Prompts

A prompt is one question a customer might ask an AI engine, such as "which CRM is best for a small agency?". Everything Rixe knows about your visibility comes from asking your prompts, so the set you track decides what you can learn. This page covers the Prompts screen: reading each question's results, adding and pausing questions, and the suggestions Rixe makes.

#Where to find it

In the sidebar, Prompts › Your Prompts. The band at the top tells you what the workspace tracks — "The Acme configuration includes 48 prompts across 6 topics · measured daily" — and how many of your plan's prompts you use ("48 of 50 prompts used"). The tabs are:

#Choosing prompts worth tracking

Each prompt is measured on every engine your plan includes and uses part of your plan's allowance, so a short, sharp list beats a long one. Three kinds earn their place:

• Category prompts — no brand named ("best CRM for a small agency"). These decide who gets discovered, and they are the ones you can win or lose. A healthy set is mostly category prompts.
• Comparison and purchase prompts — where a buyer is already choosing ("Acme vs. Globex", "which CRM should I buy?"). Losing here is expensive and usually fixable.
• Brand prompts — questions that name you, to see what engines say about you unprompted and catch wrong information early.

Rixe tags every prompt with its Prompt type (Category or Brand, worked out from the text) and its Intent — Awareness, Consideration, Evaluation or Purchase. Engines rarely name brands on awareness questions and often do on purchase ones, so cover every stage.

#Reading the Active tab

Each row is one prompt, with results for the chosen period (last 7, 30 or 90 days):

• Visibility — the share of answers, across engines, that named your brand. Filter by Visibility range: 0% — never mentioned, Below 25%, 25–50%, Above 50%.
• Engines — which engines named you for this prompt.
• Position and Sentiment — where you appeared in the answer and how it portrayed you.
• Difficulty — how winnable the question is: Open ground (nobody wins it — the cheapest ground you have), Entrenched (one brand owns it), Rotating (brands take turns) or Contested (several brands in play, a good page can put you ahead). It needs at least 10 answers; before that the row says so.
• Volatility — whether the answer stays the same between runs: Sticky, Settled or Unstable.
• Topic, Persona and Volume can be shown or hidden from Columns.

You can group rows By topic, By intent, By type or as a Flat list, search the list, filter by engine, intent and prompt type, and download the table as CSV.

A rate based on fewer than 4 runs is marked as not reliable ("Only 3 runs — this rate is not reliable") instead of being shown as a confident number. See Visibility score.

#Opening a prompt

Click a prompt to open its page. There you find:

• Answers — each engine's answer in full (Open answer), the brands it named and the sources it cited. Answers shared with another workspace that asked the same question are marked pooled.
• The sources cited by the answers in the period, and in how many of those answers your brand and your competitors were named.
• The searches the engines ran before answering this prompt.
• A run summary: how many runs completed and how many failed in the period, and why (for example "The engine did not answer in time"). The rates on the page cover only completed runs.
• Why are we losing? — for a question you never appear in, Rixe compares your page with the pages the engines cited and says what is missing.

If an engine showed no AI answer at all for the prompt, the page says so: "This is not a failure, it is a measurement result."

#Adding prompts

There are four ways to add prompts. Every one of them checks your plan's prompt limit first.

1. Prompt Research. Click Add at the top of the Prompts screen, or open Prompts › Prompt Research. Generate questions from a Seed keyword, a Topic or a Competitor brand, or use Manual entry. Generated questions are saved as proposed; in Review generated questions you choose which ones become active straight away ("Add 5 to active prompts").
2. Weekly prompt recommendations. On the Suggested tab, up to 15 new questions arrive every Monday, each with its reason: Coverage gap (an intent stage your set lacks), Market demand (a question rising in the market), Searched on your site (visitors searched it on your site and found nothing) or From fan-out. Accept or Reject each one; a rejected suggestion is never offered again.
3. Suggested prompts from demand. Also on the Suggested tab, questions clustered from search demand, shown with a demand band (low, medium, high) — never a precise volume, because nobody knows exactly how often a question is asked.
4. Setup. The questions you selected in the setup wizard (Quickstart).

New prompts are measured from the next daily measurement.

#Pausing prompts

To stop measuring a prompt without losing its history, select it in the Bulk grid (open it from search) and choose Pause; Activate brings it back. Paused prompts keep their past results and still count toward your plan's prompt limit.

#Personas

Select one or more prompts and use Assign persona to label them with an audience (for example SMB owner or Marketing manager). The Persona filter on AEO Insights then shows results for that audience only.

#Limits

The number of prompts you can track depends on your plan and is counted across all workspaces. On agency plans the credit pool is the limit that matters. When you have no free slots, Accept buttons are disabled and the screen says "Your plan has no free prompt slots." Measurement of existing prompts never stops because of a limit. See Plans and limits.

#Common pitfalls

• Many near-duplicates. Ten phrasings of the same question cost ten prompts and tell you about one topic. Use Engine searches to see how engines themselves phrase it.
• Only brand prompts. They make your visibility look high because the engine is led by your name. Use the Unbranded filter to see the real picture.
• Judging a prompt too early. Wait until there are enough runs; difficulty needs at least 10 answers.

#Related pages

• Conversations — follow-up questions asked as one conversation.
• Coverage and fan-out — gaps in your set and the searches engines run.
• Visibility score — how rates become a score.
• Sources — the pages engines cite.