AI Search Visibility Metrics: 6 KPIs That Actually Matter

The short version:

• There is no Search Console for AI engines: without defined KPIs measured on a regular schedule, you are flying blind.
• Six metrics cover the essentials: mention rate, mention position, weighted visibility score, share of voice, sentiment and citations of your site.
• Each KPI answers a different question; your mention rate can climb while your competitive position slips.
• Everything reads engine by engine: an "all AI" average hides gaps we have measured at 30% versus 100% visibility for the same brand.
• To get all six numbers at once, the free audit asks ChatGPT, Gemini and Perplexity 10 questions from your market, in one minute.

Why do you need dedicated AI visibility KPIs?

Because classic SEO metrics do not measure what happens inside a written answer. A Google position tells you whether your page ranks; it does not tell you whether ChatGPT recommends your brand, in which spot, against whom, or in what tone. And since answers vary from run to run, a one-off impression ("I tested it, we show up") is not a measurement: you need defined indicators, computed the same way at every reading.

The good news: six KPIs cover the essentials, and you can measure all of them by hand or with a dedicated tool.

KPI 1: mention rate

Mention rate is the share of answers where your brand is named, on a fixed basket of buyer questions asked without brand names. Named in 8 answers out of 10: an 80% rate. It is the existence KPI: it answers "are we in the conversation?". Compute it on the same questions at every measurement, with several runs per question, or you are comparing different baskets.

KPI 2: mention position

Being cited at the top of an answer and being cited eighth in a list do not produce the same effect: your buyers read the first recommendation, rarely the eighth. Track your brand’s average position when it appears. It is the salience KPI: "when we are mentioned, are we the recommendation or the footnote?"

KPI 3: the weighted visibility score

The visibility score combines the first two KPIs into a single number you follow over time: the share of answers where you appear, weighted by the position of each mention. As an example, Pythie gives full credit to a first-position mention, removes 10% of credit per rank, with a floor at 50% from the sixth position on. The exact scale matters less than its consistency: this is the trend KPI, the one you watch week after week to know whether you are progressing.

KPI 4: share of voice

Share of voice divides your mentions by the total mentions of all brands cited on the same questions: 12 mentions out of 55 counted, a 22% share. It is the competitive KPI: your score can rise while your share of voice falls, if a competitor rises faster. Our dedicated article details the formula, the counting protocol and the traps of the raw rate.

KPI 5: sentiment

Being mentioned is not enough: in what tone? An engine can cite you while flagging a high price, a missing feature, a mixed review it read somewhere. Track what each engine praises and criticizes when it talks about you: it is the reputation KPI, and it often explains why an engine mentions you without ever recommending you. Recurring criticisms also point to your content priorities: an objection the engines repeat is an objection your buyers hear.

KPI 6: citations of your site

Distinguish the mention (your brand is named) from the citation (your site is listed as a source of the answer): the difference is detailed here. The citation rate measures whether engines lean on your own pages to answer, or only on third parties. Read it with the engine in mind: according to the Profound study, Perplexity shows sources in almost every answer while ChatGPT rarely displays any. A low citation rate on ChatGPT is normal; on Perplexity, it is a signal.

The minimal dashboard

All of it computed per engine, on a fixed basket of questions. That is exactly the dashboard Pythie’s weekly scan fills automatically: 15 questions, 3 engines, several runs, every week.

The three reading traps

A clean set of KPIs can still produce wrong conclusions. Three traps come back constantly:

1. The cross-engine average. We have measured the same brand at 100% visibility on Gemini and 30% on ChatGPT, a gap documented here. An "all AI" average of 65% describes no reality: track each engine separately.
2. The single measurement. The same question asked twice can return two different answers. A KPI computed on one run measures a draw, not a trend: always aggregate several runs per question.
3. The score without the share of voice. A rising score can hide a competitor rising faster. Always read the trend KPI (score) together with the battle KPI (share of voice): that pair tells you whether you progress in absolute and in relative terms.

Frequently asked questions

What is a good AI visibility score?

There is no universal threshold: it all depends on how many brands the engines cite on your questions, and that varies widely by industry. The useful reference is double: your own trajectory (is the score rising?) and your gap to your market’s leader, engine by engine. An initial audit sets the starting line; the starting line is what gives every later number its meaning.

How often should you measure these KPIs?

Weekly for the first four (mention, position, score, share of voice): engines re-weight their sources continuously and rankings move within weeks. Sentiment and citations evolve more slowly: a monthly reading catches the inflections without the noise.

Can you track these KPIs without a tool?

Yes, by hand: 10 to 15 fixed questions, several runs, a spreadsheet, one to two hours per measurement, every week. The full protocol is in our article on AI share of voice. A tool brings no other magic than regularity, and regularity is what turns readings into trends. The free audit gives you the starting snapshot in one minute.