AI visibility metrics explained: mention rate, share of voice, citations and sentiment

You need new metrics because AI engines don't show a ranked list of ten links. They write one answer that names a few brands and links to a few sources. Rankings, impressions and click-through rate don't describe that answer. A brand can rank first on Google and still be missing from ChatGPT's recommendation, and a brand can be recommended in an answer without anyone clicking through.

The researchers who coined "generative engine optimization" made the same point: they measured a source's visibility inside the generated answer, by how much of the answer used it and how prominently, rather than by rank (Aggarwal et al.).

Every AI visibility metric is calculated over a prompt set: a fixed list of questions your buyers ask, such as "best project management tool for agencies" or "Asana vs Monday". The prompt set is to AI visibility what the keyword list is to SEO. Change it and every metric changes, so keep it stable and version it when you add prompts.

A useful prompt set mixes three kinds of questions:

• Category questions: "What is the best X for Y?"
• Comparison questions: "X vs Y", "alternatives to X".
• Problem questions: "How do I fix Z?", where your product is one possible answer.

Build one set per market and language. Turkish and Arabic answers often cite different sources and name different brands than English ones. Prompt research helps you find the questions people actually ask.

Mention rate is the share of AI answers in your prompt set that name your brand.

> Mention rate = answers that mention your brand ÷ all answers checked

If you run 100 prompts in ChatGPT and your brand appears in 23 answers, your ChatGPT mention rate is 23%. Calculate it per engine and per market; an average across engines hides the engine where you're missing.

When it misleads: a high mention rate on easy prompts (your own brand name, for example) says little. Exclude branded prompts or report them separately.

AI share of voice is your brand's mentions as a share of all brand mentions in the same answers.

> Share of voice = your brand's mentions ÷ mentions of all tracked brands

If the answers name your brand 23 times and your four tracked competitors 77 times in total, your share of voice is 23%. Mention rate tells you whether you're present; share of voice tells you whether you're winning the shortlist. Track it in competitor analysis, where every competitor is calculated on the same prompts.

Position describes how prominently an answer recommends you. AI answers don't have numbered ranks, so position is usually recorded in a few levels:

1. First recommendation: named first or called the best choice.
2. Listed: one of several recommended options.
3. Mentioned: named in passing, as an example or a caveat.

Report the share of mentions at each level. Moving from "mentioned" to "first recommendation" can matter more than a higher mention rate.

A citation is a link from the AI answer to a source page. A mention is your brand name in the answer text. They are independent: ChatGPT search and Perplexity show linked sources (OpenAI), and Google shows links in AI Overviews (Google Search Central), but the cited page is often a review site or a comparison article, not the brand's own site.

Track two citation metrics:

• Own-site citation rate: the share of answers that link to your domain.
• Source share: which third-party domains are cited most for your prompts. These are the sites where a mention moves the answer.

Citation tracking lists both, per prompt.

Sentiment measures how the answer describes your brand: positive, neutral or negative, and which attributes it repeats ("expensive", "easy to use", "limited support"). Score it per mention, then look at the attributes. A neutral score with a repeated "expensive" claim is a pricing-page problem you can fix. See brand sentiment.

Measure daily and report weekly or monthly. AI answers vary from run to run: the same prompt can name different brands an hour later. A single check is an anecdote; a 30-day series is a trend. Keep the method fixed:

• same prompts, engines, languages and locations;
• logged-out sessions without personal memory;
• several runs per prompt when you need tighter numbers.

Report share of voice as the headline, with mention rate and position beneath it, per engine and market. Share of voice is easy to explain ("of all the brands AI recommends in our category, we are 23%") and it moves when competitors move, which is what leadership cares about. Add citations and sentiment as diagnostics that explain why the headline moved.

AI visibility comes down to five numbers on a fixed prompt set: mention rate, share of voice, position, citations and sentiment. Measure them daily, per engine and market, read the trend, and use citations and sentiment to decide what to fix.