---
title: "Choosing prompts"
description: "How to build a prompt set that shows your visibility where buyers decide: focus on commercial queries, cover every entry point, and report by topic."
---

> Documentation Index
> Fetch the complete documentation index at: /llms.txt
> Use this file to discover all available pages before exploring further.

# Choosing prompts

Your prompts decide what Airefs measures and what it recommends.

A good prompt set is mostly commercial, covers every way a buyer can reach you, and is organized into topics you can report on.

## Commercial vs informational prompts

Prompts fall into two kinds, and they play different roles in your set.

### Commercial prompts

Commercial prompts come from people looking for a product or provider: *“best CRM for small agencies”*, *“alternatives to \[competitor\]”*, *“\[competitor\] vs \[competitor\]”*.

Airefs focuses on these queries because they are the ones that drive leads and customers. The answer usually names and compares brands, so appearing in it puts you on the buyer’s shortlist at the moment they are deciding. Being absent means a competitor is on that shortlist instead.

Make commercial prompts the core of your set. They are the prompts where a higher mention rate and a better position are most likely to turn into business.

### Informational prompts

Informational prompts come from people trying to understand something: *“what is lead scoring?”*, *“how does CRM automation work?”*.

These answers explain rather than recommend, so they rarely name brands. When they use a brand’s content, it usually appears as a source, not in the answer itself. A low mention rate on these prompts is normal and says little about your chances of winning customers.

Only track informational prompts when you want to see whether your content is being used for specific topics and questions. Judge them on citations, not mentions.

Give informational prompts their own topic. Rates are calculated across all the prompts in view, so a batch of informational prompts will pull down your overall mention rate without telling you anything new about your position with buyers.

### How they compare

|     | Commercial | Informational |
| --- | --- | --- |
| Buyer is | Choosing a solution | Learning about a subject |
| Brands appear | In the answer and in the sources | Mostly in the sources |
| Metrics to watch | Mentions, position, and citations | Citations |
| Track it to | Win leads and customers | Check that your content is being used |

### Prompts that include your brand name

Avoid prompts that name your own brand, such as *“What do you think of \[your brand\]?”* or *“Compare \[your brand\] vs \[competitor\]”*. They look useful but give you little to act on, for two reasons:

-   **The answers are diplomatic.** When a prompt asks about a specific brand, AI models almost always reply with a balanced list of pros and cons. That tells you little about whether the model would recommend you to someone who has not heard of you.
-   **They inflate your visibility.** Airefs counts every mention of your brand in an answer. If the question contains your brand name, the answer will almost always repeat it, so the prompt registers as a mention whatever the model thinks of you. A set with several of these shows a mention rate that does not reflect how often you are recommended.

Write the prompt as the buyer would before they know your brand, such as *“best CRM for small agencies”*, and see whether you appear. Competitor comparisons that do not include your own brand, like *“\[competitor\] vs \[competitor\]”*, are fine, because they show whether you are named as an alternative.

## Aim for coverage, not volume

Choosing prompts is not keyword research. There is no volume data for AI prompts, and people rarely phrase the same question the same way, so chasing “high-volume” wording is the wrong instinct.

What matters is coverage. If a route to your product has no prompt, you have no data for it, and you won’t see a competitor taking it.

Work through these entry points and write commercial prompts for each one that applies to your business.

### Service or product

What you sell, in the words buyers use rather than your internal naming: *“best email marketing platform”*, *“CRM for small business”*.

### Feature

The specific capability people shortlist on: *“email tool with automation workflows”*, *“CRM with built-in call recording”*.

### Use case

The job they need done: *“best tool for re-engagement campaigns”*, *“tools for onboarding new customers automatically”*.

### Pain point

The problem as the buyer feels it, sometimes before they know the category: *“software to stop losing leads in follow-up”*, *“what can I use to stop my emails landing in spam?”*.

### Audience or segment

Who the buyer is: *“marketing software for e-commerce brands”*, *“CRM for agencies”*.

### Location

Where the buyer is, when that changes the answer: *“best accounting software in Australia”*, *“GDPR-compliant email platform for EU companies”*.

### Alternatives

Queries anchored on a competitor: *“alternatives to \[competitor\]”*, *“\[competitor\] vs \[competitor\]”*.

## Decide how many prompts you need

A complete set is usually **25–50 prompts per market**, where a market is a country or location you sell into. Below 25 you are likely to have blind spots. Far beyond 50 you are mostly rephrasing prompts you already track, and answers to near-identical questions tend to converge.

Run the same set in each market instead of loosely adapting one list. Answers differ by country, and a matching set lets you compare markets directly when you group the Prompts table by **Country**.

## Group prompts by topic

A single prompt is a weak thing to report on. It is one question out of the many a buyer could ask, and its answer can change from run to run. A topic pools several related prompts, which gives you a steadier measure that also means something to the people reading your reports.

Add related prompts to the same topic. A prompt can belong to more than one topic, so *“CRM for agencies”* can count toward both *agencies* and *crm*.

### What you can see for a topic

Two views give you topic-level results:

-   **Prompts page, grouped by Topic** — each topic shows the share of its prompts where you were mentioned and cited in the latest answer, your average position across the prompts where you ranked, and your rate across recent answers.
-   **Overview page, filtered by topic** — shows the topic’s mention rate, citation rate, and your rank against tracked competitors for both, with the trend over time and the change from the previous period.

### Read each topic for what to do next

Together, these numbers show where you stand and what kind of work is needed. For example:

| Topic | Mentioned | Cited | Avg. position | What it suggests |
| --- | --- | --- | --- | --- |
| Alternatives | 60% | 10% | #2.0 | Recommended, but your own content isn’t a source. Publish content AI can cite. |
| Agencies | 20% | 40% | #5.5 | Your content is used, but you’re rarely recommended. Check how it positions you for this audience. |
| Automation | 0%  | 0%  | —   | Absent. Review the cited sources to see who is shaping these answers. |

This is also easier to report. *“We moved from 5th to 2nd among tracked brands for agency prompts”* tells a client or manager far more than *“we now appear for one more prompt.”*

### Set up topics you’ll want to report on

Name topics after the things you discuss in your reports, such as product lines, features, customer segments, or competitor comparisons. Give each topic several prompts, because a topic with one or two behaves much like a single prompt.

Rates cover every prompt in the topic, so adding or removing prompts can move a topic’s numbers without any change in AI answers. Keep that in mind when you compare periods, especially after expanding a topic.

Source: /docs/choosing-prompts/