Skip to main content
Mentioned is the share of answers that name your brand at any strength, from Weak up. Recommended is the share on the Leading or Recommended rung, the answers that tell the reader to pick you. An answer can name your brand without a decided strength verdict. It counts for Mentioned, but stays out of Recommended rather than counting as a no. Both run from 0 to 100 in whole points, and both leave out questions that name your brand.

How the shares are averaged

A plain count over all answers would let a question with many answers, or an engine that answers more often, dominate the result. Heralded averages in two steps instead:
  1. For each cell, take the share of its answers that name you. For Recommended, take the share that recommend you. A cell whose engine named you in 5 of 7 answers reads 5/7.
  2. Average the engines that answered each question using their usage weights, rescaled over the measured engines. Then average across questions, giving each question equal weight.
Answering more often does not increase an engine’s weight within a question. Claude’s one weekly answer carries Claude’s usage weight, and another engine’s seven answers share that engine’s weight. An engine that returns nothing on a question drops out and the remaining weights are rescaled.

Country language shares

A topic in one country weighs its languages at its frozen policy shares. Its figure is shown once every language of the market has answers and every question has answers in the languages it was asked in. A language whose answers all failed holds the figure back. A question does not need answers in a language it was not asked in. Each topic keeps its own language shares. The Language filter reads only English or local-language answers and applies the usual minimum-answer rule. When your selected period includes more than one language policy, each policy’s answers carry a share of the figure proportional to their number. This also applies when the policy changes during the period. Within each policy, the language, engine and question weights stay the same. A period covered by one policy keeps the same weights.

Europe topics

A Europe topic asks each question in a different country from one day to the next, so its figures are averaged by country first:
  1. Within each country and language, the two steps above run over the answers asked there.
  2. Each country weighs languages at the shares frozen in the topic’s language policy. English-speaking countries use English alone. If a language has no answers within one Europe country, its observed languages carry that country’s weight.
  3. Europe is the plain average of the ten countries, not adjusted for population.
With every language present in each country, the fixed shares keep the observed mix from moving the figure. Some answers are missing by design:
  • An engine with no answers in one of the countries is left out of the Europe figure, so every country is read on the same engines.
  • For an overall Europe selection, a country with no answers in the 14 days holds the figure back, and it shows as a dash. A policy language missing across that selection also holds its figure. A single topic or question has no language-completeness requirement.
  • A single question, a stage or a topic, grouped or filtered, is asked in only some of the countries, so its figure is read over the countries it was asked in. It is held back only when one of those countries has no answer.
  • Understood and Praised leave out a country where no answer named you.
When a brand has other topics too, their answers combine using the same policy weighting.

Rank

Next to each figure the app shows where you stand among your tracked competitors, each read the same way. Home and Perception rank you among the competitors you track today, so one you stop tracking leaves the rank at once. A Snapshot report ranks you among the competitors on that report’s roster, and keeps that list. Your rank is one plus the number of competitors with a strictly higher figure, so brands on the same figure share a place. A figure of 0 has no rank. Scores for untracked discovered brands do not add them to these ranks.

Questions won

A question’s winner is the brand recommended in the most of its answers, over the period and filters you chose. When two or more brands are recommended equally often, each of them wins the question. A question where no answer recommends anyone has no winner, and how often a brand is named breaks no tie. An answer from a report that recorded no reading of any competitor decides nothing, so a question only such answers cover is not measured and is not counted. The Recommended pane on Perception counts the questions each brand wins. Home’s diagnosis counts the questions a competitor you track wins and you don’t, over the last week.

Reading them together

A brand can be named in most answers and recommended in almost none. That usually means the engines list you as an option and send buyers to someone else. The gap between the two figures is often the most useful thing on the page. The Heralded Score combines both into one number. Each figure has a 95% range, and a difference between reports only counts when the paired test calls it real.