Share of Model

More and more people no longer search on Google but instead ask a chat program like ChatGPT for a recommendation. Anyone wanting to know which running shoes are any good gets an answer there with three or four brand names. That’s exactly what Share of Model is about. The metric describes how often a particular brand shows up in such answers, measured against all brands mentioned. If a program names Adidas sixty times out of a hundred test questions, Adidas has a high Share of Model. The term is still young and is mainly used by advertising agencies and market researchers.

Why brands are suddenly fighting for mentions

In the past, the crucial question was: who ranks at the very top of a search engine? There you had ten blue links, and you could click through several of them. A chatbot, by contrast, often delivers only a single answer with two or three suggestions. If a brand isn’t among them, it simply doesn’t exist for that user. Visibility is thus spread across far fewer spots than before.

On top of that, many people consider an AI’s answer to be neutral. An ad is recognized as advertising, whereas a recommendation in a chat feels like genuine advice. That’s why a mention there is considered especially valuable. Companies are now investing money to measure and increase this value.

It’s important to distinguish this from classic market share. Market share indicates how much a brand actually sells. Share of Model only indicates how present it is in machine-generated answers. The two can diverge widely: a small brand with lots of good reviews online can be overrepresented in chat answers.

How the mentions are counted

The measurement basically works like a survey, except that the respondent is a piece of software. One or more AI systems are asked hundreds to thousands of typical questions. Examples would be: “Which bicycle helmet is safe?” or “Which bank has good terms for young people?” An evaluation program then counts which brands appear in the answers. A brand’s share of all mentions yields its Share of Model.

Usually the analysis goes even further. It makes a difference whether a brand is recommended first or only mentioned in passing. Tone matters too: a model can name a brand while simultaneously warning against it. Good analyses therefore capture position and evaluation separately.

A well-known pitfall is fluctuating results. AI language models don’t answer identically every time, even given the same question. A single test therefore says little. Only many repetitions over weeks yield a reliable number. Moreover, the value always applies only to a particular system: the result on ChatGPT can turn out quite differently than on Gemini or Perplexity.

From search engine optimization to AI optimization

In news reports, the term mostly appears in the context of marketing and advertising. Agencies now sell services that promise to increase a company’s Share of Model. The field goes by names like GEO, short for generative engine optimization, meaning optimization for answer machines. It is the successor to classic search engine optimization.

The methods behind it are indirect. A language model has learned its information from texts on the internet or searches the web while generating an answer. So whoever appears frequently and positively in comparison tests, trade articles, and Wikipedia entries is more likely to be mentioned. Some providers therefore deliberately rewrite their product pages so that machines can easily extract the facts.

As a user, a skeptical eye is worthwhile. When a chatbot recommends a brand to you, there’s no real experience behind it, just statistics about texts. Popularity online is not the same as quality. And the more companies write specifically for machines, the more the picture becomes distorted. For bigger purchases, it still makes sense to check several independent sources.