Skip to content

The poll

Answer Share comes from the poll: a continuous sample of what the answer engines say on your topics. It works the way an opinion poll samples voters. Questions are drawn from each topic in proportion to its demand, so the share it reports stands for the market’s questions, and every share carries a known range.

The poll runs through the day, in small batches, sized so the plan’s monthly allowance of answers lasts the month. How many answers it reads is the plan’s monthly allowance per site, on Plans and limits. A new answer updates the topic’s share at once; there is no monthly panel to wait for.

  1. Across topics, in proportion to demand: a high-demand topic is asked more often than a low-demand one.
  2. Within a topic, the question asked least recently goes first.
  3. Each chosen question goes to every available engine, once.
  4. Then the next question. Repeats are rare until a topic’s questions have all been asked.

Many different questions beat asking the same one again. Two asks of one question mostly re-measure the same thing: the answers overlap far more than the answers to two different questions on the same topic. So for the same number of answers, a poll of many questions tells you much more than a short list asked repeatedly. Answer Share still counts a repeated question honestly; see Answer Share.

For every answer: the engine and how it was reached, the question and its topic, the answer’s text, the pages it cited and the pages it only read, the searches the engine ran to answer it, and every brand it names or cites.

Brands are read from each answer by a model, then checked against the brands the site already knows by name, so a known brand is never missed. A brand counts as cited when one of the answer’s sources is on one of the brand’s own domains. The engines themselves, and publishers cited only as sources, are not counted as brands.

Engine How it is asked
ChatGPT Through its app, as a person sees it. When the app cannot be read, its official interface with web search on answers instead, and the answer is labelled so.
Perplexity With live web search on every answer
Claude With web search on
Gemini Grounded on Google Search
Grok With live web search on
Le Chat With web search on
Google AI Overviews Observed on Google’s live results page
Google AI Mode Observed on Google’s AI Mode page
Copilot Search (Bing) Observed on Bing’s live results page

ChatGPT is read through its app because the app and the developer interface answer differently: they name different brands and cite different pages. Answer Share should be what your buyers see. The Overview’s engine strip marks ChatGPT “app” on the days its answers came from the app. See Answer engines.

An engine that fails or is unavailable is stopped for the day and records nothing, which counts neither for nor against anyone. Which engines answer depends on what is available that day.

The poll asks with search on, the way people use the engines. Separately, a small share of each day’s asks goes to the engines that can answer without searching, with search off. Those answers feed the Memory lens and never count toward Answer Share. See Why.

A topic like “best CRM for startups” in the US market is the same question for every workspace that tracks it. When another workspace asked an engine exactly the same public question from the same country within the last six hours, that answer is reused instead of asking again. Brands are matched for each workspace when it reads the answer, so reuse never mixes one workspace’s competitors with another’s. A reused answer holds the engine’s answer only, never who asked.

  • Private questions are never shared. A question your workspace keeps private is always asked live: its words go only to the engines asked, never into a shared answer. Every question taken from your Search Console is private.
  • You can switch reuse off. Under Settings, General, Reuse recent answers to the same questions off means your questions are always asked live and your answers are never reused.

Each plan includes a number of poll answers a month per site, and a number of topics. Both are on Plans and limits. The poll’s budget is its own: memory asks, source reading and the Engine Briefs are metered separately and never stop the poll. A paused plan pauses the poll.

It is a sample of answers to questions that stand for your market, not a count of how often anyone asked. The engines answer differently by time, place and person; the poll asks from your market and records the date of every answer. Limits of the method says what that can and cannot prove.