How we measure

Real answers, asked like a buyer

Each tracked question is asked in the language of its market and from that market’s location, through each engine’s own interface or API. Google AI Overviews and AI Mode are collected from search results in that country.

Every answer is archived with its citations and the background searches the engine ran, so each mention can be traced to the answer it came from.

Within each plan, every question is asked on the main engines at least once a week. Claude and Grok cost many times more per answer, so they are sampled: they ask the questions in rotation with what the plan leaves.

Mentions, positions and citations

We match your brand and competitors in the answer text, including spellings in other scripts. Visibility is the share of answers that mention a brand; share of voice is its share of all tracked mentions; citation rate is the share of answers that cite its site.

Positions are reported as distributions: how often a brand is mentioned first and how often it is in the top three. We never average positions into an “AI rank”: answers differ in length and engines in format, so the average has no meaning.

Uncertainty

Every rate has a 95% interval. Over a period we resample at the question level (cluster bootstrap), because answers to the same question are not independent; for a single day we use the Wilson interval.

Week-over-week changes are tested with a two-proportion test and corrected for multiple testing with the Benjamini–Hochberg procedure (q ≤ 0.10). With dozens of brand × engine comparisons, some would look significant by chance; the correction keeps those out of your alerts.

Language drift

Engines often research a local question in English. We record the language of each background search and the country of each cited source, so you can see when English pages decide answers in a non-English market.

What we do not claim

We do not claim that llms.txt or schema markup cause citations: we found no evidence that they do, and we label them as hygiene. A pitch audit or free check is a single snapshot, one answer per question and engine; trends need ongoing tracking.