The window is anchored on your last day of data
“Past 30 days” does not mean the 30 days ending today. It means the 30 days ending on the most recent day your workspace has data for. This is deliberate. A workspace whose tracking paused three weeks ago would otherwise show a month of zeroes and look like a collapse in visibility rather than a gap in collection. Anchoring on real data keeps the chart about your brand instead of about the scheduler. The page tells you when the window is not current: quietly within a couple of days, prominently after that. If a figure looks stale, read that notice first.Trailing presets include one extra day
“Past 7 days” covers 8 calendar days, and “Past 30 days” covers 31. Both ends of the range are inclusive, so today (or your anchor day) and the day 7 before it are both in. It matters when you reconcile against another tool. A 1-in-31 difference in the denominator is about 3%, which is enough to make two correct systems look like they disagree.Your plan floors how far back you can look
Each plan keeps a fixed amount of history. Pick a start date earlier than your plan allows and the range is clamped to the floor, including “everything”, which becomes “as far back as your plan goes”. It also decides what you open on: the default period is the longest of 7, 30 and 90 days that your plan can serve, so a short-history plan opens on “Past 7 days” rather than on an empty month. A notice above the table says when you are looking at a clamped window, on Analytics, Sources and Competitors. Prompts shows no such notice, so on a short-history plan that table is floored silently. One thing is deliberately not floored: the comparison period behind each delta. It is the equal-length window immediately before the current one, computed on the full data, so the change figure stays honest even where the chart cannot reach back that far.Branded, unbranded, all
A branded prompt contains your brand name: “is Surfaced any good?”. An unbranded prompt does not: “best AI visibility tracker”. They measure different things, and mixing them produces a number that means neither. Branded prompts return your brand nearly always, because the question already contains it. Unbranded prompts are the ones that tell you whether you are being discovered. On a workspace with both, the gap is routinely the difference between 77% and under 1%. So the control defaults to unbranded, the harder and more useful number. If you have never touched it, that is what you have been reading. Two consequences:- Unclassified prompts are in neither bucket. A prompt that has not been classified appears only under “All”. Switching between branded and unbranded on a workspace with many unclassified prompts will not account for every prompt you have.
- Most workspaces never see the control. It only appears once a workspace has at least one branded prompt. Everywhere else the scope is “All”, and a note on the page says so.