Skip to main content
Quotable recomputes metrics from the stored answers every time you open a view. That keeps every filter honest, but it also means a number you saw yesterday can be different today. This page lists every reason that can happen, so you can tell a real change in the engines from a change in what is being measured.

A run is still landing

A daily run does not arrive all at once. Answers come in engine by engine, usually within an hour, and the analysis pass runs after them.
  • While answers are arriving, the Overview shows “Gathering prompt data” with a count such as “40 of 75 answers in”. Panels fill in as answers land.
  • Share of voice and position settle shortly after the answers arrive. They are final once every brand in each answer has been identified, including brands you don’t track, and can move by a few points in the meantime.
  • Sentiment appears last, usually soon after the answers. Answers not analyzed yet are picked up on the next pass.
Judge a day once its run has finished. Use the weekly view for direction rather than a single day’s reading.

A missed day is filled in later

If a brand’s daily run fails to complete, Quotable detects the missing date and runs it again. The answers are stored under the missed date, so a gap in a chart can fill in later.

Part of a day is hidden

A date where fewer than half of the brand’s prompts produced answers is left out of trends and run-over-run changes. The most common case is adding a prompt: Quotable queues a first run for the new prompt right away, which creates a date with only that prompt’s answers. Shown as a full run, a brand named in its own new prompt would read as “100% visibility”. Coverage is measured against the prompts that existed on that date, so a smaller prompt set in the past is not treated as partial. When the daily run later lands on the same date, the date has full coverage and appears again. If a brand only has partial dates so far, they are shown rather than hiding everything.

You changed what is being measured

Most changes to your setup affect past numbers as well as future ones, because metrics are recalculated from stored answers.

You changed the view

The same data can give different numbers depending on the view:
  • Date range. Overview headline numbers cover the whole range; changes compare the last run with the one before it.
  • Daily or weekly. In the weekly view, each week is scored as one run, so “previous” means the previous week.
  • Prompt type. A prompt is branded when its text contains your brand name. Branded prompts inflate every visibility metric; the Unbranded prompts filter removes them.
  • Engines, topic and location filters narrow both the numerator and the denominator.
Topics are assigned once. Prompts from onboarding keep the topic they were generated under; prompts you add later are classified automatically the next time the Overview loads, and are not reclassified after that.

The engines changed

AI answers are not deterministic. The same prompt, asked the same way, can name different brands and cite different sources from one day to the next, and engines change their models and search behavior without notice. This is the change you want to measure. To separate it from noise:
  • Track enough prompts that one answer does not decide a number.
  • Read trends over several runs, or in the weekly view.
  • Compare the same engine over time, then compare engines.

Tracking stopped

Days with no data are gaps, not zeros. Tracking stops when:
  • The brand is paused (Pause tracking in the brand menu).
  • Every prompt in the brand is paused.
  • The workspace has no active plan, a payment failed, or it reached this month’s usage limit for its plan. Quotable shows a message explaining which one applies.

The app and the API differ slightly

The API returns each run’s totals as stored when the run finished, across all engines, prompts and locations. The app recalculates from the answers for the filters you choose, applies the refined positions from the analysis pass, and hides partial dates. For the same date and no filters, the two are usually identical, and can differ slightly when a run was still being refined.

Next steps

Metric definitions

Every metric and its formula in one table.

Write prompts

Build a stable prompt set worth trending.
Last modified on November 18, 2026