AI visibility is measured by putting a fixed set of questions (prompts) to AI platforms on a regular schedule and recording four things: whether the brand is named, whether its website is cited, how it compares with competitors, and what the AI says about it. The four matching metrics are mention, citation, share of voice and sentiment. Because AI answers vary, the value lies in the trend across many runs, not in any single response.
Key takeaways
- Start with a prompt set that reflects how customers really ask, not with a tool.
- Four core metrics: mention, citation, share of voice and sentiment.
- Keep the prompts, platforms and method the same between rounds so results are comparable.
- The result is a sample-based estimate that guides action, not an absolute number.
Why aren’t SEO metrics enough?
Rankings and clicks tell you where your page sits on a search results page. They do not tell you whether your brand comes up when someone asks an AI assistant about your industry, or how it is described. A person may also read the answer and never click through to any website. So you need a separate set of metrics, measured on the answers themselves.
What is a prompt set, and how do you build one?
A prompt set is a list of questions that represents how customers ask AI throughout the buying journey. One common way to group them:
- Learning: “what is X?”, “when do I need X?”.
- Comparing: “which type should I choose for…?”, “is A or B a better fit for…?”.
- Deciding: questions about cost, policies and trustworthiness.
- Brand: direct questions about you, such as “is brand X reputable?”.
A few principles for building it:
- Use customers’ natural language, drawn from sales, customer support and search data.
- Agree on the prompt set at the start and keep it fixed. When new questions are needed, track them as a separate group so the earlier data series stays intact.
- Record the platform, language and date for every answer.
What are the four main metrics?
- Mention: how often the brand is named, out of all the answers in the prompt set.
- Citation: how often your website is cited as a source. Note which pages get cited, and which third-party sources are being cited instead of you.
- Share of voice: your share of all the times brands in your competitive set are named, across the same prompt set.
- Sentiment: how accurately and how favorably the AI talks about the brand. Wrong or outdated information should be logged separately, because it needs fixing early.
You can also record position (named first, or last in a list) and whether the AI merely lists the brand or actually recommends it.
How do you make the results trustworthy?
- Repeat the question. The same prompt can produce different answers from one run to the next. Run each prompt several times and record how often you appear, instead of drawing a conclusion from a single run.
- Hold conditions steady. Same platform, same mode (with or without web search), same language, and as little influence from chat history or personalization as possible.
- Measure on a regular cycle. Monthly, for example, so that trends become visible.
- Compare with competitors using the same method. A number on its own means little; the comparison shows where you stand.
- Keep the raw answers. They let you re-check your findings and serve as evidence when needed.
How do you turn results into action?
- Not named in comparison questions: usually a sign of thin presence on third-party sources.
- Named but not cited: review the structure and specificity of the content on your website. See how AI chooses its sources.
- Described incorrectly: check that brand information is consistent across your website and reference sources.
- Competitors cited from the same few sources again and again: that is a list of places where you should consider being present.
What are the limits?
Official data from AI platforms on where a brand appears is still limited and uneven. Measurement therefore relies on sampling: you only see answers to the questions you asked, under the conditions you set. Traffic arriving from AI platforms in your web analytics can be a useful extra signal, but it does not capture the times your brand was mentioned without a click.
Nobody fully controls what an AI answers, so measuring is how you find out whether your efforts are changing anything. If you want a rigorous first measurement, with a prompt map, a share-of-voice comparison against competitors and a prioritized action list, take a look at our AI Visibility Audit.