Brand visibility monitoring in AI

How to read the monthly report

The report is produced every month and describes the whole period, not a single day. It begins with the brand’s presence in answers and its share of voice, and how both changed against the previous month. It then shows on which questions the brand appears often, on which rarely and on which not at all.

The following parts of the report cover the cited sources, the brand description and the comparison with competitors. It is best to read them in that order, because each part helps explain the ones before. When the brand’s share is low, we check the full answers, the questions and the cited sources. They point to the places worth analysing further. The indicator alone does not settle what caused a change.

What not to look at

• A single answer. A screenshot of one conversation shows one of many possible answers. It does not say how often the brand appears.
• A change of one question. With continuous measurement, small swings are normal. What matters is a direction that holds over several months.
• One assistant. Results from ChatGPT, Gemini, Perplexity and Claude can differ. We base our assessment on the picture as a whole, and we treat differences between assistants as information about where the gap is.

The report does not promise traffic or sales. It measures the brand’s presence in answers: how often it is named, how it is described and from which sources.

What we do with the results

Numbers alone do not improve visibility, so every report ends with a list of actions. They follow from what the measurement showed: the questions the brand loses, the sources competitors are cited from and the gaps in the company description.

Actions on your site

• Adding answers to the questions in the set, if the assistant has nothing to cite the brand from.
• Improving the description of services and categories so that it matches how the company wants to be presented.
• Tidying up company data and structured data so that a model can recognise the brand unambiguously.

Actions off your site

• Presence in the sources assistants cite most often in your industry: rankings, comparison sites, directories and industry portals.
• A consistent company description in the profiles and listings that search engines and models use.
• Correcting content in which the brand is described wrongly or out of date.

We describe these actions in detail on the AI SEO / GEO page. Later reports show whether share of voice, citations and the brand description changed after the changes were made. If you have not had an audit yet, an SEO and GEO audit also gives you a baseline.

How we build the question set

A monitoring result is only as good as the questions behind it. That is why we do not invent them at a desk. The starting point is real search keywords, together with data on how often they are typed. We turn them into the wording people use when they ask AI assistants.

Short, general questions dominate

The set is dominated by short general and ranking questions, for example “best X” or “top X”. Ranking questions show which brands assistants name when comparing offers. The length and wording of a question can change how often brands are mentioned, so we report the result against the chosen question set.

Local questions only where the company operates

We add questions with a city or region name only for the client’s real area of operation. A company that serves one city gains nothing from being measured on questions about the whole of Poland, because the result would understate its visibility in places where it does not want to sell.

The set as a shared yardstick

The same set is used to compare consecutive months. Changing the questions during measurement would change the result for reasons unrelated to what is happening with the brand.

Why we ask daily and count the result over the whole month

An AI assistant does not answer the same way every time. The same question asked twice in a row can give two different lists of brands, a different order and different sources. This happens because the model writes each answer afresh, and because an answer often relies on web search, which itself depends on the moment of the query.

If a brand were measured once a month, the result would say more about how that one answer happened to fall than about the company’s visibility. So we ask the same questions every day and count the brand’s share across all the answers in the month. Single swings average out, and the result describes the typical situation.

We ask the questions in four assistants: ChatGPT, Gemini, Perplexity and Claude, in versions with web search. Such a version uses current web pages, so what you publish on your site and in external services can influence the answers. Each assistant uses its own sources, so its answers can differ from the others.

What we measure: definitions of the indicators

Presence in answers

The share of the month’s answers in which the brand appears. The numerator is the answers that name the brand, and the denominator is all answers to all questions in the set. A result of 0 means assistants never name the brand, and growth between months means it appears in a larger share of answers. AI tracking tools call this metric visibility.

Share of voice

Your brand’s share of all mentions of the brands we track: yours and your competitors’. A mention is counted once per answer in which the brand appears. For example, if in a month your brand appears in 40 answers and competitors in 120 in total, your share of voice is 40 out of 160, or 25%. This is how we calculate share of voice in our monitoring. Other tools may use a different method. Presence tells you whether you are in the answers; share of voice tells you how much of the conversation about your category you hold against competitors.

Citations

Pages that the assistant refers to in an answer as a source. We collect them together with their URLs. That shows whether the cited page is yours, a competitor’s, a comparison site, an industry portal or a forum. The most frequently cited sources point to the places worth analysing when you plan your brand’s presence beyond its own website. Citations of your pages in Microsoft Copilot and in AI summaries in Bing are also shown by the AI Performance report in Bing Webmaster Tools.

Brand description

How the assistant presents the company: which category it puts it in, which services or products it attributes to it and what features it describes it with. We collect this description from the answers and check whether it matches the facts. A wrong category, an outdated offer or a key service missing from the description are concrete problems to fix.

Competitors in the answers

Brands that assistants name in the same answers as yours, or instead of it. For each of them we calculate the same indicator as for your company, so the result reads as a comparison. Besides the known competitors in your industry, firms you did not regard as rivals often appear, for example sites that compare offers.

Limits of the measurement

Measuring AI answers has limits, and it is worth knowing them in advance.

• Personalised answers. An assistant may take into account conversation history, account settings and the user’s location. Our measurement shows answers to questions asked under identical conditions, so it describes the typical situation. The answer a particular person sees may differ from it.
• Differences between model versions. Providers change their models and the way they search the web. After such a change the result can shift without any change on the brand’s side. That has to be told apart from the effect of your own actions.
• Scope of questions. The question set covers selected topics and does not exhaust what customers might ask an assistant. The result applies to that set.
• No guarantee. Nobody controls the answers of the models. Our actions raise the chance that a model finds reliable information about the brand, but they do not make it certain that it will recommend it.

Impressions of your pages in AI Overviews and AI Mode in Google are shown in Search Console, in the generative AI performance report.

Why measure brand visibility in AI

More and more people ask AI assistants the questions they used to type into Google: which product to choose, which company can do a job, whom to recommend. In the answer, the assistant may name brands and give the sources it relies on. A brand missing from the answer is less present in that part of the search, although the customer can still find it through other channels.

Classic SEO tools do not show this. A Google position tells you where a page stands in search results, but not whether ChatGPT mentions the brand, how it describes it and which sources it relies on. Monitoring answers those questions with numbers you can compare from month to month.

The measurement gives you three things. A baseline, meaning the brand’s current share of answers. A comparison with competitors who appear in the same answers. And a list of the sources assistants cite in answers about your industry, which we use to plan further work.