---
title: "Share of Model: The new key metric. Your brand’s visibility in AI."
author: "Peter Vidovič (GEO strategist)"
published: 2026-05-21
url: https://www.petervidovic.com/en/articles/share-of-model/
translation_of: https://www.petervidovic.com/sk/clanky/share-of-model/
language: en
---

# Share of Model: The new key metric. Your brand’s visibility in AI.

> Advertising is no longer the last voice before a purchase; that voice is now AI. Share of Model measures how often AI answers mention your brand compared with your competitors.

## Short answer: What is Share of Model?

Share of Model (SOM) is the share of AI assistant answers in a given category in which your brand is mentioned by name. It is calculated as the number of AI answers mentioning the brand divided by the total number of AI answers in the category, multiplied by 100. It only becomes meaningful in comparison with competitors, over time and broken down by AI tool.

For decades, marketers tracked Share of Voice: how big a slice of the media pie our brand grabbed compared with the competition. A number you could put on a slide and use to defend a budget. In 2026, for the first time, Share of Voice is joined by a second metric of equal weight, Share of Model. Not as a replacement, but as an indicator for the second moment in the customer’s decision, the one that takes place in a conversation with AI. This article explains what Share of Model means, how it is calculated, why it deserves attention already today, and where its limits are.

## Share of Voice made sense while people made purchase decisions on their own

Share of Voice worked because the world was simpler. You had TV spots, billboards, media articles, banners, Google ads. You counted how much of your category’s media pie went to you and how much to your competitors. If you had thirty per cent SOV for a brand with a twenty per cent market share, you knew you had momentum and that in a few months it would show in sales too.

Behind this whole logic lay one quiet assumption. An assumption nobody described in textbooks because it was trivially true: the customer makes the purchase decision alone, based on what they have seen, heard and remembered. The media help them build a preference, but in the end they decide for themselves. Media attention converted into preference, and preference into a purchase.

This assumption no longer holds as it used to. Part of the decision, often the most important part, is moving into a dialogue the consumer has with an AI assistant. And when part of the decision moves elsewhere, one metric is no longer enough.

## What has changed: people are no longer clicking on search results

In April 2026, Gracker AI published an analysis according to which 77% of Google searches on mobile already end without a click. In Google AI Mode it is as much as 93%. [1] These are not searches where people failed to find what they wanted. They are searches where people found what they wanted directly in Google. They simply did not need to continue to a website. They got the answer in the AI Overview right above the organic results.

The second figure. According to the McKinsey AI Discovery Survey from August 2025, 44% of AI assistant users already prefer an answer from ChatGPT, Gemini or Perplexity over Google as their primary source. [2] Not as a supplement. As their automatic, primary, subconscious choice.

- **77%** of Google searches on mobile end without a click [1]
- **93%** of searches in Google AI Mode end without a click [1]
- **44%** of AI assistant users prefer an AI answer over Google [2]

That the change is real is also clearly visible in the gemiusAudience Slovakia monitoring (the IAB monitor). [3] Within a year, traffic to some Slovak websites collapsed to an extent without historical precedent. Comparing Q1 2025 with Q1 2026, zdravoteka.sk lost 73% of its real users, peniaze.sk 58%, podnikam.sk 53%, touchit.sk 47%, autovia.sk 42%. These websites have one thing in common. They answered questions about health, money, business, phones and cars. Exactly what AI now does faster and more conveniently.

For a brand that sells goods or a service, however, this is only the backdrop. The real question is different: if people are giving up classic search and increasingly ask AI, which they trust more and more, who actually influences their purchase decision?

## AI is like a fantastic salesperson whom people trust

Imagine a customer walking into an electronics store. For months, advertising for a particular TV brand has been massaging them through the media. They know which brand they want. They know which model they have chosen. They come in and meet a salesperson. A good salesperson knows the range, listens to what the customer is looking for, and recommends something else. Cheaper. Better. More reliable. In nine cases out of ten, the customer leaves with what the salesperson recommended, not with what they originally planned to buy.

In this comparison, AI is that salesperson. With one difference. The customer only meets the salesperson when they go to the store. But they ask AI before they set foot there, while choosing and even after the purchase, at home on the sofa, at work over lunch. They ask it about everything they want to get. A mortgage. A car. Dishwasher tablets. A holiday. Wooden cladding for the terrace. A child car seat.

And people trust AI. They know it earns no commission on sales, has no contractual partner, has no trained sales rep in the back office. It speaks calmly, competently and always. For a growing group of people, it has become the natural first choice when they need advice on a purchase.

## Social proof in the AI era: the AI assistant does the research for the customer

In doing so, AI has also automated work the customer used to do themselves. Before buying a pram, a mortgage or a robotic lawnmower, people read discussions, went through reviews and asked friends. They have always looked for social proof this way; it just took hours or days. Today AI finds it for the customer in seconds. It reads the discussions, leafs through the reviews, weighs the opinions and returns a recommendation.

For a brand, this means one thing. If you already had social proof (because people discussed you, wrote reviews, recommended you), AI will see you too. If not, AI will not see you, because it simply has nothing to draw on.

## Where the customer’s new decision takes place

For a brand’s CMO, this means one specific thing. Your campaigns still build what they built before: brand recognition, associations, attributes. But the customer’s final decision, whether to choose your brand or a competitor’s, increasingly takes place in the customer’s conversation with AI. In a conversation you have no access to and in which nobody speaks for you, unless AI has chosen you as one of the recommended brands.

This does not mean Share of Voice has stopped working. It works, just at a different point in the funnel. When AI returns a shortlist of three brands to the customer, they subconsciously lean towards the one they already know from campaigns and that sounds trustworthy in their head. That was historically the job of SOV and it continues to do it. The catch is that you first have to get onto the shortlist. Whether you do depends on other factors, above all on social proof and the content about the brand that AI finds on the web. Share of Model is the metric that tells you how you are doing in this contest.

## Share of Model: definition, formula, what it measures

Share of Model is the answer to a question a CMO did not have to ask before: how often, and in what light, does an AI assistant mention your brand when a customer asks a question within your category. In Slovak, the term is rarely translated. A Slovak equivalent such as “podiel na modeli” or “podiel v modeli” sounds clumsy and has not caught on in practice, so the English original is used.

The formula is trivial and fits on one line:

> **Formula**
> 
> **SOM = (number of AI answers that mentioned your brand ÷ total number of AI answers in the given category) × 100**

Three elements of this formula need to be understood, because there are differing views on each of them in practice and each of them decides whether the resulting number means anything.

### What counts as a “mention”

A brand can appear in an AI answer in two different ways. Either it is mentioned by name in the text of the answer (“if you are looking for a powerful mower, take a look at the Stihl RMA 339”), or it is listed as a source from which AI drew information (a URL in the list of references below the answer). These are two different metrics and need to be measured separately. We will come back to this in the section on mistakes.

### What counts as an “AI answer”

Here it is important to know that there is no single AI. ChatGPT, Gemini, Perplexity, Claude and Google AI Overview behave differently, use different sources, weight them differently and cite them differently. Measuring SOM through a single tool is like measuring Share of Voice in a single medium. The value of the number is limited.

### What counts as a “category”

Your brand does not compete with the entire internet. It competes in the context of specific questions that make sense to your customers. For a garden equipment brand, that is not the question “what is a good brand”. It is dozens of real questions that people start typing into AI in March and April: “what mower should i buy for 800 square metres of lawn”, “is a robotic mower worth it or a petrol one”, “which chainsaw is good for occasionally cutting firewood”, “stihl or husqvarna for an amateur”.

Let us look at it with concrete numbers. Imagine you are the marketing manager of Stihl Slovakia. You take a sample of 100 real purchase questions from across the garden equipment category. You run them through three AI tools. You get 300 answers. You count how many of them mention your brand by name. Say it comes to 117. Your Share of Model is 39%.

This number says nothing on its own. It does not tell you whether 39% is a lot or a little. It only starts to make sense in context:

- **Compared with competitors.** In the same sample, Husqvarna might have 47%. Makita 12%. Bosch 28%. Mountfield 6%. Suddenly you see a ranking you will not see in Share of Voice, because they behave differently in the Slovak media market.
- **Compared over time.** Six months later, the same measurement will show whether your number is moving up or down. The trend matters more than the absolute value.
- **Broken down by tool.** If Perplexity mentions you in 62% of answers and ChatGPT in only 18%, it does not mean your SOM is 39%. It means you have a completely different image in two different brains, and you need to respond with different steps.

**Share of Model in garden equipment**

| Item | Value |
| --- | --- |
| Husqvarna | 47% |
| Stihl (you) | 39% |
| Bosch | 28% |
| Makita | 12% |
| Mountfield | 6% |

Source: illustrative example from this article

At Modrý koník, we carried out a similar measurement for April 2026, but for our own domain. The sample was 312 Slovak consumer questions entered into four different AI tools, 1,268 measurements in total. Our domain was cited as a source in 28 to 61% of the AI answers that contained sources, depending on which tool we measured. [4] The range between tools is large. For a brand manager who wants to measure their brand, this is important information even before the first measurement: expect a spread, not a single number.

## Share of Voice and Share of Model: two metrics for two stages of the purchase

Share of Voice and Share of Model are not rivals. They are two metrics that work at different stages of the purchase decision. SOV covers the stage in which the customer builds awareness and preferences. SOM covers the stage in which they choose a specific brand in a dialogue with AI. To make clear where each of them applies, it is worth putting them side by side.

| Comparison | Share of Voice | Share of Model |
| --- | --- | --- |
| What it measures | Your brand’s share of media attention in the category | Your brand’s share of AI answers in the category |
| Where it is measured | TV, print, digital, OOH, social, PR | ChatGPT, Gemini, Perplexity, Claude, Google AI Overview |
| Which moment in the decision it covers | Building awareness and associations before the purchase | A specific recommendation at the moment the customer asks |
| Who decides whether you are in the game | The media plan and budget | The AI model, based on the content that exists about you on the web |
| How quickly it can be changed | Weeks to months through a campaign | Months to a year through systematic work with content and structured data |

When looking at this table, it is important not to lose sight of two things.

**First: Share of Voice will not die. Only its ratio to Share of Model will change.** It still applies to categories where the customer decides on a product impulsively, based on an ad they have seen. It still applies to brand campaigns that build associations and emotion. It still applies in what was described in the previous section: when AI returns a shortlist of three brands to the customer, the customer subconsciously chooses the one they already know from advertising.

**Second: Share of Model is not just a different measurement of the same thing.** It is a measurement of a different stage of the purchase. SOV covers the moment when the customer is only looking around and building awareness. SOM covers the moment when they already want to buy and are looking for a specific recommendation. Performance marketing has tracked this later stage through its own metrics (CTR, conversions, CPA) and continues to do so. Share of Model, however, is the first metric that shows which members of the competitive set make it to this stage at all, and does so before the consumer clicks the buy button.

This leads to a practical point that is new for most marketing departments. Until now, the CMO had one main metric for the whole funnel and judged performance by it. From 2026, they have two. One tells them how the brand lives in people’s heads. The other tells them how the brand lives in the heads of the AI models that people consult more and more often. The two metrics can be in sharp conflict, to an extent most brands do not even suspect today.

## High Share of Voice, low Share of Model: when a brand is burning money

We come across cases where a brand has a high SOV because it has invested in television and outdoor advertising for years, yet its SOM is close to zero because there is no content about it on the Slovak web that AI could cite. AI simply does not recommend it, as if it did not exist. These brands burn money on building awareness that then gets lost in digital silence at the decisive moment of purchase. The opposite case also exists. A smaller brand with a modest media budget that systematically builds a content footprint on the web can have a higher SOM in AI than a competitor three times its size. It is precisely because these two numbers depend little on each other and can each move in their own direction that Share of Model is a separate metric, not a subset of SOV. Whoever does not measure it loses money where they cannot see it.

## How Share of Model is actually measured

Measuring Share of Model is not technically demanding, but it requires discipline at four points where most first attempts stumble.

### Step 1: A list of questions people actually ask

Marketing departments tend to make up questions. That is a mistake. The questions must be the way real customers type them into AI, with all the typos, half-sentences and informal phrasing. “What best mower 2026” is a bad question for measurement, because hardly anyone asks like that. “mower for a small garden under €500” is better. To compile the list, you can use data from Google Search Console, Google autocomplete suggestions and the “People also ask” section in Google search results, or input from people on the front line of sales and customer support.

For the Czech and Slovak market, there is one more source you will not find anywhere else. At Modrý koník, we have twenty years of consumer discussions and more than a million threads across the categories households care about: housing and kitchens, finance and mortgages, cars, fashion, gardening, holidays, food and dining. From these discussions we create structured AI summaries that pull concrete questions and answers out of each thread, in the exact language customers use when asking each other. For a brand in any ordinary consumer category, this is material that cannot be obtained elsewhere. These are not synthesised discussion topics; they are real purchase questions that customers asked one another before buying.

For a reliable measurement of one product segment (for example, lawnmowers), you need 20 to 40 questions. For a broader category (for example, all garden equipment) this number can stretch to 80 to 150. The rule is simple. There should be enough questions to capture the real breadth of what customers type into AI about the category, but not so many that you start inventing artificial variants of the same thing.

### Step 2: Run the questions through several AI tools

The minimum is the trio of ChatGPT, Gemini and Google AI Overview. These are the tools ordinary consumers use. Other tools (such as Perplexity or Claude) are fine, but only a small fraction of the typical consumer segment uses them. For brands that also sell to the corporate segment (IT services, office equipment, company fleets), it makes sense to add Microsoft Copilot as well, because in corporations that restrict ChatGPT it is often the only AI tool permitted on company computers. Technically, however, Copilot uses OpenAI models and the Bing index, so its results are often correlated with ChatGPT. Run each question through each tool separately. With 40 questions and 3 tools, that is 120 measurements for a narrow segment; with 100 questions and 3 tools, it is 300 measurements for the whole category.

### Step 3: Evaluate each answer

For each answer, you need to record two things for your brand. First, whether the brand is mentioned by name in the text of the answer (mention). Second, whether AI lists as a source a URL where the brand is written about (citation). These are two distinct metrics. A brand can be mentioned in the text while AI lists no source at all. Or the other way round, AI cites your website as a source but does not name you in the answer itself. Both situations occur in practice, but they mean different things and require different optimisation steps.

### Step 4: Normalisation and breakdown

When reporting the result, avoid aggregation that merges everything into a single number. “Our SOM is 39%” is a half-truth. The whole truth reads: “Our SOM is 62% in Gemini, 51% in ChatGPT and 18% in Google AI Overview, measured on a sample of 100 questions from the garden equipment segment in April 2026.” The second statement is hard to challenge. The first is not.

## Our Share of Model measurement at Modrý koník for April 2026

At Modrý koník, we repeat this procedure with the same discipline for our own domain. For April 2026, we carried out 1,268 measurements (312 Slovak questions across four AI tools). We recorded Modrý koník being cited as a source in a range of 28 to 61% of the AI answers that contained sources, depending on which tool we measured. [4] The range between tools is large and is itself an important finding. It shows that no single number describes the real situation.

For a brand that wants to measure SOM regularly, this is in fact a routine operation. All it takes is one process that is run once a quarter, with the result logged, compared with the previous measurement and interpreted. With a consistent method, you can track the trend, which is exactly what a CMO needs. Not an absolute number, but the direction in which the brand is moving in the heads of AI models.

## The most common mistakes in calculating Share of Model, which you will see in reports too

The market for measuring AI visibility began to open up in 2026, and with it came the first reports, presentations and slides that try to calculate and sell Share of Model. When reading them, keep your eyes open. Three mistakes recur so regularly that it is worth naming them before you see them in your own report.

### Mistake 1: Merging mention and citation into a single number

Some reports state “brand X has 47% SOM” without specifying whether this means a mention in the text of the answer or being listed as a source. In practice, these two things are often separate. AI can mention your brand by name in the answer (“I recommend brand X”) without listing any source for that information. Or, conversely, AI cites your website as a source in the list below the answer but does not mention the brand by name in the text of the answer at all. For a brand, these are two completely different situations and they require different steps.

In the first case (mention without citation), AI knows about you but does not say where from. On the Slovak market, three situations most often lie behind this. AI found you on the web in real time, but the model’s interface did not show the source URL to the user (a hidden citation, especially in ChatGPT). AI put the answer together from a synthesis of several sources and does not cite any specific one. Or, in rarer cases for local brands, AI has information about you hard-wired from its training data. GEO helps in all three situations, just on a different time horizon for each. In the first two, the result shows quickly, because AI draws from the web in real time. The third is an investment for the longer term. New generations of models are regularly retrained on fresh data, and the content you build systematically today appears in their knowledge, usually with a delay of twelve to twenty-four months. For Slovak brands, this means you have to build many mentions on a range of trustworthy domains. A single mention on your own website will struggle to get into AI’s “memory”, because during training models weight the quality and authority of sources, and Slovak content with low authority is at a disadvantage in this filter.

In the second case (citation without mention), AI cites your content but does not derive your brand from it as a recommendation in the answer. This is an important signal and needs to be read correctly. AI has recognised your content as a trustworthy and relevant source of information; otherwise it would not cite it. At the same time, however, it has decided that recommending your brand is not the right choice for that particular question. The most common reason in the Slovak environment is weak social proof. AI has read and recognised your content, but in parallel it looks for what others say about your brand, that is, reviews, discussions and ratings. If it finds little, or if the picture is neutral to negative, AI places your brand on the level of an information source, not on the level of a recommendation. In other words, AI knows who you are but does not have enough reason to tell the user “buy this”. Another reason may be marketing language that AI perceives as biased, or content that does not make clear what you recommend and for whom.

Reports that merge these two metrics will show you a falsely homogeneous picture. Always ask for the breakdown.

### Mistake 2: A single number for all tools without a breakdown

“Our SOM is 39%” is a metric that is technically correct but practically worthless. Without a breakdown by specific tool (ChatGPT, Gemini, Perplexity, Google AI Overview), no strategic decision can be made. If you have 62% in Perplexity and 18% in ChatGPT, you are not optimising one thing. You are optimising two. Each tool behaves differently, draws on different sources and weights them differently.

Our measurements at Modrý koník show that the difference between the strongest and the weakest tool for the same domain can be twofold to threefold. With a range like that, the average tells you nothing. It only says that there is a range of values whose average is this number. For a brand manager who has to prepare a plan for the next quarter, that is far too little. They need to know where the brand is weak and where it is strong, so they know where to put time and money.

### Mistake 3: A sample of questions chosen so that the result suits the agency

It happens that an agency measures SOM only for the brands it wants to make a case for to the client. The sample of questions is chosen so that the result sounds good, or, conversely, so that it sounds bad (depending on what the agency wants to sell). The protection is simple. For every measurement, ask three things: how many questions, which questions were selected and by what criteria. If someone cannot or will not answer, the number has no value.

## What this means for a brand

If Share of Model could be summed up in a single practical change, it would be this. Marketing is ceasing to be only a discipline that produces messages for people and is also becoming a discipline that produces content from which AI can derive a recommendation. These are two different jobs, and in brands that have already realised this, three concrete things have changed over the past year.

### Reporting no longer ends with the classic KPIs

Until now, the monthly report contained impressions, CTR, conversions, COS (cost of sale), and possibly CAC or LTV. From 2026, these metrics are joined by an SOM breakdown by AI tool, its development over time and a comparison with competitors. For a garden equipment brand, this means that the CMO presents sentences to the management board that were not heard there before. For example: “In Gemini we grew from 28% to 39% over the first quarter, in ChatGPT we stagnated, and in Google AI Overview we fell by three points because Husqvarna added a strong content strategy that influenced Slovak online outputs.” This is the new language in which marketing reports in 2026.

### Briefs to agencies include a new question

For every piece of content (an article, a video, a post, a campaign page), marketing asks: how will this content help AI find, understand and recommend the brand? The question is practical and has concrete criteria. Does the content clearly state what the brand does and where it is strong? Does the content include structured data that AI can read without having to interpret? Is the website where our content will be published authoritative enough for AI in our category? Is it linked to real proof (reviews, ratings, discussions)? These criteria are new for many agencies, and most are only starting to offer them in 2026.

### Investment in content is shifting from volume to what AI can use

Producing twenty product descriptions that read well for people is not the same as producing twenty descriptions that have correctly structured data, are linked to reviews and are framed by clear context. The second version is usable for AI; the first is not. In 2026, marketing is learning the difference between content for the reader and content for both the reader and AI at the same time. The second is more expensive to produce, but it directly increases Share of Model.

These three changes can be avoided. Many brands are still avoiding them, and for an understandable reason. Adding a new KPI means adding a new dashboard, a new person or a new supplier, and explaining it to management. In the short term, it is a cost. In the medium term (one to two years), however, it is the difference between AI recommending you at the decisive moment or not.

## When Share of Model does not yet make sense

To complete the picture, it is important to name the situations in which measuring Share of Model will not bring a brand a benefit proportionate to the effort. There are three.

### Categories where people do not ask AI

For B2B segments with low search volume (“industrial lubricants for food production lines”, “pressure transducers for waterworks”), practically no buyers ask ChatGPT or Gemini. The purchasing process here is personal, contractual and led by technical staff who do their research through specialised catalogues, suppliers and professional contacts. Investing in SOM measurement for such categories is premature in 2026. That may not be true in 2028, but it is today.

### Regional brands without a digital footprint

A local bakery, a regional car repair shop or a family guesthouse in the Tatras can do excellent business, but they are not something AI would have anything to say about. Most of them do not even have a proper website, they lack enough reviews, and there is no content about them that AI could evaluate. Their SOM will be close to zero, and optimisation would require an investment in digital content that exceeds what such brands need to deal with today. For them, it makes more sense to focus on classic local marketing channels and a high-quality Google Maps listing (exact address, opening hours, reviews, photos). For local searches such as “nearest bakery”, this is exactly what AI needs to know about them. But they do not need to measure SOM.

### Short-term tactical campaigns

If you want to push sales of a specific product for four weeks before Christmas, SOM will not help you on that horizon. AI models behave with inertia; their preferences form over months and years, not weeks. For a short promotional campaign, traditional tools, performance advertising and direct-sales channels are still better. SOM makes sense as a strategic metric, not a tactical one.

### Share of Model will not save a bad product

I will add one more perspective that is not often discussed. Share of Model is not a universal solution to marketing problems. But the opposite does not hold either, namely the idea that you can have a high SOM even with a bad product. Quite the contrary. When deciding which brand to recommend, AI looks for social proof, that is, reviews, ratings and discussions about your product. If it finds a negative picture, it will not recommend you. A poor-quality product shows up in SOM directly through bad reviews and discussions, sometimes faster than in sales. For brands with problems in their product or service, this means two things. First, investment in SOM will not achieve the desired result until the product itself improves, because AI will read about these problems in reviews sooner or later. Second, if the picture of the product in users’ real experiences improves, SOM will grow as a natural consequence, without further direct investment in AI visibility.

## The first step for a brand that wants to start

The cheapest and fastest way for a CMO to find out where their brand stands in Share of Model can be done in a single afternoon. You do not need many tools.

Put together a list of twenty real purchase questions from your category. Not marketing slogans, not the ones you would like to hear from a customer. Real questions that people type into AI when they are considering a purchase. Data from Google Search Console, Google autocomplete suggestions and the “People also ask” section in Google search results can help, or input from three people in sales.

Open ChatGPT and Gemini in a private browser window, without logging in. Also open Google for your questions and see what the AI Overview shows you above the organic results. This matters more than it sounds. If you are logged in to ChatGPT or Gemini, the models know your conversation history, your domain and your previous questions, and tailor their answers accordingly. For your own measurement, that is a disaster. As a brand owner, you are a completely different user for AI than an ordinary customer, and the answers you get will be biased towards you. A private window without logging in simulates an ordinary consumer who has opened AI to get advice on a purchase. The second thing: use the free versions of the models. The ordinary Slovak consumer does not pay twenty euros a month for ChatGPT Plus or Gemini Advanced. If you measure through a paid model, you are measuring a world your customer does not live in. Run those twenty questions through these three environments.

For each answer, watch three things. Did AI mention your brand by name? Did it cite any of your websites as a source? Did it mention competitors you did not expect, or in an order that surprised you?

After sixty AI answers (twenty questions times three tools), you will have a picture that no marketing report has shown you so far. You will see the ranking of brands in your category as AI sees it. You will see where your brand is strong, where it is weak, and where you are probably looking at a competitor you thought was smaller than you.

This exercise will not replace systematic measurement. A sample of twenty questions is small, the run is one-off, and the result is a snapshot, not a trend. But as a first-line diagnostic it works and tells you whether it makes sense to go further. If your brand did not come up even once in sixty AI answers, you have your answer. If it came up ten or fifteen times but a competitor twice as often, you also have your answer. And if it came up more than thirty times and still keeps you in the lead, you have the third answer, the most pleasant one, but even that is a reason to measure regularly and watch whether your position starts to crumble.

## Position in AI search is being shaped right now

Share of Voice will continue to exist. Nobody will bury it, because it has an irreplaceable role in building brand awareness. Share of Model has joined it as a second indicator for the second moment in the customer’s decision, the later one, where the choice is between three or four candidates and AI plays the role of a fantastic salesperson whom people trust.

Position in AI search is being shaped right now. AI models are currently forming preferred sources for every category and every brand. From what we see when tracking citation patterns over the past year, these preferences stabilise after a while, and replacing the preferred brand is slow. The first candidate to get in gains an advantage. The second has to fight for a place. The third will probably not get a chance. This is a hypothesis that does not yet have hard data in the Slovak context, but English-language markets are already showing citation patterns stabilising in certain categories.

For a brand that starts measuring Share of Model in 2026, it is a ticket into the game while the rules are still being written. For a brand that only starts paying attention to it once the media start writing about it and it becomes part of conference talks, it will be a race in which its team starts late. The competition will already be in the second half of the course. In Slovak marketing, we are currently in a window that is closing slowly, but it is closing.

## Frequently asked questions

### How does Share of Model differ from Share of Voice?

Share of Voice measures a brand’s share of media attention in a category and covers awareness building. Share of Model measures a brand’s share of AI answers and covers the moment when the customer chooses a specific brand in a conversation with AI. They are not rivals; they work at different stages of the purchase.

### How often should you measure Share of Model?

Regularly, for example once a quarter, using the same method and the same set of questions. The trend matters more than the absolute number.

### When does Share of Model not make sense?

In categories where buyers do not ask AI, for regional brands without a digital footprint and for short-term tactical campaigns. Share of Model will not save a bad product either, because AI reads reviews and discussions.

## Sources

1. [Gracker AI: Zero-Click Search Statistics 2025/2026 (April 2026)](https://gracker.ai/blog/zero-click-search-statistics) – Analysis of the share of Google searches that end without a click, across mobile and desktop devices, including a comparison of the classic SERP and Google AI Mode.
2. [McKinsey & Company: AI Discovery Survey 2025 (August 2025)](https://www.mckinsey.com/capabilities/quantumblack/our-insights) – Survey of AI assistant users’ preferences when looking for information.
3. [IAB Slovakia / gemiusAudience: Traffic monitor of Slovak websites (Q1 2026)](https://www.iabslovakia.sk/audit-navstevnosti/) – Monthly monitoring of real users and visits to Slovak media websites, comparing Q1 2025 and Q1 2026.
4. Modrý koník: Internal measurement of AI citation rate (April 2026) – Our own measurement of the visibility of the modrykonik.sk domain in AI tool answers: 312 Slovak consumer questions in four AI tools (ChatGPT, Gemini, Perplexity, Google AI Overview), 1,268 measurements in total.

---
Author: Peter Vidovič, GEO strategist. https://www.petervidovic.com/en/