[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"blog-prompt-engineering-cheatsheet":3},{"id":4,"title":5,"author":6,"body":7,"date":440,"description":441,"extension":442,"image":443,"meta":444,"navigation":445,"path":446,"readingTime":447,"seo":448,"stem":449,"tags":450,"__hash__":455},"content/blog/prompt-engineering-cheatsheet.md","Prompt Engineering Cheatsheet – Technischer Leitfaden für LLMs","KIana",{"type":8,"value":9,"toc":405},"minimark",[10,15,24,27,30,34,37,48,53,59,61,65,69,82,86,97,101,112,116,124,126,130,134,185,189,203,208,216,218,222,226,237,241,247,251,263,265,269,273,281,285,296,298,302,306,317,321,327,329,333,336,338,342,345,359,362,364,368,371,373,377,380,386,389,400],[11,12,14],"h2",{"id":13},"einordnung-warum-ein-prompt-engineering-cheatsheet-notwendig-ist","Einordnung: Warum ein Prompt-Engineering-Cheatsheet notwendig ist",[16,17,18,19,23],"p",{},"Moderne Large Language Models (LLMs) sind leistungsfähig – aber nicht selbsterklärend. Ihre Qualität hängt direkt von der Struktur, Präzision und technischen Ausgestaltung der Prompts ab. Dieses Cheatsheet richtet sich an Entwickler, Architekten und technisch orientierte Entscheider, die ",[20,21,22],"strong",{},"verlässliche, reproduzierbare und skalierbare KI-Ergebnisse"," erzielen wollen.",[16,25,26],{},"Kein Marketing, keine Buzzwords – sondern klare Regeln.",[28,29],"hr",{},[11,31,33],{"id":32},"_1-grundstruktur-eines-professionellen-prompts","1. Grundstruktur eines professionellen Prompts",[16,35,36],{},"Ein robuster Prompt besteht aus klar getrennten Komponenten:",[38,39,44],"pre",{"className":40,"code":42,"language":43},[41],"language-text","[Ziel / Aufgabe]\n[Kontext]\n[Rolle des Modells]\n[Anweisungen]\n[Ausgabeformat]\n","text",[45,46,42],"code",{"__ignoreMap":47},"",[49,50,52],"h3",{"id":51},"beispiel","Beispiel",[38,54,57],{"className":55,"code":56,"language":43},[41],"Aufgabe: Analysiere den folgenden Text auf Risiken.\nKontext: Du arbeitest für ein B2B-SaaS-Unternehmen im Finanzumfeld.\nRolle: Du bist ein Senior Risk Analyst.\nAnweisung: Liste nur relevante Risiken.\nAusgabeformat: Bullet-Points, maximal 5 Punkte.\n",[45,58,56],{"__ignoreMap":47},[28,60],{},[11,62,64],{"id":63},"_2-prompt-typen-technisch-klassifiziert","2. Prompt-Typen (technisch klassifiziert)",[49,66,68],{"id":67},"zero-shot-prompt","Zero-Shot Prompt",[70,71,72,76,79],"ul",{},[73,74,75],"li",{},"Keine Beispiele",[73,77,78],{},"Schnell, aber fehleranfällig",[73,80,81],{},"Einsatz: einfache Aufgaben",[49,83,85],{"id":84},"one-shot-few-shot-prompt","One-Shot / Few-Shot Prompt",[70,87,88,91,94],{},[73,89,90],{},"1–5 Beispiele",[73,92,93],{},"Hohe Stabilität",[73,95,96],{},"Einsatz: Klassifikation, Parsing, strukturierte Ausgaben",[49,98,100],{"id":99},"system-prompt","System Prompt",[70,102,103,106,109],{},[73,104,105],{},"Definiert globale Regeln",[73,107,108],{},"Erzwingt Output-Format",[73,110,111],{},"Reduziert Halluzinationen",[49,113,115],{"id":114},"rollen-prompt","Rollen-Prompt",[70,117,118,121],{},[73,119,120],{},"Steuert Fachlichkeit und Tonalität",[73,122,123],{},"Erhöht Domänenpräzision",[28,125],{},[11,127,129],{"id":128},"_3-steuerung-der-modellparameter-entscheidend","3. Steuerung der Modellparameter (entscheidend!)",[49,131,133],{"id":132},"temperatur","Temperatur",[135,136,137,150],"table",{},[138,139,140],"thead",{},[141,142,143,147],"tr",{},[144,145,146],"th",{},"Use Case",[144,148,149],{},"Empfohlene Temperatur",[151,152,153,162,170,177],"tbody",{},[141,154,155,159],{},[156,157,158],"td",{},"Klassifikation",[156,160,161],{},"0.0 – 0.2",[141,163,164,167],{},[156,165,166],{},"Analyse",[156,168,169],{},"0.1 – 0.3",[141,171,172,175],{},[156,173,174],{},"Code",[156,176,161],{},[141,178,179,182],{},[156,180,181],{},"Kreativität",[156,183,184],{},"0.7 – 0.9",[49,186,188],{"id":187},"top-k-top-p","Top-K & Top-P",[70,190,191,197],{},[73,192,193,196],{},[20,194,195],{},"Top-K",": begrenzt Token-Auswahl numerisch",[73,198,199,202],{},[20,200,201],{},"Top-P",": begrenzt kumulative Wahrscheinlichkeit",[16,204,205],{},[20,206,207],{},"Empfehlung:",[70,209,210,213],{},[73,211,212],{},"Business-Use-Cases: Top-P 0.9–0.95",[73,214,215],{},"Kreative Tasks: Top-P 0.95–0.99",[28,217],{},[11,219,221],{"id":220},"_4-chain-of-thought-cot-wenn-logik-erforderlich-ist","4. Chain of Thought (CoT): Wenn Logik erforderlich ist",[49,223,225],{"id":224},"wann-cot-einsetzen","Wann CoT einsetzen?",[70,227,228,231,234],{},[73,229,230],{},"Berechnungen",[73,232,233],{},"Entscheidungslogik",[73,235,236],{},"Mehrstufige Analysen",[49,238,240],{"id":239},"cot-trigger","CoT-Trigger",[38,242,245],{"className":243,"code":244,"language":43},[41],"Denke Schritt für Schritt.\n",[45,246,244],{"__ignoreMap":47},[49,248,250],{"id":249},"best-practice","Best Practice",[70,252,253,256],{},[73,254,255],{},"Temperatur = 0",[73,257,258,259,262],{},"Antwort ",[20,260,261],{},"nach"," der Begründung",[28,264],{},[11,266,268],{"id":267},"_5-erweiterte-reasoning-techniken","5. Erweiterte Reasoning-Techniken",[49,270,272],{"id":271},"tree-of-thoughts-tot","Tree of Thoughts (ToT)",[70,274,275,278],{},[73,276,277],{},"Mehrere Denkpfade parallel",[73,279,280],{},"Einsatz: Strategie, Planung, Szenarien",[49,282,284],{"id":283},"react-reason-act","ReAct (Reason + Act)",[70,286,287,290,293],{},[73,288,289],{},"Kombination aus Denken und Aktionen",[73,291,292],{},"Grundlage für KI-Agenten",[73,294,295],{},"Einsatz: Recherche, Tool-Nutzung, Automatisierung",[28,297],{},[11,299,301],{"id":300},"_6-strukturierte-ausgaben-erzwingen","6. Strukturierte Ausgaben erzwingen",[49,303,305],{"id":304},"warum","Warum?",[70,307,308,311,314],{},[73,309,310],{},"Maschinenlesbarkeit",[73,312,313],{},"Weiterverarbeitung",[73,315,316],{},"Reduktion von Halluzinationen",[49,318,320],{"id":319},"json-beispiel","JSON-Beispiel",[38,322,325],{"className":323,"code":324,"language":43},[41],"Antworte ausschließlich im folgenden JSON-Schema:\n{\n  \"risiko\": string,\n  \"wahrscheinlichkeit\": \"hoch\" | \"mittel\" | \"niedrig\"\n}\n",[45,326,324],{"__ignoreMap":47},[28,328],{},[11,330,332],{"id":331},"_7-typische-prompt-anti-patterns","7. Typische Prompt-Anti-Patterns",[16,334,335],{},"❌ Mehrere Aufgaben in einem Prompt\n❌ Fehlender Kontext\n❌ Keine Definition des Outputs\n❌ Kreative Temperatur bei analytischen Aufgaben\n❌ Prompts nicht dokumentieren",[28,337],{},[11,339,341],{"id":340},"_8-prompt-engineering-als-engineering-disziplin","8. Prompt Engineering als Engineering-Disziplin",[16,343,344],{},"Professionelle Teams:",[70,346,347,350,353,356],{},[73,348,349],{},"versionieren Prompts",[73,351,352],{},"testen verschiedene Varianten",[73,354,355],{},"dokumentieren Ergebnisse",[73,357,358],{},"behandeln Prompts wie Code",[16,360,361],{},"Prompt Engineering ist kein Einmal-Setup, sondern ein iterativer Optimierungsprozess.",[28,363],{},[11,365,367],{"id":366},"fazit","Fazit",[16,369,370],{},"Dieses Cheatsheet zeigt: Gute KI-Ergebnisse sind kein Zufall. Sie sind das Resultat sauberer Prompt-Architektur, kontrollierter Parameter und systematischer Tests. Unternehmen, die Prompt Engineering technisch beherrschen, nutzen KI nicht experimentell – sondern produktiv.",[28,372],{},[11,374,376],{"id":375},"cta-technische-ki-potenzialanalyse-mit-vinspire","CTA: Technische KI-Potenzialanalyse mit Vinspire",[16,378,379],{},"Viele Organisationen nutzen KI ohne klare technische Standards.",[16,381,382,383],{},"👉 ",[20,384,385],{},"Vinspire analysiert Ihre bestehenden Prompts, Use Cases und KI-Workflows technisch fundiert.",[16,387,388],{},"Sie erhalten:",[70,390,391,394,397],{},[73,392,393],{},"eine strukturierte KI-Potenzialanalyse",[73,395,396],{},"konkrete Prompt- & Architektur-Empfehlungen",[73,398,399],{},"eine belastbare Roadmap für skalierbare KI-Systeme",[16,401,402],{},[20,403,404],{},"Jetzt technische KI-Potenzialanalyse anfragen – und KI professionell einsetzen.",{"title":47,"searchDepth":406,"depth":406,"links":407},2,[408,409,413,419,423,428,432,436,437,438,439],{"id":13,"depth":406,"text":14},{"id":32,"depth":406,"text":33,"children":410},[411],{"id":51,"depth":412,"text":52},3,{"id":63,"depth":406,"text":64,"children":414},[415,416,417,418],{"id":67,"depth":412,"text":68},{"id":84,"depth":412,"text":85},{"id":99,"depth":412,"text":100},{"id":114,"depth":412,"text":115},{"id":128,"depth":406,"text":129,"children":420},[421,422],{"id":132,"depth":412,"text":133},{"id":187,"depth":412,"text":188},{"id":220,"depth":406,"text":221,"children":424},[425,426,427],{"id":224,"depth":412,"text":225},{"id":239,"depth":412,"text":240},{"id":249,"depth":412,"text":250},{"id":267,"depth":406,"text":268,"children":429},[430,431],{"id":271,"depth":412,"text":272},{"id":283,"depth":412,"text":284},{"id":300,"depth":406,"text":301,"children":433},[434,435],{"id":304,"depth":412,"text":305},{"id":319,"depth":412,"text":320},{"id":331,"depth":406,"text":332},{"id":340,"depth":406,"text":341},{"id":366,"depth":406,"text":367},{"id":375,"depth":406,"text":376},"2026-02-09","Das ultimative Prompt-Engineering-Cheatsheet: Technische Regeln, Beispiele und Best Practices für präzise, reproduzierbare KI-Ergebnisse im Unternehmen.","md","/images/blog/prompt-engineering-cheatsheet.png",{},true,"/blog/prompt-engineering-cheatsheet",8,{"title":5,"description":441},"blog/prompt-engineering-cheatsheet",[451,452,453,454],"Prompt Engineering","LLMs","KI-Strategie","KI-Potenzial","dHInREsijaEWOBEw5P8UixUodtU8bOEudxAlzoXuxuQ"]