Three AI search metrics, defined: visibility, share of voice and position
Written by
Odin Lund
Category
AI search
Date
Share this article
Visibility: are you in the answer at all?
Visibility is the share of answers to a prompt set that name the business at all. If a prompt set is run 100 times across three models and the business appears in 23 answers, visibility is 23%. It is the closest equivalent to being on the first page of a search result, with one difference: an answer that does not name you gives the customer no way to find you. There is no page two.
Visibility varies sharply by model, because each model reads a different mix of sources. For the client in our engagement log, September visibility was 26.7% on ChatGPT, 26.0% on Gemini and 15.4% on Google AI Overviews. The overall figure, 23.3%, hides that spread, which is why we always report it by model.
Share of voice: how much of the conversation is yours?
Share of voice is the business’s share of all company mentions across the same answers. If an answer names three businesses and yours is one of them, you hold a third of that answer’s voice; summed across the set, it shows how much of the conversation you own relative to competitors. It can be high while visibility is modest when a business dominates the answers it does appear in. The same client held 47.2% share of voice against 23.3% visibility: named in fewer than a quarter of answers, but usually named first and often alone when it was.
Share of voice is also the metric that moves when competitors lose ground. Between June and September, the client’s share went from 4.6% to 47.2% while the number of firms named per answer stayed roughly constant, which means the gain came out of competitors’ mentions.
Position and citations: how strongly, and on what evidence?
Position is the business’s rank among the businesses an answer names, averaged across the answers that name it. A position near 1 means the model usually leads with you. Against a fixed set of competitors it collapses into a single rank: first of fourteen, in this case. Citations count the answers in which the model retrieved or linked a page you own. They are the most direct evidence that your own content is doing the work: 1,164 citing answers in 30 days, from 63 pages, tells you exactly which pages the models trust and which they ignore.
A related measure is alignment: how closely what the models say about you matches what your site says. We read every claim in every answer against the business’s verified facts. For this client, alignment stands at 90.2%; the remaining 9.8% is a list of specific corrections to make at the source.
Reading them together, and sources
Read the three by model and by stage of the customer journey. Decision-stage prompts, where the customer is ready to choose, tend to move first: 37.8% visibility for the client above, against 6.3% on comparison prompts. That gap is the work list. Read them over time, not once. Answers are rebuilt as sources change and competitors publish; a position that is not maintained decays. And read them next to outcomes. Visibility that does not produce visits, and visits that do not produce customers, are not the goal, which is why we trace every citing answer to the page it came from and every AI referral to a lead.
Why the metrics matter more than click-through now: Ahrefs measured a 34.5% drop in click-through for the top organic result when an AI Overview was present (Ahrefs, April 2025), and Pew found users clicked a result in 8% of visits with an AI summary versus 15% without (Pew Research Center, July 2025). The click is no longer the unit of visibility. The mention is.
Sources
SourceNode tracking, aisearchmonitor.com, September 2026, by model and by customer stage.
Ahrefs, “AI Overviews reduce clicks by 34.5%,” April 2025.
Pew Research Center, 22 July 2025, click behaviour with and without AI summaries.
Share Blog
