What Is an AI Effectiveness Score (AES)?

Everyone uses AI tools now. But how do you know if you're using them well?

The AI Effectiveness Score (AES) is a product-defined composite metric from 0-100 that measures structural signals across six dimensions.

The 6 Dimensions

1. Specificity (25%) — How precisely you define what you want. "Help me with code" scores low. "Write a TypeScript function called parseCSV that handles quoted fields" scores high.

2. Context Engineering (20%) — How well you provide background. Error messages, environment details, what you've already tried.

3. Task Decomposition (15%) — Whether you break complex problems into steps instead of asking the AI to solve everything at once.

4. Iteration Quality (5%) — How effectively you refine AI responses. "Try again" scores low. "The function works but doesn't handle empty arrays — return an empty object for that case" scores high.

5. Token Efficiency (15%) — Conciseness without sacrificing clarity. No filler words, no unnecessary preamble.

6. Output Specification (20%) — Whether you define the format, structure, and style of the response you want.

Why Does It Matter?

Clear context, constraints, and output requirements can make an AI request easier to execute. AES makes those structural habits visible; it is not a guarantee of response quality.

A consistent heuristic can make changes in prompt structure easier to see. AES gives you a number, a breakdown, and coaching cues to consider; the model response remains the real test.

How Scoring Works

Baseline scoring happens on your device using deterministic heuristic analysis. The current hardened browser developer preview keeps automatic measurement local and excludes prompt bodies from score sync. A direct checksummed ZIP is available for unpacked installation; it is not a browser-store release. Separately invoked web AI features have an explicit provider boundary described in the privacy policy.

What's a Good Score?

80+: The scorer detected most of its preferred structural signals.

60-79: Several structural signals are present, with identifiable gaps.

40-59: The request leaves multiple dimensions underspecified.

Below 40: The scorer found few of its expected context, structure, or output cues.

These bands describe the current scoring model, not validated tiers of human skill or guaranteed model outcomes.