Prompt Structure Scoring Methodology

Purpose and disclaimer

Prompt Checker reports whether predefined structural cues are present in an English-language prompt. The score is a transparent rule-based structure check, not an AI judgment about the prompt's meaning or overall quality.

Version

The checker uses the unchanged Prompt Structure Score v1.0 rubric and the English ruleset en-1.1. Ruleset updates may improve detection while keeping the eight criteria, fixed points, and grade bands unchanged.

Prompt Structure Score

The score totals 100 points: Clear Goal 20, Context 15, Audience 10, Role 10, Output Format 15, Constraints 15, Examples 10, and Quality Check 5. Each criterion receives either all of its fixed points or zero points. There are no partial scores.

Criterion definitions

• Clear Goal: a recognizable action or intended outcome.
• Context: background, source material, or situational detail.
• Audience: the intended reader, viewer, customer, or user.
• Role: an explicit role, persona, or perspective for the AI.
• Output Format: the requested structure or response format.
• Constraints: limits, exclusions, or mandatory requirements.
• Examples: a supplied or requested sample or reference.
• Quality Check: an instruction to review or verify the final output.

How detection works

Each criterion uses local pattern rules to look for recognizable structural cues. A detected cue receives the fixed points assigned to that criterion. Detection does not mean that the checker understands the full meaning of the prompt, and a missing cue does not necessarily mean that the prompt is poor.

Detection examples

“Write an email for new subscribers” contains a goal, an audience, and an output format. “Review this contract” contains a goal but not a final-answer quality check. “Review your answer before finalizing” is treated as a quality-check instruction.

Grades

0–39 Needs significant improvement; 40–59 Basic structure; 60–79 Good structure; 80–100 Strong structure.

Language support

Prompt Checker v1.1 is designed for prompts written in English. Its structure score is based on predefined English wording and patterns. Prompts written mainly in other languages may receive incomplete or inaccurate results, even when they are well structured. Additional languages require separate detection rules and language-specific testing.

Edge cases and known limits

Short prompts, unusual grammar, domain-specific wording, mixed languages, and implied instructions can be missed. Labels without content do not count. A prohibition such as “Do not include an example” can count as a constraint without counting as an example.

False positives and false negatives

A recognizable phrase can occasionally be detected even when it has a different purpose, and a well-structured instruction can be missed when it uses wording outside the rules. The evidence shown with each result identifies the cue and rule that produced the score.

What the score does not measure

It does not evaluate factual accuracy, creativity, domain correctness, safety, or the final response quality of any AI model. It does not predict whether a specific AI model will produce a good answer.

Change log

Version 1.1 adds explicit English-language disclosure, language-scoped rules, broader English cues, structural-label recognition, numeric constraints, negation handling, and clearer evidence. The v1.0 score weights and grade bands remain unchanged.