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

# Choose your prompts

> Which questions to track, how many, where they come from, and how to check them after the first week.

A prompt is one question that a buyer asks an AI platform. Every figure counts the answers to the project's prompts. So the choice of prompts sets the level of every number, before any answer arrives.

This guide is the judgement. [Prompts](/reference/prompts) is the page that runs them.

## What makes a prompt measure something

<Steps>
  <Step title="Ask for a choice">
    Ask for the best providers, the alternatives, the prices or the trade-offs. A question that asks for providers gets a list of providers. A question that asks how or why gets advice, and advice seldom names a product.
  </Step>

  <Step title="Stay in this brand's category">
    Ask what a buyer of this brand's product asks. To test a topic, ask this question: when a buyer asks for providers on this subject, are this brand's competitors the answer? If not, the topic measures the brands of another market.
  </Step>

  <Step title="Say who asks and what they need">
    "The best invoicing tool for a 10-person agency" measures more than "the best invoicing tool". A [persona](/reference/personas) can send the same sentence in front of many prompts.
  </Step>

  <Step title="Keep brand names out of the wording">
    A prompt that names this brand is [branded](/reference/filters#branded-prompts), and every figure leaves it out by default. A prompt that names a competitor names it in every answer. Read those prompts apart from the rest: a topic of their own, such as "Comparisons", gives them their own row in the Overview breakdown.
  </Step>

  <Step title="Keep the wording fixed">
    A new wording restarts the prompt's trend. Leave out the year, because a prompt runs for months.
  </Step>
</Steps>

A few how-or-why prompts still earn their place. They name few brands, but they show which pages the platforms read on the subject. Read them on [Sources](/reference/sources), not for visibility.

## How many prompts

A small prompt set gives an unreliable level, because most prompts never name a given brand. More prompts narrow the uncertainty step by step. Cover every [topic](/reference/topics) that the buyers ask about, with several prompts each. Read the trend of a fixed set more than its level.

The plan counts prompts across the organization's projects, and each market counts again. See [Plan limits](/reference/limits) and [Markets and languages](/reference/markets-and-languages).

[Which prompts to track](https://fullreach.ai/guides/which-prompts-to-track) and [How many prompts to track](https://fullreach.ai/guides/how-many-prompts) hold the measurements behind this section.

## Where prompts come from

1. **The setup wizard** writes the first set, for each topic and persona. See [Get started](/guides/get-started#the-setup-wizard).
2. **Add prompt**, on the Prompts page, adds one in your own words.
3. **Questions worth adding**, on the Prompts page, proposes prompts from what the answers showed. See [Suggestions](/reference/prompts#suggestions).
4. **[Searches](/reference/searches)** shows the web searches that a platform ran before it answered. A search that names a competitor and not this brand is a question the buyers really ask.

## Check them after a week

* **Names a brand.** On the Prompts page, this column counts the answers that name any monitored brand. When most answers name a competitor and not this brand, the competitors win that question. Keep it: it is the work.
* **names nobody.** A prompt with this mark measures nothing about visibility. Reword it so that it asks for providers, or accept the suggestion that replaces it.
* **The breakdowns.** The Overview splits visibility by topic and persona. A topic where no answer names a brand is outside the category.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.