\u003C\u002Fp>\n\u003Ch2>You don't have to rebuild the scaffold every time\u003C\u002Fh2>\n\u003Cp>The good news: you don't have to assemble this structure yourself each time. Our free \u003Ca href=\"\u002Fen\u002Fprompt-generator\">Prompt-Generator\u003C\u002Fa> does exactly that — it turns your idea into a clearly structured prompt with role, task, context and format that you can drop straight into ChatGPT, Claude & co. No login, no upload.\u003C\u002Fp>\n\u003Cp>And if "better prompts" should become a real process — AI that knows your knowledge and runs inside your workflows — \u003Ca href=\"\u002Fen\u002Fkontakt\">talk to us\u003C\u002Fa>. You'll get an honest assessment of where the effort actually pays off.\u003C\u002Fp>\n",1782055540746]