How Sondar measures AI visibility

Sondar asks each AI engine the questions your buyers ask, grouped by intent. A full Panel holds between 50 and 100 Prompts written for one vertical, and it is versioned: a Panel that changes gets a new version, never a quiet edit. Only the number of Prompts per intent is ever published, never the Prompt texts.

Every Prompt goes to ChatGPT, Perplexity, Gemini and Google AI Overviews. Each answer is stored verbatim with its timestamp and citations, so every published figure can be traced back to the exact response behind it.

Each answer is classified by how it mentions a business (recommended, listed, absent and negative) and whether it cites the business's own website. Each class carries a disclosed value, citation presence carries a disclosed weight, and the Visibility Score is the average of those values. The Weight Version and its full table are published here once Gate Zero has been decided, and every later change appears in a public changelog.

AI engines are non-deterministic, so the Panel is run repeatedly across one gate week and the spread between those Runs becomes a Variance Band per Engine and overall. The number of Runs is published with the first Methodology Version. Bands use the mean plus or minus two sample standard deviations across Runs. Only a move larger than the band is reported as real; a move inside it is measurement noise.

The gate has one pre-registered pass rule, written down before any Run: Band half-width must be smaller than the minimum improvement the retainer promises to detect.

Once the gate is decided, this page fills in from the published Methodology Version: Prompt count per intent, run count and dates, Engines covered, the weight table, the bands per Engine and overall, the API-versus-Consumer agreement summary and the publication date. Until then you can request a free scan and receive verbatim excerpts of what the engines say about you today.