Konkrete Details hinzufügen

Konkrete Details dienen als Anker

Wortanzahl als konkrete Vorgabe

import anthropic

client = anthropic.Anthropic(api_key='sk-ant-your-key-here')

response = client.messages.create(
    model='claude-opus-4-5',
    max_tokens=200,
    messages=[{
        'role': 'user',
        'content': (
            'Write a product description for a noise-cancelling travel pillow in exactly 60 words. '
            'After the description, output the word count in parentheses on a new line.'
        )
    }]
)
print(response.content[0].text)

Zielgruppe als konkretes Detail

import openai

client = openai.OpenAI(api_key='sk-your-key-here')

concrete_audience_prompt = (
    'Explain what a REST API is.\n\n'
    'Audience: a non-technical product manager who uses Salesforce daily '
    'and understands the concept of a form submission on a website. '
    'They will use this explanation to brief their engineering team tomorrow. '
    'Use one real-world analogy. Max 100 words.'
)

response = client.chat.completions.create(
    model='gpt-4o',
    messages=[{'role': 'user', 'content': concrete_audience_prompt}]
)
print(response.choices[0].message.content)

Benannte Entitäten statt Verallgemeinerungen

import anthropic

client = anthropic.Anthropic(api_key='sk-ant-your-key-here')

response = client.messages.create(
    model='claude-opus-4-5',
    max_tokens=400,
    messages=[{
        'role': 'user',
        'content': (
            'Compare FastAPI vs Flask vs Django REST Framework '
            'for building a production API that will serve a React frontend. '
            'Context: 2-person startup team, Python 3.11, needs async support, '
            'expects < 50 ms response time, and will deploy on AWS Lambda. '
            'Format: 3-column markdown table with rows: setup complexity, '
            'async support, performance, learning curve, community size.'
        )
    }]
)
print(response.content[0].text)

Quantitative Vorgaben

import openai

client = openai.OpenAI(api_key='sk-your-key-here')

response = client.chat.completions.create(
    model='gpt-4o',
    messages=[{
        'role': 'user',
        'content': (
            'Generate exactly 8 onboarding email subject lines for a B2B SaaS product. '
            'Requirements:\n'
            '- Exactly 8 lines, no more, no fewer\n'
            '- Each subject line between 40 and 60 characters\n'
            '- 4 must emphasize value, 4 must create urgency\n'
            '- Label each line: [VALUE] or [URGENCY]\n'
            '- No emojis\n'
            '- No duplicate words across all 8 lines'
        )
    }]
)
print(response.choices[0].message.content)

Frist und zeitbezogener Kontext

import anthropic

client = anthropic.Anthropic(api_key='sk-ant-your-key-here')

response = client.messages.create(
    model='claude-opus-4-5',
    max_tokens=256,
    messages=[{
        'role': 'user',
        'content': (
            'Write a project status update for our AI chatbot project. '
            'This will be read aloud at a 5-minute Monday morning stand-up. '
            'The audience has zero technical background. '
            'Talking time: max 2 minutes (approximately 300 words). '
            'Cover: what we shipped last week, one current blocker, one upcoming milestone. '
            'Tone: confident and factual. No jargon.'
        )
    }]
)
print(response.content[0].text)

Konkrete Beispiele im Prompt

import openai

client = openai.OpenAI(api_key='sk-your-key-here')

prompt = '''
Write 3 feature announcement tweets for a new developer tool.

Here is an example of the style I want:
"We just shipped inline type errors in the editor. No more tab-switching to find 
what broke. Your flow stays unbroken. Try it now: [link]"

Characteristics of this style:
- Opens with what shipped, not with excitement words
- Describes the user benefit in the second sentence
- Ends with a CTA
- No hashtags. No emojis. Under 200 characters.

New feature: autocomplete that learns from your codebase.
'''

response = client.chat.completions.create(
    model='gpt-4o',
    messages=[{'role': 'user', 'content': prompt}]
)
print(response.choices[0].message.content)

Fachspezifischer Wortschatz

import anthropic

client = anthropic.Anthropic(api_key='sk-ant-your-key-here')

# Generic prompt vs domain-vocabulary prompt
generic = 'Explain caching strategies.'

domain_anchored = (
    'Compare write-through, write-behind, and cache-aside patterns '
    'for a microservices architecture using Redis 7.x as the cache layer and '
    'PostgreSQL 15 as the source of truth. '
    'Focus on consistency guarantees, latency tradeoffs, and failure modes. '
    'Audience: senior backend engineers familiar with CAP theorem.'
)

response = client.messages.create(
    model='claude-opus-4-5',
    max_tokens=400,
    messages=[{'role': 'user', 'content': domain_anchored}]
)
print(response.content[0].text)

Vorgaben dazu, was ausgeschlossen werden soll

import openai

client = openai.OpenAI(api_key='sk-your-key-here')

response = client.chat.completions.create(
    model='gpt-4o',
    messages=[{
        'role': 'user',
        'content': (
            'Suggest 5 ways to speed up a slow Python script. '
            'EXCLUDE:\n'
            '- Suggestions requiring third-party libraries (stdlib only)\n'
            '- Suggestions about hardware upgrades\n'
            '- Rewrites of the entire script\n'
            '- Suggestions that take more than 30 minutes to implement\n'
            'Each suggestion: one sentence describing what to do, one sentence on expected speedup.'
        )
    }]
)
print(response.choices[0].message.content)

Alles zusammenführen

import anthropic

client = anthropic.Anthropic(api_key='sk-ant-your-key-here')

concrete_prompt = (
    'Write a job description for a Senior ML Engineer position '
    'at a Series B AI startup in London. '
    'Target candidate: 5+ years ML experience, strong Python, familiar with LLMs. '
    'Length: 350-400 words. '
    'Structure:\n'
    '1. 3-sentence company intro (no fluff — focus on product and mission)\n'
    '2. Responsibilities: exactly 8 bullet points\n'
    '3. Requirements: exactly 5 bullet points\n'
    '4. Nice-to-haves: exactly 3 bullet points\n'
    'EXCLUDE: phrases like "fast-paced", "passionate", "ninja", "rockstar", "self-starter".'
)

response = client.messages.create(
    model='claude-opus-4-5',
    max_tokens=600,
    messages=[{'role': 'user', 'content': concrete_prompt}]
)
print(response.content[0].text)

Die Gewohnheit, Prompts anzureichern

# Prompt enrichment in code form: before/after comparison
before = 'Give me ideas for my presentation.'

after = (
    'Generate 6 opening hook ideas for a 15-minute conference talk '
    'titled "Why Your AI Prompts Are Failing You" at PyCon 2025. '
    'Audience: Python developers, mostly backend, some ML experience. '
    'Each hook: 2 sentences max. '
    'Mix: 2 provocative statistics, 2 counterintuitive questions, 2 short stories. '
    'EXCLUDE: "Did you know..." openers and rhetorical questions about the future of AI.'
)

print('BEFORE:', before)
print()
print('AFTER:', after)

Wissensüberprüfung

Konkrete Details ergänzen – Zusammenfassung