Skip to main content
Prompts lists the questions Surfaced sends to the AI engines on your behalf, with your results for each one. It is the screen that turns “visibility is 47%” into “we lose the pricing question and win the integration question”.

What is on the page

One row per prompt per language. A question tracked in six languages is six rows, each with its own numbers, grouped under a parent you can expand. Clicking a row opens the detail panel with the answers themselves. Reading three or four of them teaches you more about a bad number than any chart.

Controls

Search, sorting on every column, and group by: funnel stage, tags, language, asked-from country, brand visibility band, or branded/unbranded. Grouping by brand visibility is the fastest way to see your worst questions together. The toolbar also shows your response budget rather than a prompt limit: how many tracked responses this prompt set will use against what is left this month. It warns as you approach the limit and never blocks you from adding. There is no cap on how many prompts you may save, since what costs is running them. See Billing and usage.

Adding prompts

Three ways:
  • Quick add. One line in the table, defaults applied.
  • Add prompt. The full sheet. It always creates a group: the question in your workspace’s primary language, plus any language-and-country variants you add, with auto-translation you can edit or re-run.
  • Suggest. Surfaced proposes a handful of questions based on your brand and site. Nothing is saved until you accept one.
If the workspace has no prompts at all, the empty state generates a set for you to choose from. Generated prompts appear in the table inactive: nothing is tracked, and nothing costs you a response, until you activate the ones you want.

Grouping variants

Select two or more rows and choose Group as variants to bind differently-worded versions of the same question together, picking one as the source. They then report as one parent with expandable children, which stops a question being counted as several.

What to watch out for

This table is not filtered by brand scope. It lists every prompt and badges the branded ones, while the Analytics KPIs default to unbranded. The two populations are different on purpose, and that is the usual reason a row count here does not reconcile with a figure there. “Mentioned by” hides absence. It lists the engines that mentioned you. An engine that ran the prompt and never named you is not in the column at all, so it cannot show you where you are losing. The per-platform charts on Analytics can. Deleting a prompt deletes its history, permanently. Every answer recorded for it goes with it, and there is no undo. If a question has stopped mattering but its past is worth keeping, untrack it instead. Untracking is not deleting. Removing one language of a group keeps the row and its history and moves it to an Untracked prompts section with a “Track again” button. That is the safe way to stop paying for a question without losing what it told you. Paused is a budget state. A Paused prompts section appears when prompts have stopped running because the workspace reached its monthly response limit. Nothing was deleted, and they resume when the quota resets or the plan changes.