How the AI Prompt Optimizer Analyzes a Prompt (Heuristics, Not an LLM)
AnyTool’s prompt optimizer takes an existing rough prompt the user pastes and analyses it with rule-based heuristics — keyword and structure pattern-matching — running entirely in the browser with no AI, LLM, server or network call. It scores the prompt against eight best-practice dimensions drawn from prompt-engineering research: Task clarity (does it start with a clear action verb such as write, summarize, analyze?), Role / persona (cues like “you are” or “act as”), Context / background, Output format (cues like “bullet points”, “table”, “JSON” or a word count), Audience, Constraints (do/don’t rules and limits), Examples (few-shot), and Specificity (penalising vague filler words and very short input). Each dimension is marked Good, Weak or Missing with a short reason and a concrete fix, and the dimensions are weighted and summed into a score out of 100 with a grade band. Because the analysis is heuristic, the tool is explicit that it does not understand the prompt or judge its quality the way a human or an AI would — the score is a checklist guide, not a verdict.
- Rule-based heuristics — no AI, LLM, server or network call
- Eight dimensions: Task, Role, Context, Format, Audience, Constraints, Examples, Specificity
- Each marked Good / Weak / Missing with a reason and a concrete fix
- Weighted checklist tally produces a transparent score out of 100
- Honest: it does not understand or grade the prompt like an AI
