Why Traditional Share of Voice Metrics Miss the AI Search Layer
Share of voice has always been a useful way to understand how much of a given conversation your brand owns compared to competitors. For years, that meant tracking mentions in media, ad impressions, or organic search rankings. Now there is a third channel that most brands are completely ignoring: the answers that AI engines like ChatGPT, Perplexity, and Gemini generate in response to user questions.
These answers are not a list of ten blue links. They are a single, synthesised response. One brand gets named. Several get ignored. The difference between being cited and being invisible can be worth enormous amounts of revenue, and yet almost no brand has a systematic way to measure it.
This guide covers how to actually track your brand's share of voice in AI-generated answers, what to measure, how to build a process around it, and what to do when the numbers are not going your way.
Defining AI Share of Voice
Before you can measure something, you need a clear definition. AI share of voice (AI SoV) is the percentage of relevant AI-generated responses that mention or recommend your brand, compared to the total number of responses that mention anyone in your category.
For example: if you query 20 questions that a customer might ask in your category, and your brand appears in 8 of those responses while competitors appear across all 20, your raw AI SoV is 40%. That is a meaningful, trackable number.
The key difference from traditional SoV is that AI responses are not just about reach or frequency. They carry implicit endorsement. When ChatGPT recommends a product, the user typically acts on it with a high degree of trust. So even a small AI SoV can have an outsized commercial impact compared to a banner ad impression.
Building Your Query Set
The foundation of any AI SoV measurement programme is a well-constructed list of queries. These are the questions you believe your target customers are asking AI engines right now.
How to identify the right queries
Start with your existing keyword research. Look at informational queries, comparison searches, and "best of" type questions. These are the formats that AI engines tend to answer with named recommendations. Questions like "what is the best project management tool for small teams?" or "which CRM is easiest to set up?" are exactly the kind of prompts that generate a list of brands in the response.
Layer in questions that match different stages of the buying journey. Awareness-stage questions ("how does email marketing work?") are unlikely to name brands. Consideration and decision-stage questions ("which email platform should I choose for a Shopify store?") are far more likely to generate brand-specific answers. Focus your query set on the latter.
Aim for a minimum of 30 queries for a meaningful baseline. If you are in a competitive category with many sub-niches, 50 to 100 is more realistic. Organise them by topic cluster so you can identify where you are strong and where you are invisible.
Standardise your query format
Slight variations in how a question is phrased can produce different AI responses. For consistent tracking, use the same exact wording every time you run your measurement cycle. Keep a spreadsheet with the exact query text, the date it was first run, and any notes about why it was included.
Running the Queries and Recording Results
This part of the process is still largely manual, though tools are beginning to emerge to help automate it. Here is a practical workflow you can start using immediately.
Which AI engines to track
At minimum, track ChatGPT (GPT-4 and above), Perplexity, and Google's AI Overviews (the AI-generated summaries that now appear at the top of many Google search results). Gemini is worth adding if your audience is heavily Google-centric. Each engine has different training data, different citation behaviours, and different user demographics, so your SoV can vary significantly between them.
Do not try to track all engines at once when you are starting out. Pick the two your audience uses most and build consistency there first.
Recording the output
For each query, record the following: the full AI response text, which brands are mentioned, whether your brand is mentioned, the position of your brand in the response (first, second, last, etc.), whether a source link is cited, and the date. A simple spreadsheet works well. Use one row per query per engine per date.
Position matters. Being mentioned first in an AI answer carries more weight than being the fourth option listed. Track this separately so you can see not just if you appear, but how prominently.
Frequency of measurement
Run your full query set once a month at minimum. AI engines update their models and training data regularly, and your rankings can shift without any obvious trigger. Monthly tracking gives you enough data to spot trends without consuming all of your time. If you are in a fast-moving category or have recently made significant changes to your website's structured data and content, fortnightly tracking is worth the extra effort.
Calculating Your AI Share of Voice Score
Once you have a month's worth of data, you can calculate your score. Here is a simple method that works well for most brands.
For each query, assign a point value based on whether you are mentioned: 1 point for any mention, 2 points for a first-position mention, 0 points for no mention. Do the same for each competitor you are tracking. Sum the points across all queries and divide your total by the combined total of all brands tracked. Multiply by 100 for a percentage.
This gives you a weighted AI SoV that rewards prominent mentions over buried ones. You can adjust the point weights based on what matters most to your business. Some brands care more about citation volume; others care almost entirely about being mentioned first.
Track this score over time in a simple chart. Three to six months of data will start to reveal clear patterns: which query clusters you own, which ones competitors are dominating, and whether your optimisation efforts are making a difference.
What Influences Your AI Share of Voice
Understanding the levers behind your AI SoV score is what turns measurement into action. Several factors consistently influence whether AI engines cite your brand.
Structured data and schema markup
AI engines parse structured data to understand what your brand does, what it sells, and how it is rated. Brands with well-implemented schema markup, including Brand schema, product schema, and review schema, are far easier for AI systems to categorise and recommend with confidence. If your structured data is thin or absent, you are essentially asking AI engines to guess what you are about.
At FlinnSchema, we have seen brands dramatically shift their AI SoV after implementing structured data correctly. It is not the only factor, but it is one of the most controllable ones.
Third-party mentions and citations
AI engines are trained on vast amounts of web content, including review sites, forums, news articles, and directories. The more your brand appears in authoritative third-party sources, the more likely it is to be included in AI-generated responses. This is the AI equivalent of link building. A mention in a well-regarded industry publication carries more weight than ten mentions on low-authority blogs.
Review signals
Aggregate ratings and review content are frequently used by AI engines to assess brand quality. Implementing AggregateRating schema ensures your review data is machine-readable and attributable to your brand. Brands with strong, consistently positive reviews tend to appear more frequently in AI answers that involve recommendation queries.
Content clarity and topical authority
If your website content is vague, jargon-heavy, or poorly structured, AI engines will struggle to extract clear facts about your brand. Specific, well-organised content, particularly FAQs, comparison pages, and structured product descriptions, gives AI engines the building blocks they need to include you in relevant answers.
Competitive Intelligence from AI Monitoring
Measuring your own AI SoV is valuable, but the competitive intelligence you gather in the process is equally useful. When you track which brands appear in response to your target queries, you start to understand what those brands are doing differently.
Look at the content structure of top-cited competitors. Do they have detailed FAQ sections? Are their product pages heavily structured? Do they have a strong presence on review platforms? Are they regularly cited in trade press? These observations translate directly into an action list for improving your own AI visibility.
Pay attention to which competitors appear consistently across multiple AI engines. A brand that ranks well across ChatGPT, Perplexity, and Google AI Overviews simultaneously has almost certainly invested in their structured data and content quality in ways that are worth studying.
If you want a clearer picture of where your brand stands right now, the free AI visibility audit from FlinnSchema is a good starting point. It surfaces gaps in your structured data and content that are likely suppressing your AI SoV.
Setting Benchmarks and Goals
Raw numbers are only useful if you know what good looks like. When you are first starting out, your goal should simply be to establish a baseline. Do not try to set a target SoV percentage until you have at least two months of consistent data.
Once you have a baseline, a realistic improvement target is a 10 to 15 percentage point increase in AI SoV over six months, assuming you are actively working on the factors above. Brands in less competitive niches can move faster. Those in saturated categories with well-established competitors may need to be more patient.
Break your targets down by query cluster. You are unlikely to win every category at once. Focus first on the clusters where you are closest to the threshold of being mentioned, then build from there.
Frequently Asked Questions
How often do AI engines update their responses?
This varies by engine. Perplexity retrieves live web content, so its responses can change frequently. ChatGPT's responses are influenced by its training data, which is updated periodically but not in real time. Google's AI Overviews draw on its live index. Because of this variation, monthly tracking is the minimum recommended frequency, and you should always note the date when recording results.
Can I automate AI share of voice tracking?
Some specialist tools are beginning to offer automated AI monitoring, though the space is still maturing. For most brands, a structured manual process in a spreadsheet remains the most reliable method. The key is consistency: using the same queries, the same engines, and the same recording format every time you run the process.
Does schema markup directly affect what AI engines say about my brand?
Schema markup does not dictate what an AI engine says, but it significantly improves the engine's ability to understand and correctly attribute information about your brand. Brands with clear, accurate structured data are more likely to be cited correctly, with the right product names, ratings, and descriptions, compared to brands whose information has to be inferred from unstructured text.
What should I do if my AI share of voice is very low?
Start with an audit of your structured data and on-page content. Then look at your third-party presence: are you being mentioned on review platforms, industry directories, and relevant publications? After that, review your content for specificity. Vague content is hard for AI engines to cite confidently. Improving these three areas in combination tends to produce the most meaningful improvements in AI SoV over a three to six month period.