Skip to main content
When someone asks an AI platform “which CRM suits a small agency”, the answer they read is assembled in a few steps. Koalr records each of those steps for the prompts you track, so it helps to know what they are.

​
1. The platform decides whether to look things up

Some platforms answer from what they already know. Most now run one or more web searches first, in the background, before writing a word. Those hidden searches are usually shorter and more literal than the question the person typed: “best CRM small agency”, “CRM pricing comparison”. Koalr shows them as Background Searches on the prompt detail page. They matter because they are the queries your pages need to win for the platform to find you at all.

​
2. It reads a handful of pages

The searches return candidate pages. The platform reads some of them, usually a small number, and treats them as its evidence for this answer. Koalr calls these sources. The Sources page lists them and shows what kind they tend to be in your category: editorial round-ups, review sites, directories, community threads, or brands’ own pages. See Sources.

​
3. It writes the answer and names brands

The platform composes an answer that names the brands it considers relevant, in an order, with a tone. Being named is a mention; that is what Prompt Visibility Rate counts. Where you sit among the brands named is your Share of Voice, and how the answer talks about you is your Sentiment. A platform can name a brand it never read a page about, because it already knew the brand. It can also read your page and still not name you. Both happen often, which is why Koalr tracks mentions and citations separately.

​
4. It sometimes cites its sources

Some answers include links to the pages they used. A link to your site in an answer is a citation. The Sources page shows which of your pages earn them and which third-party pages are cited instead of you. See Citations and mentions.

​
Why the same question gets different answers

  • Platforms differ. Each has its own search behaviour, its own index and its own habits. Being strong on one and invisible on another is normal, and the Platform Breakdown on the Dashboard is where you see it.
  • Location matters. Platforms answer differently in different countries, so Koalr asks every prompt from the country a site tracks. See Countries and languages.
  • Answers vary from run to run. Ask the same question twice and the brands, order and sources shift. That is why Koalr measures a rate across many answers, and why a trend over several runs means more than any single run. See Runs and data freshness.
  • Prompts that include your name behave differently. If the question names you, the answer almost always does too. Koalr keeps those branded prompts out of your headline rate.

​
What Koalr keeps from each answer

The full text, the brands named and in what order, the sentiment towards each, the sources read, the pages cited, and the background searches. Everything on the Dashboard, the Prompts page and the Sources page is computed from that record. Nothing is estimated. See How Koalr collects data.
The fastest way to build intuition is to open a prompt on the Prompts page and read three or four of the actual answers under AI Chats. Five minutes there teaches more than any chart.