Updated 2026-07-06
AI visibility metrics that actually matter
Five metrics describe your AI visibility: citation rate, brand-mention rate, share of voice, sentiment, and answer position. Each answers a different question and fails differently when misread. Here's what each one means, what moves it, and which to act on first.
Citation rate — "can engines use my content?"
Citation rate is the share of monitored answers that link your domain as a source. It is the strictest metric and the most directly actionable: it responds to retrievability and content structure within days. Low citation rate with healthy mentions means engines talk about you from other people's pages — a content-shape problem, not an awareness problem.
Brand-mention rate — "do engines know me?"
The share of answers naming your brand, linked or not. Mentions draw on training data and the internet's accumulated description of you, so they move slower than citations and respond to reputation work: third-party coverage, consistent entity facts, category association. Mentions without citations is common and fine early; citations without mentions usually means you're cited as a generic source while competitors get named as the recommendation — a positioning gap.
Share of voice — "am I winning or just growing?"
Share of voice is your slice of all tracked-brand appearances across the prompt set. It's the market-share view that keeps solo metrics honest: your mention rate can climb while your SOV falls because competitors climb faster. Read it per platform — engines disagree about categories more than most teams expect — and per prompt cluster to find where the answer is genuinely contested.
Sentiment — "how am I framed when I appear?"
Presence isn't endorsement: "X is popular but users report billing issues" is a mention you'd rather not win. Sentiment classifies how answers frame you. It matters most as a trend and an alarm — a sentiment dip on one platform, with the answers as evidence, is an early warning that some source engines trust has turned.
Position — "am I the answer or the footnote?"
Where you appear in the answer: first recommendation, one of several options, or a trailing caveat. Position separates "technically present" from "actually recommended", and it's the tiebreaker metric when rates look healthy but pipeline doesn't feel it.
- Low citations + good mentions → content-shape problem: build liftable pages.
- Good citations + low mentions → positioning problem: strengthen brand-category association.
- Rising mentions + falling SOV → competitors rising faster: check battleground prompts.
- Healthy rates + weak position → recommended-second problem: sharpen differentiation claims.
The rules that keep metrics honest
Every metric needs a denominator of scheduled runs — single checks are anecdotes (prompt monitoring). Compare within a platform over time, not across platforms in absolutes: Perplexity cites on every answer, ChatGPT only when it searches, so their baseline rates differ by design. And insist on evidence links — a number you can't trace to the answer that produced it is a number you can't defend in a review.
Frequently asked questions
Which metric should a team start with?
Citation rate and share of voice. Citation rate is the one you can move fastest with content; SOV tells you whether moving it mattered competitively.
What cadence should I measure at?
Weekly is enough to see trends for most categories; daily makes sense during launches or when a platform is actively shifting. Match cadence to how fast you can act.
Are these metrics comparable across AI platforms?
Directionally, not absolutely. Platform mechanics set different baselines — trend each platform against itself, then compare the trends.