The 6 Prompt Scoring Dimensions, Explained

The AI Effectiveness Score isn't a black box. It's built from 6 independently tracked dimensions, each detecting a defined prompt-structure signal. Here's what each one measures and how to change that signal.

1. Specificity (25% weight)

What it measures: How precisely you define what you want.

High-scoring signals:

Named technologies ("TypeScript", "PostgreSQL", "Next.js 16")

Defined input/output types ("takes a string, returns an array of objects")

Measurable criteria ("under 500ms response time")

Constraints and boundaries ("must work with Node 20+")

Quick win: Replace every instance of "some" and "stuff" with specific terms.

2. Context Engineering (20% weight)

What it measures: How well you provide background information.

High-scoring signals:

Role assignment ("You are a senior backend engineer")

Relevant code or error messages

Environment details

What you've already tried

Quick win: Always include the error message, stack trace, or relevant code when debugging.

3. Task Decomposition (15% weight)

What it measures: Whether you break complex work into steps.

High-scoring signals:

Numbered steps or bullet points

Sequential language ("First... Then... Finally...")

Focused scope (one concern per prompt)

Building on previous steps

Quick win: If your prompt is more than 3 sentences, add structure with numbers or bullets.

4. Iteration Quality (5% weight)

What it measures: How effectively you refine AI responses.

High-scoring signals:

Specific references to prior output ("The function works but...")

Clear keep/change instructions ("Keep the recursive approach, but fix the edge case")

Quoted sections from prior responses

Targeted follow-up questions

Quick win: Never say "try again." Instead, say exactly what was wrong and what to preserve.

5. Token Efficiency (15% weight)

What it measures: Conciseness without sacrificing clarity.

High-scoring signals:

High information density (no filler words)

No unnecessary preamble ("I was wondering if you could maybe...")

Appropriate length for the task

No restating information the AI already has

Quick win: Delete the first sentence of your prompt. If nothing is lost, it was filler.

6. Output Specification (20% weight)

What it measures: How clearly you define the response format.

High-scoring signals:

Format type ("Return as JSON", "Format as a markdown table")

Length constraints ("In 2-3 paragraphs")

Structure ("Include sections for: Overview, Implementation, Testing")

Tone and audience ("Explain for a non-technical PM")

Quick win: End every prompt with one sentence about what format you want the answer in.

---

Putting It Together

You do not need to score 100 on every dimension. Use the weakest dimensions as review prompts, then judge the actual model output against your task. A dimension change affects the composite only according to that dimension's published weight.

Start tracking with Measurably to see your personal dimension breakdown and get targeted coaching tips.