Glossary

AI share of voice

The share of relevant AI answers that cite your brand, relative to the competitors named alongside you.

Updated May 2026Definition
MentionRadar TeamAI search visibility researchers tracking how LLMs cite the web.
The short answer

AI share of voice (AI SOV) is the proportion of relevant AI answers that cite or mention your brand, measured against the competitors named in the same set of answers. Conceptually it borrows from the classic marketing metric — your slice of the conversation — but the “conversation” is now the set of citations that answer engines (ChatGPT, Gemini, Grok) produce for a defined group of buyer queries. If you run 100 relevant queries across the models and your domain is named in 22 of those answers, your AI SOV for that set is roughly 22%. It is the headline competitive metric for AI search because, in a world of one synthesised answer per query, what matters is not your rank but whether you made the answer at all — and who you shared it with. Always anchor the number to a defined query set and a date.

How is AI share of voice calculated?

The mechanic is simple once the inputs are defined. You need a query set (the buyer questions relevant to your category), a set of models to run them through, and a way to detect whether each answer cites you. Then:

  • Numerator — the count of answers in which your domain is cited or mentioned.
  • Denominator — the total answers across the query set (or, for a relative SOV, the total mentions across you and your named competitors).
  • Segment it. SOV is most useful split by model (ChatGPT vs Gemini vs Grok often disagree) and by intent, so a high score on informational queries doesn’t mask a low score on bottom-of-funnel buying questions.

Why is AI share of voice different from a visibility score?

A visibility score is a black box: it samples some prompts and returns a number you cannot trace. A well-built SOV is auditable — every point traces back to a specific query and a real answer you can read. That is the difference between a metric you can defend in a client meeting and one you cannot. The deeper you go, the more SOV stops being a single number and becomes a list — which is exactly what reverse AI search gives you.

What are good AI share-of-voice benchmarks?

There is no universal “good” number — SOV is only meaningful relative to your category and your competitors, because how often AI cites anyone in a space varies widely. We publish category-level reference ranges from our index in category AI share-of-voice benchmarks (labelled as estimates). Treat any benchmark as directional and date-stamped, because AI answers are volatile.

How do you track AI share of voice over time?

Because AI answers change frequently, a one-off SOV reading ages fast. The useful approach is to fix a query set and re-measure on a cadence so you can watch the trend. That requires the underlying record — the AI query index that logs which domains get cited for which queries over time. To see where you stand right now, run the free Domain Check: it returns the real queries a domain is cited on across all three models, and the competitors named beside it — the raw material every SOV figure is built from.

Worked example

Define a 20-query set for “project-management tools.” You are cited in 6 of the resulting AI answers, so your AI share of voice is 30% for that set. A competitor is cited in 12 of the same 20 — a 60% share — so even though 6 citations sounds healthy in isolation, the comparison shows you trail badly. Re-running the same fixed 20 queries a month later and seeing your share move to 45% is real, defensible progress; a different query list would not be comparable.

Frequently asked questions

How is AI share of voice different from a citation count?

A raw citation count measures activity; AI share of voice measures competitive position. Being cited on fifty queries means little until you know whether rivals are cited on five or five hundred of the same set.

How do you calculate AI share of voice?

Define a query set for your category, count the answers that cite you, and divide by the total number of answers in the set — optionally weighted by prompt volume.