Report · How to read it
What is in an AI visibility report? How to read the numbers
By Kristian Stig Henriksen · Dear Future
An AI visibility report answers one question: how often, and how, do the AI assistants mention your brand when customers ask the questions that lead to a purchase? It does so with a fixed set of questions, measured every day, and it shows the numbers alongside the competitors' and with the uncertainty that belongs to them. Here is what is in it, part by part, and what to do with it.
The prompt set: the foundation for everything else
The report begins with the questions that were asked: how many, how they are spread across topics and buying stages, and that none of them names the brand. It is the most important section to read first, because the numbers only mean something for the questions that were asked. A set of 120 questions built as a market survey covers the category; a set of 20 chosen by you covers your assumptions.
The three signals per answer: mentioned, cited, tone
Every answer is measured on three things. Mentioned: is the brand's name in the answer. Cited: does the answer link to one of the brand's pages, and which. Tone: is the brand recommended, mentioned neutrally, or advised against. The three are counted separately because they tell different stories: many mentions without citations means the model knows you from training but does not find your pages; citations without mentions means your page is used as a source without you being recommended.
Share of voice: the share of the whole answer
Share of voice is your share of all mentions in the answers, you and the competitors combined. It is the number that can be compared over time and against competitors, because it accounts for answers getting longer or shorter. The report shows it per week, per assistant and per topic, and always with the competitors alongside.
The uncertainty: the interval on every number
Every number has an interval because it is measured on a sample of answers. A share of 2% with an interval from 0.4% to 3.6% is not 2%; it is somewhere in that range. The report only calls a change real when the intervals for before and after do not overlap, the direction holds across the days of the week, and the control group has not moved. The guide to the confidence interval explains the calculation.
The sources: which pages the models read
The report lists the pages and domains the assistants cite in your category: your own, the competitors' and all the others, overviews, trade media, documentation. It is the market's source map, and it answers two questions: which of your pages work, and where you need to be mentioned to get into the answer.
The competitor gap: where they are mentioned and you are not
For every question the report shows who leads and whether the lead is statistically real. Questions where competitors are mentioned and you are not are the work list. Questions where nobody is mentioned are empty seats. Questions where you lead are the ones to keep an eye on. This is also where new market opportunities show themselves: questions nobody answers well today.
The roadmap: from numbers to action
The report ends with recommendations, and each recommendation is tied to the questions it is meant to move, with the tasks that belong to it and a date for when the effect can be read. That makes it testable: when the read date arrives, the linked questions show whether it worked.
What to do in week one
- Read the prompt set and check that it resembles your customers' questions.
- Find the five questions with the largest gap where you have a page that almost answers.
- See which sources are cited in those answers, and what they do that your page does not.
- Choose one change and note the read date. Two changes at once cannot be measured.
- Leave the control group alone.
Read on: the confidence interval behind the numbers, share of voice in AI and the method behind the measurement.
From anatomy to your report
Want to see the report for your own category?
Get an AI visibility analysis: prompt set, daily measurement, competitors, intervals and a roadmap, ready within a day.
Get the analysis →Frequently asked questions
What information does a report on a brand's performance in AI answers contain?
The prompt set and its distribution, the three signals per answer (mentioned, cited, tone), share of voice per week, assistant and topic, the uncertainty on every number, the sources the assistants cite, the competitor gap per question, and a roadmap with recommendations tied to the questions they are meant to move.
How do you track how often a brand appears in AI assistants?
By putting a fixed set of customer questions to the assistants every day and recording for every answer whether the brand is mentioned, cited and in what tone. The shares are read in windows of weeks with an interval, because a single answer carries almost no information.
What are the next steps after an AI visibility analysis?
Read the prompt set, find the five questions with the largest gap where you have a page that almost answers, see which sources are cited in those answers, choose one change with a read date, and leave the control group alone. The roadmap in the report is built for exactly that order.
Can the AI answers reveal gaps in the market?
Yes. Questions where no provider is mentioned, or where the answers are thin and without sources, are demand nobody covers today. The competitor gap in the report shows them side by side with the questions the competitors already win.