Natural Language Processing (NLP) Tool for Analyzing User Input Prompts
Analyze the quality of your LLM system prompt according to Anthropic and OpenAI guidelines: role, output format, few-shot examples, edge cases, safety guardrails. Score 0-100 + issue list + rewrite suggestion. 100% client-side, no data sent.
Federico Calò's System Prompt
How it works
Paste the prompt
Please paste your text into the box below and I'll translate it to English:
Analysis of the 7 dimensions
The tool checks for role presence, optimal length and clarity of the text, format examples, edge cases, safety guards, and formatting guidelines.
Score + rewriting
Receive a score of 0-100, the list of issues with severity levels and (if below 70) an outline for rewriting with placeholders.
Come utilizzare AI System Prompt Analyzer
Copy system prompt
Paste your agent’s or chatbot's system prompt (max 8000 characters) into the provided area: it defines model behavior and is usually passed as a system message before conversation starts.
Start analysis and read score
Click “Analyze Prompt” to get a score from 0-100 and evaluation on 7 dimensions: role clarity, output format, examples, edge cases handling, security, length, and structure.
Apply suggested rewrites
Tool proposes an improved prompt with main issues corrected; compare original version against the one proposed by tool, adapt changes to your use case, and test updated prompt again.
Suggerimenti
- Always start system prompt defining the role precisely (e.g., "You’re a specialized Italian tax law assistant for SMEs"). A specific role reduces hallucinations and improves coherence in responses.
- Add at least 2-3 concrete input/output examples (few-shot) for complex or non-standard tasks: these are most effective way to communicate desired format and tone to model.
- Separate system prompt into sections with titles (ROLE, RULES, OUTPUT FORMAT, EXAMPLES) for better readability and maintainability; a structured prompt is also easier to update over time.
Domande frequenti
What criteria does the score of a prompt rely on?
Analysis is based on 7 dimensions: role clarity (who model is), task specificity, output format structure vs. free-form, presence of few-shot examples, edge cases handling, security guardrails, and appropriate length; each dimension contributes to final score from 0-100.
Is analysis performed locally or server-side?
100% client-side: Analysis occurs in browser through rules, heuristics, regex without sending prompt text to any external servers; your prompts remain completely private.
Does a high score guarantee better model responses?
Score measures structural quality of the prompt according to best practices for prompt engineering, not actual performance of the model used or complexity of task at hand: well-structured prompt statistically increases coherence and accuracy but final result depends on both factors mentioned above.
What are most common signs of a poorly structured prompt?
Most frequent issues include absence of defined role ("You're an assistant" without specificity), lack of instructions for output format, no concrete examples provided for complex tasks, prompts too long with contradictory instructions, and absence of guardrails on sensitive topics or out-of-scope subjects.