--- language: en title: "What Is Share of Voice in AI Responses" description: "Learn how to calculate AI Share of Voice, distinguish share from presence rate, and compare mentions, recommendations and citations using the same set of questions and brands." canonical: "https://tropk.ai/en/blog/what-is-share-of-voice-in-ai-responses" last_modified: "2026-10-12T16:05:23.873Z" --- # What Is Share of Voice in AI Responses Learn how to calculate AI Share of Voice, distinguish share from visibility rate, and compare mentions, recommendations, and citations using the same set of questions and brands. [Luiz Gardelin]() Published on 2026-07-01T02:34:20.000Z Categorias: GEO/AEO [**AI Share of Voice](/en/resources/glossary/ai-share-of-voice) (AI SOV)** is the metric that measures how often your brand appears in [generative AI](/en/resources/glossary/generative-ai) responses compared with competitors that appear in those same responses. It is Share of Voice in a world where search has become a conversation. The formula is simple, but its denominator must be explicit. A visibility rate divides responses featuring the brand by the total number of responses. Share of Voice divides the brand’s appearances by the appearances of all eligible brands in the same set. Confusing these measures makes a brand look more dominant than the data shows. Meanwhile, [ChatGPT sent 6.1 million visits to Brazil’s 10 largest e-commerce sites](https://exame.com/marketing/ia-generativa-leva-6-mi-ao-e-commerce-e-tem-mercado-livre-na-lideranca/) between January 2023 and August 2025. That traffic was real. And it was concentrated among brands that already appear. This article explains what AI SOV is, how to distinguish it from visibility rate, and how to measure organic share using documented questions, brands, and execution conditions. ## Key points - AI Share of Voice measures a brand’s share of appearances among eligible brand appearances in the same AI responses. - Presence rate uses total responses as its denominator; SOV uses total brand appearances and cannot be inferred from presence frequency alone. - Define eligible brands, consolidate name variations and count each brand no more than once per response to avoid distortion from repetition. - Measure recommendations, mentions and citations separately; exclude ads from the organic share indicator. - Compare platforms and periods under equivalent conditions, with stable questions and repeated measurements; a single ranking or percentage does not demonstrate leadership. ## The Share of Voice You Already Know **Share of Voice (SOV)** is the metric that measures a brand’s relative presence in a market or channel compared with competitors. What share of the market’s total conversation is about you? ### The Metric That Measures Who Dominates the Conversation In the classic model, the formula is straightforward: SOV (%) = your mentions divided by total market mentions, multiplied by 100. If the market generated 1,000 category mentions and 200 were about your brand, your SOV is 20%. The logic of relative share can be applied to paid search, organic presence, media, and social media, always with a defined unit and universe. Relationships observed in historical media and market-share studies should not be transferred as a causal formula for AI SOV. Presence in a response is an exposure signal, not a direct measure of revenue. The same question about relative presence can be asked of [AI responses](/en/resources/glossary/ai-response), with specific sampling and comparison considerations. ## What Changes When Search Becomes a Conversation Traditional search presents results and may include ads. Organic and paid positions are governed by distinct mechanisms; paying more does not guarantee any position in an auction. Users can click, read an answer on the page itself, or continue searching. [AI search](/en/resources/glossary/ai-search) can generate a direct answer, a comparison, or a list of options, with supporting references. Ad-supported experiences also exist: [OpenAI says](https://openai.com/index/new-chatgpt-ads-format-and-measurement/) that paid placements in [ChatGPT](/en/resources/glossary/chatgpt) are identified and distinct from responses. This article addresses organic AI SOV: mentions and recommendations in response content should be measured separately from paid placements. Buying advertising does not demonstrate an increase in a brand’s organic share. ### From Ranking to Citation The numbers show the scale of the change in Brazilian consumer behavior: - [ChatGPT has 310.67 million monthly visits in Brazil](https://www.meioemensagem.com.br/marketing/chatgpt-domina-mercado-de-inteligencia-artificial-no-brasil) – growth of 124.58% over 12 months through August 2025 - [58% of consumers already turn to generative AI for product and service recommendations](https://hbr.org/2025/06/forget-what-you-know-about-seo-heres-how-to-optimize-your-brand-for-llms) , compared with 25% in 2023 - A response can be read without a site visit; the zero-click percentage should be tracked by platform and methodology, without applying a single rate to all AI experiences A response can satisfy a question without generating a click, but zero-click metrics depend on the platform and sample. So track organic presence, cited links, and observed traffic as separate indicators. Don’t confuse a lack of clicks with a lack of influence, or paid presence with an organic recommendation. ### The Problem Brazil Has Yet to See To understand the Brazilian market, the best starting point is a sample relevant to your category, in Portuguese, and with local conditions. Percentages from a survey about [GEO](/en/resources/glossary/geo) adoption or monitoring do not automatically show what proportion of all brands appear in responses. **GEO** ([Generative Engine Optimization](/en/blog/what-is-geo-generative-engine-optimization-complete-2026-guide)) brings together practices to improve the information available, its discoverability, and its usefulness in AI experiences. The effect needs to be monitored; inclusion is not guaranteed. Meanwhile, ChatGPT sends real traffic. Zara, for example, [leads in AI visibility for women’s fashion in Brazil](https://acontecendoaqui.com.br/comunicacao/zara-lidera-visibilidade-em-respostas-de-ia-sobre-moda-feminina/) – and this is not a theoretical research finding. It’s about which brand appears when someone asks ChatGPT where to buy clothes. Starting to measure now can reveal inaccurate descriptions, important sources, and information gaps. These observations guide content and reputation actions. There is no evidence that a [citation](/en/resources/glossary/ai-citation) automatically generates new citations through a learning cycle across all models. ## What Is Share of Voice in AI Responses **AI Share of Voice (AI SOV)** is how often your brand appears in generative AI responses, divided by the total appearances of all brands in the same [prompts](/en/resources/glossary/prompt), multiplied by 100. Formula: AI SOV (%) = appearances of your brand divided by total appearances of eligible brands in the same responses, multiplied by 100. First define what counts as an appearance. One useful convention is to count each brand no more than once per response, consolidating name variations for the same [entity](/en/resources/glossary/entity). The denominator should include the category brands actually mentioned, with criteria for excluding irrelevant names, confusions, or products not part of the comparison. If no eligible brands appear, the SOV for that sample is undefined: don’t present it as zero market share. ### Definition and Difference from Traditional SOV The fundamental difference is not just where the metric is measured. It’s the environment in which it operates: | Aspect | Traditional SOV | AI Share of Voice | | --- | --- | --- | | What it measures | Positions, impressions, clicks | Mentions in generated responses | | Environment | Structured SERP, ads | Dynamic, probabilistic responses | | What influences it | Organic systems or auctions, depending on the metric | Models, search, sources, and response context | | Is the position stable? | Can vary by query and context | Can vary between runs | | How to measure | Traditional tools (SEMrush, Ahrefs) | Specialized tools (Profound, Scrunch, Peec AI) | | Paid presence | Ads measured separately | Ads measured separately from organic AI SOV | With traditional [SEO](/en/resources/glossary/seo), you can know with reasonable certainty that if you rank #1 for a keyword, you will appear for that search. With AI, the response changes each time someone asks the same question. There is no fixed position. There is an appearance frequency over time. ### The Two Types of AI SOV No One Distinguishes Not every [brand mention](/en/resources/glossary/ai-mention) represents the same event. There are two distinct mechanisms: **Entity-based SOV** (entity-based SOV – recommendation): AI recommends your brand as a solution to a problem. When someone asks “which financial management tools do you recommend for SMBs?” and ChatGPT names your company as an answer, that is entity-based SOV. It is the commercial goal – you are the answer, not just a source. **Citation-based SOV** (citation-based SOV – source): AI cites your publication as an information reference. When [Perplexity](/en/resources/glossary/perplexity) answers a question about market trends and includes a link to a report on your blog, that is citation-based SOV. It builds authority and credibility, but does not necessarily recommend the product. In practice, a citation as a source and a recommendation as a solution serve different goals. Neither guarantees conversion. Measuring only one of the two means seeing half the picture. ## The AI Share of Voice Formula (and the Mistake Most People Make) The formula itself is simple. The key is to recognize the metric and the comparison universe. Different denominators can produce different numbers even when the response has not changed. ### Visibility Rate and SOV Use Different Denominators One formula used to measure visibility is: appearances of your brand divided by the total number of prompts tested, multiplied by 100. It seems reasonable. But it has a fundamental problem: it does not account for how many other brands appear in the same responses. Numerical example: in 100 responses, your brand appears in 40. The visibility rate is 40%. Suppose that across the whole set, there are 600 appearances of eligible brands, including 40 for your company and 560 for other brands. Your AI SOV is (40 ÷ 600) × 100 = 6.7%. Without the total appearances of other brands, you cannot derive SOV from the visibility rate. The first calculation is **visibility rate** (visibility rate in responses); the second is **AI Share of Voice** by appearances. Check each tool’s labels and documentation before interpreting a score. The distinction matters: visibility rate measures “your brand appeared in this prompt” – AI SOV measures “of all brand mentions in this set of prompts, what percentage were yours”. [Waikay](https://waikay.io/ai-brand-visibility-guide/share-of-voice/) documented this methodological distinction in detail: the correct formula uses brand mentions in the numerator and total mentions of all brands in the denominator, not the number of prompts. The difference matters because you may think you dominate a space while sharing it equally with 4 competitors. ### The Closed Denominator Problem The second mistake is subtler and more common. Some AI SOV platforms let you manually list your competitors when setting up monitoring. You define the comparison pool. This is the **closed denominator** problem. What this means in practice: the calculated SOV is share within the group you specified, not share of the entire conversation AI is having about the topic. This selection can be useful for tracking a specified group, but it has three limitations: First, it can be manipulated. Exclude the biggest competitors from the list and your share rises artificially. The numbers improve without anything real changing. Second, it hides new entrants. A new competitor that AI starts mentioning frequently will not appear in reports if it was not on the initial list. You may be losing ground without realizing it. Third, it is not comparable across companies. Each company builds its list using its own criteria. “40% AI SOV” at one company says nothing about the real market when compared with “35% AI SOV” at another. To measure share across all observed brands, use an open denominator: extract competitors directly from AI responses – the brands AI actually mentions in the context of your prompts. This reflects the model’s reality, not your perception of your competitors. ### Why a Single Position Does Not Demonstrate Leadership Instinctively, appearing first on a recommendation list seems better than appearing third. For generative AI, that intuition is misleading. Rand Fishkin and Patrick O’Donnell ran [2,961 runs of 12 prompts](https://waikay.io/ai-brand-visibility-guide/share-of-voice/) in ChatGPT, [Claude](/en/resources/glossary/claude), and Google’s generative systems. The conclusion: as Waikay reports, the probability of repeating exactly the same ordered list of brands was less than 1 in 1,000 in that test. This applies to the complete list, not to the chance of an individual brand maintaining its position. LLMs don’t retrieve lists from databases. They generate text token by token, based on probability distributions that vary with each run. The order changes. The stability of presence also needs to be measured; it should not be assumed for every category or assistant. Frequency across multiple runs provides more context than a position observed just once. Position can also be tracked, provided its distribution and variation are shown. Appearing in 70% of 200 runs of the same prompt is useful information. Appearing in first place in one specific run is not. ## Why this matters for your brand now The Brazilian AI search market is already generating real commercial traffic. But the impact of AI SOV goes beyond direct traffic. ### AI as an invisible filter in the buying journey AI is acting as a gatekeeper before buyers visit any website or speak to any salesperson: - [89% of B2B buyers will use generative AI in purchasing decisions in the next 2 years](https://mundodomarketing.com.br/decisoes-mediadas-por-ia-e-a-mudanca-na-jornada-de-compra) – Forrester, 2025 - [39% of consumers have already used AI in an online purchasing process](https://mundodomarketing.com.br/decisoes-mediadas-por-ia-e-a-mudanca-na-jornada-de-compra) – Adobe, 2025 - [73% of B2B buyers use AI for research today](https://www.shadow.inc/resources/ai-search-visibility-pr) – Shadow.inc, 2026 The pattern that is emerging: buyers ask AI before visiting any website. AI filters, recommends, and rules things out. If your brand doesn’t appear in a relevant query, it may miss a consideration opportunity in that context. This does not prove exclusion from the buyer’s entire journey. [B2B brands appear in fewer than 30% of relevant category queries](https://llmpulse.ai/blog/glossary/share-of-voice/), regardless of traditional SEO performance, according to an analysis of 2.4 million AI responses by [LLM](/en/resources/glossary/llm) Pulse. A good Google ranking does not guarantee an appearance in ChatGPT. ### AI traffic in Brazil is already real These figures are not future projections: - [ChatGPT sent 6.1 million visits](https://exame.com/marketing/ia-generativa-leva-6-mi-ao-e-commerce-e-tem-mercado-livre-na-lideranca/) to the 10 largest Brazilian e-commerce sites between January 2023 and August 2025. Mercado Livre led this traffic. - Zara leads [AI visibility](/en/resources/glossary/ai-visibility) for women’s fashion in Brazil – when someone asks ChatGPT where to buy women’s clothing, Zara is the most frequent answer. - [A MonitorCNPJ client increased its presence in ChatGPT from 41% to 78% in 3 months](https://monitorcnpj.com.br/solucoes/presenca-em-ia/) after taking a structural approach to its content. This traffic is growing alongside the adoption of AI tools. And it is concentrated among the brands that appear in the responses. ## How each AI engine works differently ChatGPT, Perplexity, and Google [AI Overviews](/en/resources/glossary/ai-overviews) are different experiences that may select sources and generate different responses. This is why presence and SOV should be calculated by platform, keeping the language, questions, and conditions consistent. An improvement hypothesis for one platform should not be treated as a universal rule. A [Profound study of 680 million citations](https://www.tryprofound.com/blog/ai-platform-citation-patterns), from August 2024 to June 2025, shows differences among the sources cited. The Wikipedia (47.9%) figure for ChatGPT and Reddit (46.7%) for Perplexity represent their share only among each platform’s ten most cited sources. Across all citations in the sample, the figures are 7.8% and 6.6%. It is important to preserve the denominator and period; these data do not demonstrate the current behavior of every query. ### ChatGPT – verifiable content and retrieved sources In Profound’s historical sample, Wikipedia was ChatGPT’s most frequently cited source: 7.8% of the total observed. The encyclopedia’s presence is consistent with the value of factual, verifiable content, but does not demonstrate a mandatory formatting rule or a fixed preference for any question. To be found when ChatGPT uses search, keep content publicly accessible, use clear links, and cite sources that support your claims. There is no official confirmation that 87% of its citations must come from Bing, or that a Bing ranking alone determines a brand recommendation. The number of brands, section length, and presence of links vary with the question and mode of use. Observe these events in the responses you collect; don’t impose a fixed number of brands or words as a ChatGPT standard. ### Perplexity – research with sources and context In the same Profound sample, Reddit was Perplexity’s most frequently cited source: 6.6% of the total observed, and 46.7% among the ten most cited sources only. These denominators are not interchangeable. The distribution of sources varies by topic and collection period. In practice, investigate the sources Perplexity cites for questions relevant to your business. Clear, verifiable, and up-to-date content when needed helps readers; participation in communities should be authentic and relevant. A strong Google result does not allow you to infer your share of presence in Perplexity. Don’t adopt a fixed number of brands per response or a universal percentage of links. Comparison, definition, and research questions can produce different formats. Assess citation frequency, context, and source relevance in your own sample. ### Google AI Overviews – SEO fundamentals and eligibility To appear as a supporting link in AI Overviews or [AI Mode](/en/resources/glossary/ai-mode), the [official Google documentation](https://developers.google.com/search/docs/appearance/ai-features) requires the page to be indexed and eligible to appear with a snippet. It does not require a top-10 ranking or special schema. [Internal links](/en/resources/glossary/internal-links), useful content, and information available in text are among good practices; images and videos can complement it when relevant. [An analysis of 75,000 brands by Ahrefs](https://ahrefs.com/blog/ai-overview-brand-correlation/) found a correlation of 0.326 between Domain Rating and visibility in AI Overviews – significant, but not dominant. This correlation does not prove that DR has a causal influence. [Schema markup](/en/blog/schema-markup-for-ai-visibility-the-7-types-that-matter-and-how-to-implement-each) should reflect visible, accurate information and does not guarantee selection for a response. ### Comparison table of the three engines | Aspect | ChatGPT | Perplexity | Google AI Overviews | | --- | --- | --- | --- | | Most cited source in the Profound sample, Aug 2024–Jun 2025 | Wikipedia: 7.8% of total | Reddit: 6.6% of total | Reddit: 2.2% of total | | Number of brands and links per response | Varies with the question and mode of use | Varies with the question and mode of use | Varies with the query and summary display | | Universal Google top-10 requirement | Not confirmed | Not confirmed | Not required by official documentation | | What to measure | Mentions, recommendations, and citations | Mentions, recommendations, and citations | Presence and links in the AI section | Historical sample source: [Profound](https://www.tryprofound.com/blog/ai-platform-citation-patterns). The table describes a sample of citations, not a list of official selection factors. For reliable AI SOV, measuring all three separately is not optional. One aggregated figure hides variations that can completely change the diagnosis. ## How to measure your Share of Voice in AI No tool does this completely automatically and without configuration. But the manual process is simpler than it seems – and you can get started at no cost. ### Step 1 – Build your prompt set The [prompt set](/en/resources/glossary/prompt-set) defines what you are measuring. The formula is simple. The prompt set is what separates a measurement that says something from one that says nothing. [Alex Birkett](https://alexbirkett.com/ai-share-of-voice/), from Omniscient Digital, organized this into three layers: **Layer 1 – Category prompts:** “What are the best [your category] for [customer profile]?” – maps where your brand appears compared with the entire category. **Layer 2 – Specific-need prompts:** based on how your actual customers describe the problem – forums, reviews, support questions. “How do I [specific problem] without [common constraint]?” – captures long-tail queries that AI answers more frequently. **Layer 3 – Comparison prompts:** “[Your brand] vs [Competitor A]” – reveals how AI positions the brand in a direct comparison. An editorial starting point could be 20 to 30 questions, but this number does not guarantee statistical consistency. Audience and buying-situation coverage matter more than a fixed quantity. Test the same prompts on the three engines separately. ### Step 2 – Run and record For each prompt, in each engine, record in a spreadsheet: - Date and time of the run - Engine name (ChatGPT, Perplexity, Google AI Overviews) - All brands mentioned, in order - Whether your brand was mentioned (yes/no) - Whether it was a direct recommendation (entity-based) or a source citation (citation-based) Repeating each question, for example three to five times at first, helps you observe variation, but this number does not guarantee representativeness. The randomness of LLMs means that a single result is not representative – the average across multiple runs is what matters. Suggested frequency: weekly to detect variations, daily only if there has been a recent content change or an event relevant to the brand. ### Step 3 – Calculate the right formula With the data collected: 1. Add up your brand’s appearances across all responses, counting it no more than once per response 2. Add up the appearances of all eligible brands in the same responses, using the same counting rule 3. Divide the first by the second and multiply by 100 This is AI SOV based on appearances within the defined universe. Calculate it separately for each engine—the numbers differ, as do the actions to take based on them. Compare entity-based (direct recommendations) and citation-based (source citations) separately. A strong citation-based result but weak entity-based result means you are an informational authority but are not being recommended as a solution. ## Benchmarks: what is a good number? It depends on the context—category, market maturity, product type. But there are useful reference points. ### International reference points Articles from vendors and consultants may present indicative ranges, but they are not comparable without knowing the denominator. A 40% [presence rate](/en/resources/glossary/ai-presence-rate) is not equivalent to 40% AI SOV. Interpret any range alongside the number of eligible brands, the question set and the data collection conditions. A category with ten brands appearing equally will have shares close to 10%; with two, close to 50%. This shows why there is no universal leadership range. Compare shares within the same universe and investigate the commercial context of the responses, as well as the percentage. ### The Brazilian context changes the calculation For Brazil, use questions in Portuguese with local situations and purchasing criteria. A competitor not being monitored does not prove that its SOV is zero. A positive number also does not guarantee a competitive advantage: the value depends on the questions, the comparison and what the AI says about the brand. Cases published by vendors, such as [MonitorCNPJ GEOscore](https://monitorcnpj.com.br/solucoes/presenca-em-ia/), should be read in light of their methodology and scope. Presence percentages reported in a case are not automatically AI SOV and do not promise the same result for another company. ## Tools for measuring AI Share of Voice ### International tools | Tool mentioned | Use described in the article | Current platforms | Current pricing | | --- | --- | --- | --- | | [Profound](https://www.profound.com/) | AI presence monitoring and analysis | Check with the vendor | Consult the vendor | | Scrunch | AI presence monitoring and analysis | Check with the vendor | Consult the vendor | | Peec AI | AI presence monitoring and analysis | Check with the vendor | Consult the vendor | | Otterly.AI | AI presence monitoring and analysis | Check with the vendor | Consult the vendor | | Semrush Enterprise | AI presence monitoring and analysis | Check with the vendor | Consult the vendor | ### Brazilian tools Language and market context should guide your choice of tool, regardless of its origin. **Oria** is one of the tools mentioned in the Brazilian context for investigating brand presence and the sources used in responses. Before choosing a solution, confirm its current offering, methodology and data collection conditions. An investment announcement does not prove coverage or measurement quality. [**MonitorCNPJ GEOscore**](https://monitorcnpj.com.br/solucoes/presenca-em-ia/) is among the Brazilian solutions mentioned for tracking brand presence. Confirm which platforms are monitored and how the system distinguishes mentions, recommendations and sources; the importance of Reclame Aqui should be assessed within your question set, without assuming a fixed weighting. \*\*[Ipsos Synthesio GEO](https://www.ipsos.com/pt-br/desbloqueie-a-visibilidade-da-sua-marca-na-era-da-ia-com-o-ipsos-synthesio-geo) \*\* is the enterprise option, with simulated prompts from real Brazilian consumers and analysis of brand image in responses. For companies with high volumes and a need to segment by region or consumer profile. For an [AI visibility analysis focused on the Brazilian market](/en), [Tropk.ai](/en) organizes the analysis of questions, responses and sources, as well as content auditing, to guide the assessment of the brand and the information on the site. When comparing tools, assess language, location, purchasing contexts and relevant sources in Brazil, such as reputation platforms when they appear in responses. This matters for brands operating exclusively in the domestic market. ## Sources for further reading - [Original Tropk article]() - [Waikay: AI Share of Voice denominators and methodology]() - [Profound: citation patterns, August 2024 to June 2025]() - [Ahrefs: AI visibility correlations across around 75,000 brands]() - [Google Search Central: AI features and your website]() - [OpenAI: ChatGPT ad format and measurement, October 2026]() ## Frequently asked questions ### What is AI Share of Voice in AI responses? AI Share of Voice is a brand’s share of total eligible brand appearances in a set of AI responses. One formula is brand appearances divided by the total appearances of all eligible brands, multiplied by 100. The indicator should specify the questions, platforms, period, counting rule and comparison universe. ### What is the difference between AI Share of Voice and AI presence rate? AI presence rate divides the number of responses mentioning a brand by the total responses analyzed. AI Share of Voice divides a brand’s appearances by the total appearances of eligible brands in the same responses. If a brand appears in 40 of 100 responses, its presence rate is 40%; to calculate SOV, you still need to know how many times other brands appeared. ### How do you calculate AI Share of Voice manually? To calculate AI Share of Voice manually, choose relevant questions and record responses by platform and period. Identify eligible brands, consolidate names belonging to the same entity, and count each brand no more than once per response. Divide your brand’s appearances by the total appearances of all eligible brands and multiply by 100. If no brands appear, SOV is undefined for that sample. ### Is it better to calculate AI Share of Voice using selected competitors or all observed brands? Calculating with selected competitors shows your share within that specified group. Calculating with all observed eligible brands shows your share in the universe produced by the responses and may reveal new competitors. Both views can be useful, but they need explicit names and denominators; changing the group changes the result even if the responses do not change. ### What AI Share of Voice result proves a brand is a leader? No isolated AI Share of Voice percentage proves universal leadership. The result depends on the number of brands, the questions, the category, the platform and the calculation rule. A 20% share means something different in a group of two brands than in a group of ten. Compare within the same universe, track trends and check the context of the recommendations. ### Are ads included in the organic AI Share of Voice calculation? Ads should not be included in the organic AI Share of Voice calculation. Mentions and recommendations in response content need to be separated from paid or sponsored placements. A brand appearing in advertising shows paid exposure in that experience, but does not prove the AI has recommended the company organically or that its organic share has increased. ## About the author [Luiz Gardelin]() Luiz Gardelin is the founder of Tropk.ai and works in GEO/AEO, product vision, strategy, and growth, helping brands understand and improve their presence in AI responses. --- [All pages in Markdown]()