Methodology

The 6 Dimensions of
AI Effectiveness

The AI Effectiveness Score (AES) is a composite metric from 0-100 that measures structural signals in a prompt. It is calculated from six explicitly defined dimensions. The current validation establishes deterministic behavior on test fixtures; it does not prove causal improvement in model outcomes or human productivity.

How AES is Calculated

Baseline scoring runs entirely on your device, producing a 0-100 score across all six dimensions. Raw prompt text is not included in analytics or sync. If you explicitly invoke an AI feature, the requested text crosses the provider boundary described in our privacy policy. The dimension scores are weighted and combined into a single AES:

25%

Specificity

20%

Context Engineering

15%

Task Decomposition

5%

Iteration Quality

15%

Token Efficiency

20%

Output Specification

Weights are product-defined defaults used consistently by the current scorer. They are not presented as a validated causal ranking of model-output quality.

Specificity

25% weight

How precisely you define what you want. Vague prompts like 'help me with code' score low. Specific prompts that name the language, framework, function signature, and expected behavior score high.

Low Score Example

“Help me write a function”

High Score Example

“Write a TypeScript function called `parseCSV` that takes a string of comma-separated values with a header row and returns an array of objects where keys are the header names and values are the corresponding row data. Handle quoted fields containing commas.”

What We Look For

Names specific technologies, frameworks, or tools
Defines input/output types or formats
States measurable success criteria
Includes constraints and boundaries

Context Engineering

20% weight

How well you provide the background information the AI needs to give a useful response. This includes relevant code, error messages, environment details, and what you've already tried.

Low Score Example

“My app is broken, fix it”

High Score Example

“I'm getting a TypeScript error TS2345 in my Next.js 16 app when passing props to a Server Component. Here's the component code [code], the parent [code], and the full error. I've tried adding 'use client' but that breaks the data fetching.”

What We Look For

Includes relevant code snippets or error messages
Describes the environment and setup
Mentions what's already been tried
Provides business context or user requirements

Task Decomposition

15% weight

Whether a complex request is separated into manageable steps instead of asking the AI to solve everything at once. A well-decomposed prompt gives the model a narrower, more explicit unit of work.

Low Score Example

“Build me a full e-commerce site with payments, auth, inventory, and email notifications”

High Score Example

“Let's build this in steps. First, create the database schema for products with these fields: name, price, description, stock_count, category. Use Supabase and include the migration SQL.”

What We Look For

Breaks large tasks into sequential steps
Focuses each prompt on a single concern
References previous steps or builds on prior work
Uses numbered lists or clear ordering

Iteration Quality

5% weight

How effectively you refine and build on AI responses through follow-up prompts. Great iterators don't just say 'try again' — they explain what was wrong, what to keep, and what to change.

Low Score Example

“That's wrong, try again”

High Score Example

“The function works but has two issues: (1) it doesn't handle the edge case where the input array is empty — it should return an empty object, and (2) the variable naming is inconsistent — use camelCase throughout. Keep the recursive approach, that's good.”

What We Look For

Identifies specific issues in prior responses
States what to keep vs. what to change
Provides additional context discovered during iteration
Asks targeted follow-up questions

Token Efficiency

15% weight

How concisely you communicate your intent without sacrificing clarity. Token-efficient prompts avoid unnecessary filler, repetition, and irrelevant information while remaining complete enough to get good results.

Low Score Example

“I was wondering if you could maybe help me with something. So basically what I need is, like, a way to sort a list of numbers. I know there are many ways to do this but I'm not sure which one would be best. Can you help?”

High Score Example

“Implement quicksort in Python for a list of integers. Optimize for average-case performance. Include type hints.”

What We Look For

High information density per token
No filler words or unnecessary preamble
Avoids restating information the AI already has
Appropriate length — not too short, not padded

Output Specification

20% weight

How clearly you define what format, structure, and style you want in the AI's response. Without output specification, you get whatever the AI defaults to — which may not be what you need.

Low Score Example

“Explain React hooks”

High Score Example

“Explain the useState and useEffect hooks in React. Format as a comparison table with columns: Hook, Purpose, When to Use, Common Pitfalls. Keep each cell under 20 words. Add a code example for each after the table.”

What We Look For

Specifies output format (table, list, code, prose)
Defines length or detail level
Requests specific structure (headings, sections)
States the intended audience or tone

Ready to measure your AI effectiveness?

Try the local scorer now, or install the checksummed browser developer preview. Signed desktop packages remain gated on platform acceptance.

Browser PreviewGet Started FreeTry the Demo

© 2026 Measurably by AshlrAI. All rights reserved.