AI share of voice shows how much of the brand visibility in AI-generated answers belongs to your business compared with competitors. Measure it with a stable set of real customer prompts, a clearly defined competitor group, and the AI platforms, locations, and time period that matter to your audience. For each response, record whether each brand appears, then divide your qualifying appearances by the total across the tracked brands; review citation presence, answer position, and sentiment separately rather than hiding them inside one score. Repeating the same test regularly matters because changing prompts or counting repeated name drops can make an apparent gain meaningless.
AI Share of Voice Meaning and Core Formula
Numerators and denominators to define
AI share of voice, or AI SOV, measures a brand’s relative presence in answers generated by AI search and assistant platforms. It answers a simple question: when people ask the prompts that matter to your market, how often does your brand appear compared with the brands you track?
A reliable calculation starts by defining the unit being counted. For basic mention share, count an answer as one qualifying appearance when it names a brand at least once. Do not inflate the score because a model repeats the same name several times in one response.
The numerator is your brand’s qualifying appearances across the selected prompt set. The denominator is the total qualifying appearances earned by every brand in the same defined competitor set. Keep the prompt list, brand-matching rules, AI engine, market, language, and reporting period consistent. Otherwise, the comparison is not meaningful.
This matters because AI answers can vary by platform and context. Search-enabled answers may cite or link to sources, while other answers may simply recommend or mention brands. Google also notes that its AI search features can surface different responses and links from traditional results, making AI visibility a distinct measurement layer alongside conventional SEO. AI features in Google Search
Mention share calculation formula
Use this core formula:
AI Share of Voice (%) = (Your brand’s qualifying mentions ÷ Total qualifying mentions for all tracked brands) × 100
For example, if your brand appears in 24 qualifying answers and the full competitor set receives 120 qualifying appearances, your AI share of voice is 20%.
[ (24 \div 120) \times 100 = 20% ]
This formula measures relative visibility, not authority, traffic, sentiment, or conversion impact. A brand can have a high mention share but rarely appear first, receive few citations, or be recommended only for a narrow use case. Those signals need their own metrics, covered later in this guide.
AI Share of Voice Metrics Beyond Brand Mentions
Answer appearance rate and citation share
Answer appearance rate measures how often your brand appears in the AI responses tested. Unlike mention share, it is not relative to competitors.
Answer appearance rate (%) = (Responses that mention your brand ÷ Total responses tested) × 100
For example, appearing in 35 of 100 sampled answers produces a 35% appearance rate. This is useful for seeing whether a brand is broadly visible across relevant prompts, including answers where no competitor is mentioned.
Citation share tracks how often an AI response cites, links to, or otherwise identifies your website as a supporting source. It is especially relevant in search-grounded experiences such as Google AI Overviews, AI Mode, and ChatGPT Search. Google’s AI features may show supporting links, while ChatGPT Search can include inline citations when web search is used. Google’s guidance for AI features in Search makes clear that inclusion is not guaranteed, even for pages that meet technical requirements.
Measure citation share separately from brand mentions. An AI answer might name your company without citing your site, or cite your content without naming the brand prominently.
Recommendation share and position-weighted visibility
Recommendation share measures how often an AI system actively recommends your brand for a defined need. Count only clear endorsements, such as “best for,” “a good choice for,” or inclusion in a shortlist the answer presents as recommended options.
This metric is usually more commercially meaningful than a neutral mention. However, it needs a strict annotation rule. A factual reference, comparison point, directory listing, or warning should not count as a recommendation.
Position-weighted visibility adds importance to where a brand appears in a ranked list or ordered set of options. A simple model might assign three points for first position, two for second, and one for third. Divide a brand’s points by all available points across the sample to calculate its weighted share. Record the visible order exactly as presented, rather than assuming the first company is always the strongest recommendation.
Why AI visibility metrics should not be conflated
Each metric describes a different type of AI visibility. Mention share shows relative presence. Appearance rate shows reach across the prompt set. Citation share indicates source visibility. Recommendation share captures commercial preference, while position-weighted visibility reflects prominence within ordered answers.
Combining them into one “AI rank” can hide useful changes. A brand may gain mentions but lose citations. It may be recommended more often for one use case while becoming less visible overall. Keep the metrics separate on the reporting dashboard, then interpret them together with prompt intent, engine, location, and time period.
Prompt Sampling for Reliable AI Visibility Measurement
Customer journey, use case, and audience coverage
A useful AI share of voice report begins with prompts that reflect how prospective customers actually research, compare, and choose solutions. Sampling only broad “best” queries may create an impressive-looking number, but it rarely explains where your brand is visible or missing.
Build the prompt set around the customer journey. Early-stage prompts explore a problem or category, such as “how to improve website performance.” Mid-stage prompts compare approaches, features, or providers. Late-stage prompts focus on suitability, pricing, implementation, support, or alternatives.
Cover the main use cases your product serves. A technical buyer, a small business owner, and an agency may ask different questions about the same category. Segment prompts by audience where those differences affect the brands or sources an AI system is likely to surface.
Use enough prompts within each group to avoid treating a single response as a trend. Keep a prompt library with an intent label, target audience, journey stage, and business priority. Review it routinely, but avoid changing too many prompts at once. Consistency makes movement in AI visibility easier to interpret.
Branded, non-branded, geographic, and language prompts
Include a balanced mix of branded and non-branded prompts. Branded prompts test whether AI systems describe your company accurately, recognize key products, and position you against alternatives. Non-branded prompts reveal whether the brand earns visibility when the user has not named it.
Geographic prompts are essential for local, regional, regulated, or market-specific offers. Results can change with location, and search systems may use signals such as a user’s location and language to determine relevance. ChatGPT Search may also use approximate IP-based location to improve local results. ChatGPT Search
For multilingual businesses, translate prompts for the way people naturally search in each market rather than relying on literal translations. Record the language, country or city, and any location settings used for every run. This makes AI share of voice reporting more reproducible and prevents one market’s results from being mistaken for another’s.
AI Response Collection and Share Calculation Method
Repeat runs, timestamps, and model-version tracking
AI answers are not fixed rankings. The same prompt can produce different wording, brands, citations, and ordering across runs. Search-enabled tools may also retrieve fresh sources, while model updates and product settings can change how an answer is generated.
Run every prompt more than once within a defined collection window. Three to five runs per prompt is a practical starting point for identifying obvious variation without making the process unnecessarily expensive or slow. Use the same signed-in state, location, language, browsing or search setting, and prompt wording wherever possible.
For each response, save the full output and record:
- Prompt ID and exact prompt text
- AI engine and product surface
- Model name or selected mode, when visible
- Date, time, timezone, market, and language
- Whether web search was enabled
- Brand mentions, citations, recommendation status, and listed position
Track model and product changes as a separate field, not a footnote. Platforms regularly update available models and response behavior, as shown in ChatGPT’s release notes. When a meaningful change occurs, mark the reporting period so readers do not confuse a platform shift with a change in brand performance.
A compact AI share of voice calculation
First, apply the same qualifying rule to each collected answer. For example, count one appearance when a brand is named in the main response or a clearly presented recommendation list. Count it once per answer, even if its name appears repeatedly.
Then calculate:
AI Share of Voice (%) = (Your qualifying appearances ÷ All qualifying appearances across the competitor set) × 100
If 50 prompts are run four times, the sample contains 200 responses. If Screpy appears in 42 qualifying responses and all tracked brands receive 210 qualifying appearances, Screpy’s AI share of voice is 20%.
[ (42 \div 210) \times 100 = 20% ]
Report the underlying counts beside the percentage. A move from 2% to 6% can be meaningful, but it needs different interpretation when it reflects three additional appearances rather than thirty.
Competitor Sets and AI Engine Reporting Rules
Defining and maintaining the competitor set
Your AI share of voice is only as useful as the competitor set behind it. Define competitors by the customer problem and use case being measured, not simply by the companies you consider rivals. A website monitoring platform, for example, may compete with different brands for technical SEO prompts, site-speed prompts, and agency-focused prompts.
Include direct competitors, established category alternatives, and any brands that appear frequently in the sampled AI answers. Exclude unrelated companies that share a name, marketplaces that only list products, and brands that do not serve the prompt’s underlying need.
Keep the core competitor set stable for a reporting period. If a new competitor becomes consistently visible, add it at the start of a new period and clearly note the change. Recalculate historical results only when you need a like-for-like trend line. Otherwise, a changed denominator can look like a shift in AI visibility when the real change is simply the list of brands being measured.
Use a separate “other brands” field for unexpected mentions. This protects the analysis from blind spots without constantly changing the official comparison set.
Reporting results by engine before aggregation
Report each AI engine separately before creating a combined AI share of voice figure. Google AI Overviews and AI Mode, ChatGPT Search, and other AI answer experiences can retrieve, cite, structure, and recommend information differently. Google states that AI Overviews and AI Mode may use different models and techniques, so their answers and supporting links can vary even within Google Search. Google AI features in Search
For every engine, show the prompt count, repeat runs, appearance rate, mention share, citation share, recommendation share, and date range. This reveals whether a brand’s overall result is broadly supported or driven by one platform.
Only aggregate engine-level results when the sampled prompts, markets, and collection rules are comparable. Use a simple average when each engine has equal strategic importance. Use a weighted average when traffic potential, customer usage, or business relevance differs by engine. State the weighting method beside the result.
A combined score is helpful for executive reporting, but engine-level data is where the actionable SEO insight usually sits.
AI Share of Voice Benchmarks, Trends, and Limits
Interpreting changes over time
There is no universal “good” AI share of voice benchmark. A strong result depends on the category, prompt intent, competitor set, AI engine, and size of the sample. A 15% mention share may be meaningful in a crowded market with many established brands, while the same figure may signal limited visibility in a narrow category.
Start by creating a baseline from several consistent reporting periods. Then compare changes using both percentages and raw qualifying appearances. For example, a rise from 12% to 18% is more credible when appearances increase from 30 to 45 across a stable prompt sample than when the change represents only a few answers.
Look for patterns by prompt group. Growing visibility in high-intent comparison and recommendation prompts is usually more valuable than a gain concentrated in broad informational queries. Compare AI visibility with organic impressions, clicks, engaged sessions, and conversions where possible. Google’s Generative AI performance report can help website owners evaluate visibility in Google’s generative AI features alongside wider Search performance.
Treat sudden shifts carefully. Check for changes in prompts, competitors, location, language, collection settings, model versions, and product interfaces before attributing movement to SEO work.
Why AI share of voice is a directional metric
AI share of voice is a directional metric because AI responses are dynamic. Results can vary by run, and AI platforms may use different retrieval methods, models, source sets, and answer formats. Google notes that AI Overviews and AI Mode can return different responses and supporting links, even for the same underlying topic. Google AI features in Search
It also does not measure demand, traffic, revenue, or factual accuracy on its own. A brand can be frequently mentioned without earning clicks or being the best fit for the user. Conversely, a lower share may still produce valuable visits if citations appear on a small set of high-conversion prompts.
Use AI share of voice to identify visibility opportunities, competitive gaps, and changes worth investigating. Pair it with traditional SEO data, on-site engagement, conversion metrics, and manual quality checks. That combination gives a more realistic view of how well a brand is being discovered and represented in AI-driven search.