Prompts

A prompt is one question your buyers ask AI, tracked in one segment. Heralded asks it of every engine exactly as written: ChatGPT, Gemini, Google AI Overviews and Google AI Mode daily, Perplexity and Microsoft Copilot twice a week, and Claude weekly. Your prompts decide what every figure measures, so they are the most important choice you make in Heralded.

​

Choosing prompts

Good prompts sound like a buyer who doesn’t know your brand yet, such as “best project management tool for agencies” or “how do I reduce churn in a SaaS product”. During setup Heralded drafts a set from your site and your market, and you tick the ones to track. After that, you manage them on the Prompts page. If keyword research is unavailable for your market and language, setup uses questions drafted from your site. Their topic search volumes still come from Google Ads where it has data. Heralded never rotates or rewords a prompt. A prompt’s wording is fixed once it is tracked. To ask it differently, archive it and track the new wording.

​

Type and intent

Each prompt carries a type:

• Organic, a category question that names no brand. Most of your prompts should be organic.
• Brand, a question that names your brand, such as “Is Acme any good?”. It doesn’t count toward your score; see which answers count.
• Competitor, a question that names a competitor.

Each prompt also carries an intent: Problem, Educational, Comparison or Transactional. Heralded sets both, and a Member or above can correct them. Heralded also sorts prompts into topics, which you can filter by.

​

Pausing and archiving

A prompt is active, paused or archived.

• Pause stops asking the prompt and keeps its last values. Resume starts it again.
• Archive retires the prompt. Its daily answers are kept for 90 days.

Paused and archived prompts free their prompt slot.

​

The Prompts page

The page has three views: Tracked, Discover and Archived. Under the title, a line says what the page holds: the questions your buyers ask AI, asked every day. The filter bar selects the period, engine, type, intent, topic and segment. You offers Any, Named and Not named. Not named lists prompts with answers in the period where no engine names your brand. The four cards show Mentioned, Recommended, Understood and Praised under those filters. Open a card to read that figure. Tracked lists your active and paused prompts, ten per page. Search prompts searches their wording. Each row shows the prompt with its kind of question, engine marks, Score, Trend and Searches/mo, the monthly searches for its topic, such as “~1,000”. The kind of question is the intent plus the word question, such as “Comparison question”, and it names the type first when the prompt is a brand or competitor prompt, as in “Brand · Comparison question”. A paused prompt shows its pause date under its text. When you select another page, the current rows stay dimmed while Loading these prompts… appears over the table. The pager is disabled until the new rows arrive. If they cannot load, the current rows return with an error above the table, and you can try again. Each dot is your Heralded Score for the prompt on one AI engine. The legend shows the five score bands and their ranges, plus Not measured. Hover a mark for its score, figures and answer count. Open it to read that engine’s most recent answer in the period. Open the row to read the prompt’s pane. A prompt you track from the Track a prompt pane or from Discover is read within minutes on every engine its segment reads, without waiting for the next daily update. Prompts you track within a minute of each other are read together. The next time you open Prompts, the prompt’s pane shows each engine’s answer. Its marks stay Not measured until a weekly report scores it. A brand whose first report has not started yet reads the prompt in that first report. Every pane has the same header: its title and read-only period and filter scope on the left, and close on the right. Its actions sit in a footer pinned to the bottom of the pane, with a ⋯ menu at the bottom-left. The header stays visible as you scroll. Use that menu to pause, archive or correct a prompt’s type and intent. Pause and archive ask you to confirm first. On an archived prompt, Restore prompt is the primary action at the bottom-right. A message confirms each saved change or explains a failure. Reversible changes offer Undo. The score shows ”–” when too few answers were measured. Trend shows an arrow and the size of a real rise or fall, or Steady with the signed change. It shows Calibrating with one to three earlier reports and New for a prompt first read in the period. Searches/mo stays empty when nobody has a number, and starts highest first. Select the Prompt, Score, Trend or Searches/mo header to sort; select it again to reverse the order. Discover lists prompts worth tracking, ten per page. Each row shows the prompt, its question type (Short, Conversational, Written as a buyer, Comparison or Alternatives), a star on the ones best to track, short labels saying why it is suggested (Search Console, People search this, Rivals named, not you, Nobody named yet, Based on competitors, Engine searches) and Searches/mo, empty when unknown. Segment picks the segment a suggestion adds to, and Question type narrows the list to one question type. The list starts with the ones Best to track first. Select Searches/mo to sort highest first, again for lowest first, and a third time to return to the best-first order. If Search Console is not connected, the first row offers to connect it. Track adds a suggestion, uses one prompt slot and reads it within minutes. Under the Tracked table, Worth adding shows the top three as cards, with Adds to naming the segment a Track writes into. Archived holds retired prompts.

​

Inside a prompt

Open a tracked prompt to see its Heralded Score and the answers behind it. The four tiles show Mentioned, Recommended, Understood and Praised, each with a whole count of answers read as Yes. Understood and Praised count only answers naming you. Partial agreement contributes to the Understood percentage but does not count as a Yes answer. Open a tile to see that figure with the prompt fixed in its scope. The figure pane keeps its standard blocks under that prompt and the period and engine filters. Its figures, trend, claims, competitor rows and actions follow that scope. The closing link for Understood and Praised opens only the answers naming you. Score shows one bar per weekly report. With one or two reports, dashed placeholders mark the remaining space. When the next report is scheduled, Trend starts shows when the third report is due. Competitors lists the brands named on this prompt, highest score first, with their scores, Mentioned, Recommended and Trend. Your brand is tagged You. Open a row for that competitor’s pane. Each engine card shows your score on that engine, the four figures coloured by their own bands and the score by week. Open a card for that engine’s pane, fixed to this prompt. An engine filter leaves one card and a Filtered to note. Sources lists the five hosts cited most for the prompt, their types, whether they name you and their citation counts. Open a source for its pane fixed to this prompt, or follow All sources cited for this prompt to Sources with the prompt filter set. Actions holds the work tied to the prompt. Show answers opens the answers behind the pane.

​

Answers

Open All answers at the right of the Prompts page, or follow Show answers at the end of a prompt, figure, engine, competitor, source or page pane. The pane’s link keeps its subject and filters selected. The link has no count and stays available when there are no matching answers. The Answers page shows the empty state. The Answers table shows one engine answer per row, 25 rows per page. Narrow it by period, engine, prompt type, intent, topic or segment, then by Prompt, You, Names or Cites. Prompt includes archived prompts with answers in the selected period. Names lists the brands the answers name, tracked competitors first. Cites lists Your site once for your owned hosts, then other cited hosts. Search answers searches retained answer text. When a Sources group or page opens Answers, Cites shows that subject. Clear removes it. Each row shows Mentioned, Recommended, Understood and Praised, the answer excerpt, other brands named, how many cited sources are from your site, or None from your site, and the time in your time zone. The states are Yes, Partly, No and Not measured. Understood and Praised are not measured when the answer does not name you. The Prompt column disappears when one prompt is selected. Sort by Sources for answers that cite the most sources first, or When for the newest answers first. Open a row to read the full answer. The list, full answer and answers CSV use the same frozen measurements and detected brand names. Cited page labels use the title stored when Heralded read the page. A running scan’s answer uses live measurements until the scan finishes. When its text has passed retention, the row says Text no longer kept; its measurements remain.