How to measure AI visibility: seven metrics, and what each one tells you

Start from questions, not keywords

AI assistants answer questions, so the unit of measurement is a question asked of an assistant, from a country, in a language, on a date. Pick the questions customers ask, in their words, and ask the same set every time. A number from a different set of questions is not comparable.

For example, a dentist in Leeds might ask: "Who is the best dentist in Leeds for emergency care?" and "How much does a root canal cost in Leeds?" These are the questions real patients type or say. Write them down and keep them fixed.

Answers vary between runs, so one answer is a sample, not a verdict. Read the pattern over several checks. If you ask once and see your name, that does not mean you are always recommended. If you ask once and do not see it, that does not mean you are never recommended.

This is why measurement needs a schedule. Ask the same questions weekly or monthly, and look at the trend, not a single result. The questions themselves become your measurement instrument.

Mention rate and citation rate

Mention rate is the share of answers that name your business at all. Citation rate is the share that link to or cite your website as a source. A business can be named without being cited, and cited without being named.

For example, an assistant might say "Several options include Acme Plumbing and Beta Plumbers" without linking to either website. That is a mention but not a citation. Or an assistant might cite your blog post about boiler repair but not name your company. That is a citation but not a mention.

Mention rate is the number that matters for a service business chosen by name. If you run a local law firm, being named is what brings the call. Citation rate matters for a publisher or an online shop whose page is the destination. If someone clicks through to your product page, that is a citation doing its job.

Track both separately. A rise in mentions without citations means assistants know you exist but do not yet trust your site as a source. A rise in citations without mentions means your content is useful but your brand is not yet the answer.

Share of voice, or share of model

Among all the businesses named in the answers to your questions, your share of the mentions. Some call this share of model. It is the AI equivalent of a share of the search results page, and it is the metric that shows a competitor gaining even when your own mention rate holds.

For example, if ten businesses are named across your set of questions and you appear in three of those mentions, your share of voice is three out of ten. If a competitor appears in five, they have a larger share. Your mention rate might stay the same while your share of voice drops because a new competitor entered the answers.

It needs the whole list of businesses named, not just yours, so collect every answer in full. Do not just note whether you appear. Note who else appears and how often. This is the metric that reveals competitive shifts before they show up in your own numbers.

Position and sentiment

Position: when you are named, whether you are first, second or an afterthought. Assistants often list three to five businesses, and the first is the one most people act on. If you are always third, you are visible but not preferred.

Sentiment: how you are described, from a plain listing to a recommendation to a warning. A single negative sentence about pricing or reliability travels into every answer that repeats it. For example, if one source says "Acme is expensive but reliable," that phrase may appear in multiple assistants' answers for months.

Track position as a simple average or as a count of first, second, and third placements. Track sentiment by reading the sentences around your name and grouping them as positive, neutral, or negative. Neither is perfectly precise, but both show direction.

A drop in position often precedes a drop in mentions. If you slip from first to fourth, you may still be named, but fewer people will act. Sentiment shifts can be harder to reverse because they come from sources you do not control.

Source coverage

The sites the assistants cite when answering your questions, and whether you appear on them. This is the leading indicator: a mention on a cited site tends to show up in answers later. Group sources by type: your site, review sites, directories, comparison articles, forums, press.

For example, if an assistant cites a local news article about the best cafes in your city, and you are not in that article, you are missing a source that drives answers. If a competitor is in that article, they have a source advantage.

Report the top cited sites for your questions and which of them mention competitors but not you. This tells you where to focus your outreach or content updates. It is more actionable than a single mention rate because it points to specific pages and sites.

Review this monthly. Sources change as assistants update their training data or search the web. A site that was cited last month may not be cited this month, and a new source may appear.

AI referrals and the trend

In Google Analytics, visits that arrive from chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and similar can be grouped as an AI channel. It is small for most businesses and does not capture people who were told your name and typed it into Google, so pair it with the answer-based metrics above.

Google Search Console does not separate AI Overviews and AI Mode; Bing Webmaster Tools has an AI Performance report for Copilot. See Bing Webmaster Tools AI search guide.

Report every metric per assistant and over time. The useful sentence for a boss is: named in this many of these questions, up or down from last month, with this competitor named most often instead. That sentence combines mention rate, share of voice, and trend in one line.

Do not rely on referrals alone. They show the smallest part of the picture. The answers themselves are where the real movement happens, and they need to be collected directly.

Collecting it without a spreadsheet

Vasukai asks your questions of ChatGPT, Gemini, Google AI Mode, Google AI Overviews, Perplexity, Claude, DeepSeek and Sarvam on a schedule, keeps every answer, and reports mention rate, citations, share of voice, position, sentiment, the cited sites and the trend per assistant and per question. Plans from $19 a month; the free check shows one round with no card. See also AI rank tracker guide.

This replaces manual checking and ad-hoc notes. You pick the questions, the assistants, and the frequency. The tool does the rest and shows you the full answers so you can read exactly what was said.

For a small team, this is the difference between guessing and knowing. You can see which fixes move which metric, and you can show progress without building a custom dashboard.