Skip to main content
Mentioned sends each tracked prompt to four AI engines. Each engine returns an answer. Mentioned then reads every answer and records four things.
  • If the answer names your brand.
  • The sentiment toward your brand.
  • Your position, if the answer is a ranked list.
  • The other brands in the answer, and the role of each brand.
An AI model does this analysis. Every metric on your dashboard starts from it. This page tells you how the analysis works, and how Mentioned checks it.

​
The mention check has two steps

First, the analysis model reads the answer and decides if your brand is there. Second, Mentioned looks for your brand name in the answer text and in the cited links. If the name is not there, Mentioned changes the result to no mention. The second step can only remove a mention. It can never add one. This step exists because of a known problem. A brand name that is also an ordinary word causes wrong mentions. The analysis model reads the ordinary word as a reference to the brand. To count, the name in the answer text must have a capital letter, or it must be part of a web address. For example, for a brand named Summit, “Summit” and “summit.com.au” count. The word “summit” in a normal sentence does not count. Mentioned uses the same rule when it highlights brand names in an answer. If your brand name in Settings has no capital letters, Mentioned accepts the name in all forms.
This rule removes most wrong mentions for a common-word name. It cannot remove all of them. A capital letter at the start of a sentence or in a heading still counts. Read your mentions in context if your brand name is an ordinary word.

​
The role of each other brand

Mentioned puts each other brand in the answer into one of two roles.
  • An option. The answer offers this brand as a choice for the question. Competitive Ranking uses these brands.
  • A reference. The answer names this brand in passing. Examples are a platform, a material, or a publisher. Competitive Ranking does not use these brands.
This boundary controls which brands become competitors in your results.

​
How Mentioned checks the analysis

Two people read the same set of AI answers. Each person labels the answers independently. They do not compare notes. A third step settles the answers where the two people disagree. These human labels become the reference set. Mentioned then compares the analysis model against the reference set. The comparison covers mentions, sentiment, rank, and the role of each other brand. The independence rule is necessary. It shows if the two people agree with each other. If two trained people disagree about an answer, the answer is difficult, not incorrectly analysed.

​
Current status

Mentioned does not publish accuracy figures for the analysis yet. The first check uses a small set of answers, approximately 20. A set of this size can show a large problem. It cannot give a precise accuracy figure. Mentioned set the pass and fail levels for this check before it saw any result. Mentioned will publish the figures, their limits, and the known problems when the first check is complete.
Read this page with Brand Health Score and Mentions and Mention Rate. Those pages tell you what each metric measures. This page tells you how Mentioned produces the evidence behind them.