> ## Documentation Index > Fetch the complete documentation index at: https://docs.mavel.ai/llms.txt > Use this file to discover all available pages before exploring further. # Metrics overview > How Mavel's four core brand metrics (Visibility, Position, Sentiment, and Share of Voice) are computed from every day's captured answers, and how to read them together. Every day, Mavel runs your prompt set against your enabled AI surfaces, captures the full answers, and rolls them up into four metrics. Each metric is computed from the same daily fact rows, so they can be sliced by the same filters and compared side by side for every brand you track. | Metric | Question it answers | | ------------------ | -------------------------------------------------------------- | | **Visibility** | In what share of answers does my brand appear? | | **Position** | When I appear, how early in the answer? | | **Sentiment** | How does the answer characterize my brand? | | **Share of Voice** | How much of all tracked-brand mentions in my category is mine? | ## How the numbers are built Each prompt runs once per enabled surface per day. The raw answer is stored in full, then a classifier pipeline extracts brand mentions, their position within the answer, and a sentiment score for each mention span. Those per-mention facts roll up into daily brand-level aggregates: the grain the dashboard reads from. This means every metric descends from the same underlying measurements. A brand's visibility ratio, its position average, its sentiment index, and its mention share are all computed from the same set of captured answers for the same date range and filter. ## How to read them as a set The four metrics tell different parts of the same story, and gaps between them are often the most informative signal: * **Visibility without Position data** means you appear, but position ranks haven't been recorded for those mentions, typically answers captured before position tracking was active for that surface. * **High Visibility but negative Sentiment** is a warning sign: AI engines are mentioning you, but not favorably. * **Rising Visibility with flat Share of Voice** means the category is growing: competitors are growing with you. * **Improving Position with stable Sentiment** is the compounding pattern you want: showing up earlier without the narrative degrading. Read the four together on the Overview rather than in isolation. ## Filters that apply to all four Every metric can be filtered by: * **Date range**: Last 7, 30, or 90 days (default: 7). Each range also shows period-over-period deltas against the equal-length window immediately before it. * **AI surface**: filter to a single surface (e.g. ChatGPT only) to see where you are strong or absent. * **Topic**: a group of prompts (e.g. "pricing questions") you have defined in [Topics and tags](/setup/topics-and-tags). * **Tag**: a cross-cutting label you have applied to prompts. Filter by topic and surface before reading the blended number. A strong overall visibility can hide very weak numbers on the surface or prompt set that matters most. ## Non-determinism and trend reading AI answers are not deterministic. The same prompt sent twice can produce different answers. Daily runs turn that noise into measurable trends. Judge weeks, not single days. A one-day spike or dip is an anecdote until the trend over 7-30 days confirms it. Deltas compare the selected window against the equal-length period immediately before it, so a 7-day view compares this week against last week, and a 30-day view compares this month against the prior month. ## Where to go next * [Visibility](/results/visibility): exact definition and how to use it * [Position](/results/position): what rank 1.0 versus 4.5 actually means * [Sentiment](/results/sentiment): how the 0-100 index is built from LLM scoring * [Share of Voice](/results/share-of-voice): why the competitor set matters so much