What is AI visibility?
AI visibility is how often, and how accurately, a brand appears in AI-generated answers. It is measured across six metrics: citation rate, brand mention rate, share of voice, prompt coverage, brand accuracy and source concentration.
The definition, in full
AI visibility describes your presence inside AI-generated answers, in the same way that search visibility describes your presence in search results. The difference is that search visibility has an agreed unit, the ranking position, and AI visibility does not.
That missing unit is why the term is used so loosely. Vendors report a single blended visibility score, or a citation rate that quietly includes unlinked brand mentions, and both make the number look better than the reality. Neither is a lie exactly, which is the problem.
So the working definition has to be plural. AI visibility is not one number. It is six, and they move independently.
The six metrics
Full formulas, what counts, what does not, and the way each metric is usually inflated are in our published measurement standard.
Prompt coverage
How much of the buying conversation you are measuring at all. Every other metric is meaningless without it, because a high score on ten flattering prompts is not a measurement.
Citation rate
The share of answers that link a page on your domain. Never blended with brand mentions, which roughly doubles the figure.
Brand mention rate
The share of answers that name you at all, linked or not. On Gemini and other assistants that hide sources, this matters more than citation rate.
Share of voice
Your mentions as a proportion of all mentions across a competitor set declared before measurement began.
Brand accuracy
The share of answers naming you that contain no material factual error. Being visible and wrong is worse than being invisible.
Source concentration
Which domains the citations actually come from. This is what tells you whether the work belongs on your site or off it, and almost nobody measures it.
How to establish a real baseline
- 1Write down the questions your buyers actually ask, in their words, including the follow-up turns. Thirty to fifty is a workable set.
- 2Declare your competitor set before you look at anything. Changing it later invalidates the series.
- 3Run every prompt at least five times across a two-week window, because model outputs are non-deterministic.
- 4Record which platform and which model version produced each answer.
- 5Score against fixed definitions, keeping citation rate and mention rate separate.
- 6Report per platform. Never average ChatGPT with Perplexity into one number.
Our position
A baseline that takes an afternoon is a screenshot. Ours takes ten working days, and the reason is the fifth item on that list: a single run of a non-deterministic model tells you almost nothing.
What good looks like
There is no universal benchmark, and anyone quoting one is selling something. Citation rates vary enormously by category, by how much community discussion exists, and by whether the category has strong review-platform coverage.
What is comparable is your own series over time, and your share of voice against a fixed competitor set. Those two are where the signal is.
The one number worth watching in absolute terms is brand accuracy. A material error rate above roughly a quarter usually means a stale third-party listing is propagating, and that is fixable in days rather than quarters.
Related questions
Get your AI visibility baseline, free
Related answers
How to monitor AI search visibility
Build a prompt set of 30 to 50 buyer questions and freeze it.
Read the answerDefinitionWhat is generative engine optimisation (GEO)?
Generative engine optimisation (GEO) is the practice of getting a brand cited and accurately described in AI-generated answers.
Read the answerDecisionHow to report AI search visibility to a board
Report four things: share of voice against a fixed competitor set, brand accuracy, source concentration, and citation rate per platform.
Read the answer