What percentage of the AI market do you hold?
How much weight do you carry in AI when it comes to your market? For the first time, marketing—which has always relied on estimates based on indirect signals—has a metric to gauge a brand’s value as a quantity calculated within artificial intelligence. It’s called Share of Model, which lets you measure how much of that space belongs to you, compared to other brands in the same market, and how much you’re leaving to your competitors.
In fact, competition is now “mathematical”: for an AI engine like ChatGPT or Gemini, your brand is a point in a space of associations—linked to a category, competitors, and certain attributes—and is referenced with a higher or lower probability each time it constructs a response about your market.
What Is Share of Model
Share of Model measures how much weight a brand carries in the representation that an AI creates of a market, relative to other brands it associates with the same topic. It is a relative metric that makes sense when compared to other names associated with the same category, the same need, or the same cluster of intents.
It does not estimate the physical space on a page or the visible space on a SERP, but rather the position you occupy in the statistical representation through which the AI links markets, categories, needs, competitors, sources, and attributes.
When a user asks ChatGPT, Gemini, or Perplexity for advice on an industry, the AI does not list all the names it knows nor does it draw from a fixed list: it searches for brands, sources, products, alternatives, editorial sites, marketplaces, comparison sites, reputation signals, and content it considers relevant to the response. And it mentions some brands frequently, others rarely, and others never.
The Share of Model quantifies that disparity and indicates what share of that market, in the AI’s eyes, belongs to you. It doesn’t matter how strong you are in absolute terms, but how strong you are compared to your competitors on the same topic: you can appear frequently and still have a low share if the brands the AI mentions before you are stronger, just as you can appear less often but maintain a high share if you’re one of the category’s established leaders. The share reflects yourposition within the group, not your visibility in isolation.
In practical terms, take a thousand different questions related to the same intent—all about long-term car rentals, for example—and count how many times the AI includes your brand: if it happens in three hundred, your Share of Model in that area is 30%.
The signals that give weight to the brand
The final percentage is the synthesis of three factors to be examined separately, because a brand can be strong in one area and lacking in others, and understanding where the weakness lies tells you what to work on.
The first is presence—that is, the fact that the AI has linked you to that market: when the topic comes up, your name is part of the material it draws from. This is the starting point, because a brand that the model doesn’t associate with the topic at all won’t appear in any responses, and asking how much weight it carries makes no sense until it exists in that space.
The second is relevance—that is, the weight you carry once you’re there. Many brands coexist within the same topic, and the AI does not treat them all the same: it cites some as solid references, while keeping others in the background. Relevance measures the strength of your connection to the category compared to that of your competitors, and it’s the reason why you can be present yet still carry little weight if there are names alongside you that the model cites more consistently.
The third is predictive frequency—that is, the consistency with which you reappear. There are many questions about the same market, phrased in different ways, and a brand truly connected to the topic resurfaces in most of them, while one with a weak connection appears here and there and disappears elsewhere. It’s the nature of your market share that determines whether your position is stable or depends on the right question at the right time.
The three dimensions influence one another: you dominate the category, hold your own against competitors, and appear regularly. You may be present on a topic but carry little weight, because there are more established names alongside you; you may have a strong connection to the category but appear rarely, because in concrete responses the model tends to favor others. As long as they remain misaligned, the presence exists but has not yet solidified into a position.
The percentage is derived from the observed responses
The Share of Model can vary from ChatGPT to Gemini, from Perplexity to Google AI Mode, and can shift over time: each system operates on a different memory, uses different sources, and reconstructs the market with its own balance.
The internal weights of the model remain beyond direct observation; what we can measure is the frequency with which the brand reappears in the outputs generated within a controlled scope of prompts, engines, language, time period, and competitive players.
This is why a single response tells us very little. It may depend on the wording of the question, on a source retrieved at that moment, or on how the prompt narrows or broadens the context. A series of responses to related questions, however, reveals whether the brand truly belongs to the market’s representation or whether it appears only under fragile conditions.
You need to query the engine on the full set of questions that make up a market—informational requests, comparisons, and questions from those about to make a choice—and record how it treats you each time: whether your name appears, in what role, and alongside which competitors. From that behavior, repeated across many questions, you can derive an estimate of how strongly the AI associates you with that market. AI operates on probability, and the same repeated query may include your name one time and omit it the next. A reliable measure comes from consistency. If you appear consistently while the queries change in form and angle, that consistency is a sign that the AI keeps you tied to the topic; if you appear intermittently, your presence depends on a single lucky query rather than on an established position.
Marketing and the Search for a Brand Metric
For decades, marketing has measured brand strength by looking at people’s behavior: how many mentioned it, how many searched for it, and how many chose it when prices were equal. These were human signals, collected from the outside and interpreted—a survey revealed how many people remembered you, brand searches showed how many sought you out by name, and sales figures provided the final tally—from which one could infer the brand’s weight in the public’s mind. Today, Share of Model measures that same strength from within the model itself, but the idea behind it predates artificial intelligence: measuring a brand by the share of space it occupies, and not just by its immediate sales.
Share of Voice was the first response to this need. It measures what portion of a sector’s overall presence belongs to a single brand: if there are 100 mentions in a market and 20 of them are about you, your Share of Voice is 20%. It originated in media, advertising, and communications, and in the digital realm, it has expanded to include organic, paid, editorial, and social media reach. Its strength lies in a correlation observed in the field: brands that hold a Share of Voice higher than their market share tend to grow, while others tend to erode. In short, voice precedes sales.
Share of Search has applied the same logic to a cleaner, less manipulable metric: how many searches in an industry include a brand’s name, compared to competitors. It’s not advertising that drives this; it’s people’s spontaneous demand, and that’s precisely why many marketers use it as a telltale indicator: brand search share tends to shift before market share, and anticipating it helps predict where an industry is headed. It’s a proxy, not a certainty, but it’s one of the most reliable ones the industry has found.
Share of Model is the next step in this same family of metrics. It changes where the share is measured, but not the underlying question, which remains the same as ever: of all the space at stake, how much is yours? Share of Voice measured space in the media, Share of Search measured the share in explicit search queries, and Share of Model measures the position within the representation that the model constructs of the market.
Appearing often may not be enough
The term has already taken the international marketing world by storm.
Jellyfish, part of the Brandtech group, launched a platform called Share of Model™ in late 2024 to analyze how different LLMs perceive brands, products, and services, and to understand whether a brand manages to secure recommendations within generative models. Hallam describes Share of Model as the brand’s presence in LLMs relative to total category mentions; Tom Roach links it to the “Share of” family of metrics, explaining it as the ratio of brand mentions to total brand mentions in the same category, while also focusing on positive and negative associations.
Some operational interpretations, however, focus solely on appearances—how many times the brand appears out of the total—and mistake that percentage for the actual metric. But counting appearances, as we’ve seen, tells you that you’re there—not why you’re there—and the same frequency can indicate opposite positions.
That’s why the interpretation we propose brings together four factors that tell you whether your position is solid or about to falter—the very information that a percentage of appearances, on its own, hides from you.
Why and How the Percentage Changes
The Share of Model percentage only makes sense when the scope is specified. The number changes if the observed category, prompt cluster, search engine, language, time period, market, or competitive landscape under analysis changes.
A brand may have a 30% Share of Model on a cluster of informational prompts and a much lower share on prompts related to decision-making. The number, on its own, simply indicates that the brand appears with some regularity; it’s the broader context that explains where it appears and what that presence means.
Take a business management software provider. For questions where a person is seeking information—how do I organize staff shifts at a restaurant—that brand may occupy a significant space, because it has published guides, articles, and materials on those topics, and the model has learned to recognize it as an authoritative voice. Apply the same brand to questions involving a decision—what is the best software for managing shifts—and it may almost disappear, because there the model recalls the names it associates with the purchase decision, and those names are different.
These are two both valid Share of Model metrics, measured for the same brand on the same day, that tell two opposing stories: a strong presence where the user researches the problem, and a weak presence where they decide whom to entrust it to. The trap is to focus only on the first one, see a high number, and feel secure, while at the point that really matters—the one where the sale is closed—you’re losing ground without even realizing it. A share figure only makes sense if you know which stage of the journey it refers to.
The scope also includes the players in the comparison. The model can associate direct competitors, marketplaces, publishers, review sites, communities, comparison sites, and companies that the brand doesn’t consider rivals—but which the AI uses to build its response—with a category. Share of Model doesn’t just measure the game the company has in mind; it also shows the game the model is making it play.
That’s why the measurement must include both comfortable and uncomfortable prompts—the questions where the brand is already strong and those where it risks losing ground. The most useful share emerges when the observed field resembles the real market, not its most favorable version.
Value Depends on the Observed Field
Within the same scope, two mentions can have very different values. A brand can be used as a source, listed in a list, indicated as an alternative, recommended for a use case, associated with a strong attribute, or mentioned in passing. The percentage indicates how much space it occupies; the role indicates what kind of space it is.
Being a source while remaining invisible in the text produces a specific effect. The page contributes to the answer, but the brand does not enter the user’s memory. The site is useful to the algorithm and invisible to the market.
A textual mention places the brand within the answer but leaves the question of its role open. A name at the bottom of a list is worth less than a name around which the answer organizes the comparison. A generic mention is worth less than a reference tied to a specific need. A neutral mention may indicate presence, without giving the user a reason to choose.
A recommendation brings the brand closer to the decision. When the model flags a brand as a suitable option, it links it to a use case, a benefit, or a feature relevant to the query. This is where the share begins to demonstrate its competitive value: not just being there, but being used to guide the choice.
Then come the attributes. A brand can achieve a high share yet remain associated with characteristics that are inconsistent with its positioning: budget-friendly when it aims to be premium, simple when it aims to be advanced, local when it’s building a national presence, technical when it needs to be perceived as accessible. The share grows, but in the wrong direction.
Stability rounds out the analysis. A single survey may show a favorable presence; monitoring over time reveals whether the connection holds up. Share of Model becomes truly useful when it shows trends: where the brand is gaining ground, where it’s losing it, which clusters are consolidating, and which attributes are shifting.
How to Apply Share of Model in Real-World Work
Imagine discovering that your brand scores low on the questions that matter to your market. Taken on its own, that data doesn’t tell you what to do, because there may be opposing causes behind the same low presence: the model might not associate you with the topic at all; it might associate you but rate you lower than competitors; or it might recognize you very well but with attributes that push you toward a different market.
The question is why it weighs that way, where it loses strength, and what action can change the representation that the AI constructs. Each one requires a different approach, and SEOZoom’s SEO for AI tools help you figure out which of the three you’re dealing with before taking action.
What metric am I measuring? This is the first thing to clarify, because a metric read in the wrong context will lead you to fix a problem you don’t actually have. AI Prompt Research reconstructs the actual questions people use to present a need to the AI and groups them by intent, separating informational queries from those closer to a purchase—the Money Prompts. This is the moment when you choose whether to observe the market where you’re comfortable or where it hurts: measuring only the topics where you’re already strong produces a reassuring but false diagnosis.
How do I factor into the answers? Once you’ve defined the field, you now need to understand how the problem appears from your perspective. The AI Prompt Tracker tracks the responses to those questions and distinguishes what a simple count overlooks: when you’re used as a source without being named, when your brand appears in the text, when you receive a recommendation, in what tone, and alongside which competitors. And when a recommendation comes up, it tells you whether it stems from one of your strengths or merely from the weakness of a rival described as less suitable—the difference between a position that holds up and one that crumbles as soon as that rival recovers.
How much space is mine, and how much belongs to others? Here, the comparison becomes explicit. AI Competitor Analysis shows, through the AI Market Share of Voice, how presence is distributed among all the brands that the model associates with your market: you’re not looking at a disguised SEO ranking; you’re seeing which brand occupies more space in the generative narrative, which is stronger semantically, and where yours loses ground in terms of perception, trust, or the likelihood of being chosen.
How does it perceive my identity? If you’re entering the market but fear being misrepresented, the GEO Audit tells you which attributes and identity the AI associates with you, and where that perception diverges from what you want to be—the scenario, as we’ve already seen, of a high market share built on the wrong attributes.
How strong am I in answer engines?This is where the AEO Audit comes in, verifying how pages are used by answer engines that generate responses through live search. It’s a check on the present: how you’re treated when AI engines need to synthesize up-to-date information about your industry.
None of these numbers stand alone, and this is where Share of Model must be guarded against the illusion of being a magic metric. Your overall presence in AI shows where you rank across AI Overview, AI Mode, and conversational engines; the pages involved reveal which content underpins that presence; Search Console, Analytics, and brand search trends anchor it to real signals, because growth in model representation only truly matters if it correlates with traffic, branded search volume, and conversions.
Viewed this way—over monthly or quarterly periods rather than day by day—the share ceases to be an isolated data point and becomes a direction. If it grows on informational queries, you’re building brand recognition; if it grows on comparative queries, you’re getting closer to the decision-making stage; if it’s high but no page confirms it, the model is calling you out, and the site isn’t living up to the expectation it has created. It’s this analysis that tells you where to intervene—which content to strengthen, which attribute to correct, which cluster you’re losing ground in—even before telling you by how much.
Measurement is useful when it aids decision-making
Measuring Share of Model means making visible a part of the competition that until now has been difficult to interpret. As long as a brand’s weight within the responses remains an impression, you defend it based on intuition: you strengthen what seems weak, safeguard what seems important, and hope you’ve guessed right. When it becomes a readable number, you stop relying on gut feelings. You see on which topics the model treats you as a reference and on which it considers you peripheral or interchangeable; you understand whether a presence stems from your own strength or from a competitor’s temporary weakness; and you shift your efforts to where they’re needed rather than where it makes you feel secure.
From this analysis, concrete priorities emerge. If you’re absent from informational search queries, you need to strengthen your coverage of basic needs. If you’re losing out on comparative search queries, you need to create content better suited to the user’s choice. If you’re used as a source but not named, the site is helping to provide the answer without transferring recognition to your brand. If the model associates you with the wrong attributes, you need to address both proprietary and external signals.
Share of Voice has attempted to measure how much of the conversation a brand occupies. Share of Search measures how much explicit search demand is focused on the brand’s name. Share of Model adds the missing layer to this framework: the share within the representation that machines currently use to reconstruct markets and guide choices. It does not replace traffic or conversions, which remain the ultimate proof; rather, it measures the groundwork upstream, where the choice begins to take shape. And when that point becomes observable, it also becomes manageable.
