Loading page content
Analyze a system prompt for the problems that make models appear to ignore instructions: contradictory rules, redundant restatement, missing output specifications, and the token cost you are paying on every single request.
Paste any AI system prompt to get an instant structural analysis. Detects personas, safety rails, capability grants, tool definitions, and complexity. Generates optimization tips. All analysis runs client-side, so nothing is sent to a server.
Analyzes system prompt structure using pattern matching for persona definitions, safety restrictions, capability grants, tool references, and section headers. Token estimate uses a 1.3x word multiplier. All processing is client-side.
When people say a model ignores instructions, the prompt usually contains two instructions that cannot both be satisfied. Be concise and be thorough. Never speculate and always give a recommendation. The model resolves the conflict somehow, and the resolution looks like disobedience. Finding those pairs is the highest-value thing an analysis pass does.
The cost dimension is the one people underweight. A system prompt is sent on every request, so a 2,000-token prompt at a million requests a month is two billion input tokens before any user has typed anything. Prompt caching mitigates it substantially, but the cheapest token is still the one you deleted.