AI share of voice (AI SOV) is the percentage of brand mentions a company receives across AI-generated responses, relative to all brand mentions for that category on those platforms. The formula: AI SOV = (your brand mentions / total brand mentions across tracked prompts) x 100. Unlike traditional share of voice tied to ad spend or media coverage, AI SOV quantifies how often a brand appears when buyers ask ChatGPT, Perplexity, Gemini, or Google AI Mode about solutions in a given category.
Last updated: June 28, 2026
The canonical definition of this term, with the Machine Relations measurement contract and its interpretation rules, is AI Share of Voice in the MR glossary. This page is retained as an earlier reference treatment of the same question.
June 2026 update: Google made AI SOV explicit in Merchant Center #
AI share of voice moved from an experimental brand-monitoring metric to a platform-native reporting concept in May 2026. Google Merchant Center's AI performance insights now describes share of voice as brand visibility across AI-driven shopping experiences in Search and Gemini, benchmarked against similar brands, with product-term share of voice and journey-stage reporting for discovery, evaluation, and purchase (Google Merchant Center Help, May 27, 2026).
That change does not settle the formula problem. It makes the denominator problem more important. A platform-native SOV report can show how often one brand appears in a specific surface, but cross-engine AI SOV still requires open-set measurement across ChatGPT, Perplexity, Gemini, Claude, Google AI Mode, and AI Overviews. Brands should separate platform-specific SOV from category-level SOV before using the number in board reporting.
AI share of voice defined #
AI share of voice is a competitive metric, not an absolute one. A brand mentioned in 40 of 100 AI responses has a 40% visibility rate, but that number says nothing about competitive position. If three competitors each appear in 60 of those same responses, the brand's actual share of the conversation is far smaller than 40% suggests.
The distinction matters because AI-generated answers operate differently from traditional search results. When a buyer asks an AI engine "what is the best CRM for startups," the response typically names four to eight brands in a single answer. Every brand mentioned in that response competes for the same attention. Traditional search produced 10 blue links where position determined visibility. AI responses produce synthesized answers where mention frequency and context determine observed visibility.
Gartner projected a 25% decline in traditional search volume by 2026 as buyers shift to AI-powered answers (Gartner, 2024). Bain found that 80% of search users now rely on AI summaries at least 40% of the time (Bain, 2025). Those findings support a behavior-shift context for AI visibility measurement; they do not prove that an AI SOV number forecasts market share.
Three formulas in use, two of them flawed #
Not all AI SOV calculations measure the same thing. Three distinct formulas are in active use across the measurement ecosystem, and confusing them produces misleading competitive intelligence.
| Formula | What it measures | Limitation |
|---|---|---|
| Brand appearances / total prompts tracked | Visibility rate (how often you show up) | Ignores competitive density; a mention alongside nine competitors counts the same as a solo mention |
| Brand mentions / total brand mentions (closed set) | Share within a predefined competitor list | The denominator is manually defined; emerging competitors are invisible until someone adds them |
| Brand mentions / total brand mentions (open set) | Competitive share derived from AI responses | Requires entity normalization across product names, abbreviations, and brand variants |
The first formula is not share of voice at all. It measures visibility rate: the percentage of prompts where a brand appears. A brand showing up in 30 of 100 prompts reports "30% share of voice," but if each response mentions six brands, the actual share of conversation is roughly 5%. Visibility rate and share of voice diverge by an order of magnitude in crowded categories.
The second formula introduces structural bias through the closed denominator. When a platform requires users to manually define their competitor set, the resulting SOV is calculated within that predefined pool. If a new brand begins appearing frequently in AI answers but is not on the tracking list, the metric will not capture the shift. In fast-moving categories, this means competitive threats emerge in the data long before they appear in dashboards. The denominator must come from the AI responses themselves, not from a configuration screen.
The third formula is the most defensible. It derives the competitive landscape directly from AI outputs, identifying every brand and product mentioned across responses and using the full set as the denominator. This produces an open denominator that reflects the market as AI engines actually represent it.
How AI share of voice differs from traditional SOV #
| Dimension | Traditional SOV (paid/organic) | AI share of voice |
|---|---|---|
| Input signal | Ad spend, keyword rankings, media impressions | Entity recognition, source coverage, content structure, and model-specific retrieval behavior |
| Channel | Google Ads, organic SERP, social media | ChatGPT, Perplexity, Gemini, Claude, Google AI Mode |
| What it observes | Media or search visibility in a defined channel | Brand mentions in AI-generated answers for a declared prompt panel |
| Stability | Relatively stable between measurement periods | Varies across model runs, platforms, and prompt phrasing |
| Position value | #1 ranking captures disproportionate clicks | Mention order within a single response is unstable across generations |
The most important structural difference: AI SOV is observed, not bought directly. Paid search SOV scales with budget. Organic SOV tracks ranking improvements from SEO investment. AI SOV records how consistently AI models mention a brand for specific category prompts. Available public source-composition studies suggest that unpaid and third-party sources are often prominent in AI citations, but source mix is not the same as placement causality.
Muck Rack / Generative Pulse public copy describes a broad earned-media taxonomy rather than a placement-causality test. The available public material references 84% under that broad earned-media grouping and 27% journalism in the May 2026 edition (Muck Rack/Generative Pulse, 2025). Ahrefs' study of 75,000 brands reported that brand web mentions correlate 3x more strongly with AI Overview visibility than backlinks (0.664 vs. 0.218 correlation coefficient), while explicitly leaving causation unproven (Ahrefs, 2025).
AI share of voice vs. share of citation #
AI share of voice and share of citation measure related but distinct phenomena.
| Metric | What it measures | Best use case |
|---|---|---|
| AI share of voice | Brand mention frequency relative to competitors across AI responses | Competitive benchmarking: "Are we present in the AI conversation relative to competitors?" |
| Share of citation | Citation-source presence or depth under a declared query, engine, matching, and time-window contract | Source strategy: "Which source domains or assets are answer engines citing?" |
AI SOV tracks the output mention layer. It answers: when AI engines discuss our category, how much of the brand conversation includes our brand?
Share of Citation tracks the cited-source layer. It answers: when AI engines cite sources, how often a declared source domain or asset appears under the measurement contract?
The two metrics can move together, but they diagnose different problems. A brand can be mentioned without being cited as a source. A source domain can be cited without the brand being recommended. Treat mentions, citations, recommendations, and business outcomes as separate evidence units unless a disclosed composite intentionally combines them.
Within the Machine Relations framework, coined by Jaxon Parrott in 2024, both metrics sit at Layer 5 of the MR Stack: measurement and optimization. Share of Citation is a scoped citation-source metric; AI Share of Voice is the competitive mention-breadth metric.
How to measure AI share of voice accurately #
A defensible AI SOV measurement process requires four components:
1. Define a representative prompt set. Build 30-50 prompts covering informational queries ("what is [category]"), comparison queries ("best [category] tools for [use case]"), and evaluation queries ("[brand] vs [competitor]"). The prompt set determines the topic space within which SOV is measured. Biased prompt selection produces misleading SOV numbers.
2. Run prompts across multiple AI platforms. Yext's analysis of 17.2 million AI citations checked Claude, Gemini, Perplexity, and SearchGPT and found that citation behavior diverges by model and prompt class (Yext, 2026). Platform-specific SOV reveals where a brand leads and where gaps exist.
3. Use an open denominator. Derive the competitive set from the AI responses themselves. Every brand mentioned across the response set contributes to the denominator. This captures emerging competitors automatically and prevents the structural bias of closed-set measurement.
4. Normalize entity mentions. A brand and its products must be counted as a single entity. If an AI response mentions "HubSpot" and "HubSpot CRM" separately, those are two mentions of the same brand, not two brands. Without normalization, the denominator becomes fragmented and SOV calculations drift.
Why position within AI responses is unreliable #
Mention frequency across many AI responses carries meaningful signal. Position within a single response does not. AI models generate responses probabilistically. The same prompt asked twice may produce the same brands in a different order. Research from Moz analyzing 40,000 queries found that 88% of Google AI Mode citations do not appear in the traditional organic top 10 (Moz, 2026), confirming that AI engines construct their source hierarchies independently from traditional search rankings.
Forrester has identified a "visibility vacuum" where research shifts into answer engines and marketers lose visibility into buyer questions, activity, and intent (Forrester, 2026). AI SOV is one metric that measures observed brand mentions inside that vacuum.
The practical implication: track mention frequency over time rather than optimizing for position within individual responses. A brand that appears consistently across 70% of relevant prompts has a different AI SOV profile than one that occasionally appears first but only in 30% of prompts.
What drives AI share of voice #
AI share of voice is not driven by the same inputs as traditional share of voice. The signals associated with AI SOV, based on available research, should be reported with source and sample boundaries:
-
Unpaid and journalistic sources. The Fullintel-UConn study presented at IPRRC found that 47% of citations came from journalistic sources and that more than 95% of citations were unpaid sources/media, while also including corporate, university, health-network, and association sources outside journalism (Fullintel/UConn, 2026).
-
Entity clarity and structured data. AI engines must resolve brand names to specific entities. Brands with consistent naming, clear schema markup, and corroborating entity references across independent sources are easier to measure and less likely to fragment across variants.
-
Content structure optimized for extraction. The GEO study (Aggarwal et al.) found that methods such as adding statistics improved measured source visibility and position-adjusted impressions in generated answers (Aggarwal et al., 2024). That is a source-visibility result, not proof of a universal citation-rate lift.
-
Cross-platform corroboration. Yext's 17.2 million citation analysis found model-specific citation divergence in its checked set of Claude, Gemini, Perplexity, and SearchGPT (Yext, 2026). Brands visible across multiple source types should still report per-engine results rather than collapsing engines into an unexplained average.
AI share of voice by the numbers #
- 80% of search users regularly rely on AI summaries for research (Bain, 2025)
- 84% appears in Muck Rack / Generative Pulse public copy under a broad earned-media taxonomy, with 27% journalism in the May 2026 edition; treat that as source composition, not placement causality (Muck Rack, 2025)
- 88% of Google AI Mode citations are not in the organic SERP top 10 (Moz, 2026)
- 3x stronger correlation between brand mentions and AI Overview visibility than backlinks, correlation not causation (Ahrefs, 2025)
- 47% of citations sourced from journalistic content; more than 95% from unpaid sources/media in the Fullintel-UConn sample (Fullintel/UConn, 2026)
- 17.2M citations analyzed across Claude, Gemini, Perplexity, and SearchGPT in the Yext sample (Yext, 2026)
- May 27, 2026 Google Merchant Center announced AI performance insights with share-of-voice reporting for AI Mode, AI Overviews, and Gemini shopping journeys (Google Merchant Center Help, 2026)
Frequently asked questions #
What is a good AI share of voice percentage? #
There is no universal benchmark. AI SOV depends on competitive density and category size. In a category with five competitors, equal distribution would give each brand 20%. With ten competitors, each brand's mathematical parity point drops to 10%. The more useful diagnostic is excess AI SOV: the gap between your AI share of voice and your market share. Research from Les Binet and Peter Field through the IPA found that brands whose traditional SOV exceeds their market share tend to grow share over time, with each 10 percentage points of excess SOV correlating with roughly 0.5% annual market share growth. Whether that dynamic transfers to AI-driven channels remains an untested hypothesis unless a direct AI-channel study is supplied.
How often should AI share of voice be measured? #
Monthly at minimum. Weekly provides sharper trend detection in fast-moving categories. AI models update their training data on varying schedules, so SOV can shift without any change in a brand's own behavior. Quarterly measurement misses these model-driven shifts entirely. For the full prompt-set design methodology and a board-ready dashboard structure that includes AI SOV alongside citation share and financial proxies, see the CMO AI Visibility ROI Dashboard.
Can you improve AI share of voice through paid media? #
Not directly in the same way as paid search impression share. Public source-composition studies report large unpaid-source shares in their samples, but they do not prove that any specific placement causes brand recommendations. AI engines prioritize sources and entities differently by model, query, and vertical. The practical lever is earned authority: building a pattern of third-party coverage and machine-readable entity clarity that AI engines can observe. This is the foundation of Machine Relations as a discipline.
How is AI share of voice different from AI visibility? #
AI visibility is the broader concept: whether a brand appears in AI-generated responses at all. AI share of voice is a competitive refinement of visibility: what percentage of brand mentions in those responses belongs to your brand versus competitors. Visibility answers "are we present?" SOV answers "are we winning?"