The six dimensions and what each is worth
The weights are fixed. They come from what experienced communication coaches weight when they evaluate a speaker, and the methodology page is the canonical reference for every number below.
| Dimension | Points | Sub-metrics | How it is scored |
|---|---|---|---|
| Structure | 250 | Logical Sequence 50, Transitions 50, Signposting 50, Opening Quality 50, Closing Quality 50 | LLM evaluation (transcript) |
| Conciseness | 200 | Word Choice 70, Word Economy 70, Sentence Length 60 | LLM evaluation (transcript) |
| Confidence | 150 | Hedging Frequency 50, Uptalk 50, Assertiveness 50 | Multimodal LLM (audio and text when available) |
| Pronunciation | 150 | Pronunciation Clarity 150 | Azure deterministic scoring plus LLM feedback |
| Filler Rate | 150 | Filler Frequency 150 | Deterministic formula plus LLM feedback |
| Pace | 100 | Words Per Minute 50, Pause Timing 50 | Deterministic formula plus LLM feedback |
Which points are measured and which are judged
Four hundred points are deterministic. Pronunciation, Filler Rate, and Pace run on measurable signals, so the same audio returns the same numbers on a second pass. Pronunciation Clarity is scaled from Azure's word-level accuracy. Filler Frequency is fillers per minute with an exponential decay. Pace combines active speaking speed, which excludes pauses, with pause timing, which is why the number can read higher than a plain words-per-minute count.
Four hundred and fifty points, Structure and Conciseness, come from an LLM reading the transcript. Those are judgment calls about meaning: bounded rather than exactly reproducible. Prompts and rubrics are versioned alongside the code, so any score traces back to the prompt version, model, and input audio behind it.
Confidence sits between the two at 150 points, using a multimodal model with the audio track inline when audio is available and a documented text-only fallback when it is not.
Why Structure carries the most
A well-paced speaker with a disorganized message loses the room faster than a slightly clumsy speaker with a tight structure. Structure and Conciseness together are 450 of the 1000 points, which puts nearly half the scale on the idea layer before delivery is considered at all. Pronunciation, filler control, confidence, and pace still matter. They multiply an answer that already has a shape.
The limits of a single score
The Index reads how an answer was organized and delivered. Whether the argument itself holds up sits outside what it measures.
One recording is a sample of one. A mock interview, a technical explanation, and a 60-second drill make different demands, so the useful comparison is your own trend inside the same practice type.
It is a coaching score for practice sessions. IELTS, TOEFL, and CEFR serve formal language assessment, a different job, and the full guide to the Index lays out where those boundaries sit.
A score that falls after a scoring update usually means the ruler got stricter. Compare recordings inside the same scoring era.
About the scores on the speaker pages
The speaker pages apply the same rubric to well-known speeches, and all 35 of them currently carry expert estimates produced from that rubric rather than a production pipeline run on the original audio. Each page says so above its breakdown. Treat them as worked examples of how the dimensions land against a real transcript. Michelle Obama, Brené Brown, Maya Angelou, and Bill Gates each show a different pattern under the same six headings.
Reading your own score
Start with your lowest dimension and pick one drill that targets it. The dimension breakdown usually matters more than the total: the total tells you where you are, the breakdown tells you what to do tomorrow.
Two dimensions have free standalone tools if you want a number before recording anything: the filler word counter for Filler Rate and the speaking speed test for the words-per-minute half of Pace. What makes someone sound articulate at work covers the behaviors the Index is built to surface, how we rate AI speech coaches applies the same scoring stance to other products, and our review of AI-assisted speech coaching is direct about what the evidence base does and does not yet support.
