AI share of voice: how to measure your brand's citations across the AI engines

AI share of voice is the metric that answers a simple question: when someone asks the AI engines about your category, how often do they name you instead of a competitor? It is the citation-era cousin of the old share-of-voice idea, and the formula is deliberately plain, your brand's mentions divided by the total mentions across every tracked brand for the same prompts, times 100. The hard part was never the arithmetic. It is the sampling. As Semrush notes, a serious score weighs not just how often you appear but where in the answer you land, because the first brand named is the one a reader remembers.

This guide gives you the worked formula, a defensible prompt-panel design, the per-engine reality that blended dashboards hide, and the benchmarks that tell you whether a number is good. The information-gain element that listicle-style top results skip is a copy-ready measurement design: exact panel size, cadence, and position weights, so your score is reproducible next month.

What is AI share of voice, exactly?

AI share of voice is your brand's proportion of all brand mentions that the AI engines produce for a defined category, across a fixed prompt panel and a fixed competitor set. It is a relative metric: every tracked brand's share adds up to 100 percent, so your number only means something next to the competitors you chose.

Keep three related terms distinct. A mention is your brand name appearing in an answer. A citation is your domain being linked as a source. Share of voice is the aggregate: your slice of the total mention pool. A brand can be cited as a source without being recommended by name, and recommended by name without being linked, which is why treating these as one number quietly distorts the picture. If you are still mapping the vocabulary, our generative engine optimization guide lays out the full stack, and answer engine optimization covers the on-page side of getting named.

What is the AI share of voice formula?

The base formula is one line:

AI share of voice = (your brand mentions / total brand mentions for all tracked brands) x 100

Run it across your whole prompt panel and every engine you track, not on a single query. A worked example: you track six competitors, run 120 prompts, and each prompt is asked to five engines, for 600 answer instances. If your brand is named 138 times and all seven brands together are named 1,000 times, your raw share of voice is 13.8 percent.

Raw counting treats a throwaway mention at the bottom of a paragraph the same as an explicit top recommendation, which is misleading. That is why measurement should be position-weighted. Assign a higher weight to the first or explicitly recommended brand and a lower weight to trailing mentions:

Semrush's Enterprise reporting weighs both frequency and position for this reason, treating a first recommendation as worth more than a buried one. Apply the weights to each mention, sum them per brand, then run the same division. Report both the raw and the weighted number so you can see when you are present but never the top pick.

Why measure it per engine instead of one blended score?

Because the engines disagree far more than most dashboards admit. A single blended average hides the gaps that actually cost you customers. An analysis of AI visibility audits by ReSO found the engines agreed on the same number-one brand only 5.4 percent of the time across ChatGPT, Perplexity, and Google AI Overviews, and that 77.3 percent of brands surfaced on just one of the three engines, according to ReSO's published findings. A brand that looks strong on average can be invisible on the specific engine its buyers use.

The practical rule: compute a separate share of voice for each engine, then look at the spread, not just the mean. If you win Perplexity but never appear in Google AI Overviews, that is a content and authority problem you would never see in a blended figure. Our breakdowns of Perplexity citations and Google AI Overviews explain why each engine sources answers differently.

How do you build a prompt panel that gives a reliable score?

A prompt panel is the fixed list of buyer questions you ask every engine, every measurement cycle. It is the single biggest driver of whether your score is real or noise. A one-off manual check tells you almost nothing, because the same prompt can return different brands on repeat runs.

A practical, defensible design:

• Size: aim for roughly 100 to 200 real buyer-intent prompts. Below that, one volatile query swings your whole number.
• Intent mix: cover category discovery ("best tools for X"), comparison ("X vs Y"), and problem-first phrasing ("how do I solve Z"), because each surfaces a different brand set.
• Stability: freeze the panel. Adding or dropping prompts between cycles breaks your trend line. Change it deliberately, on a schedule, and note the version.
• Repetition: ask each prompt more than once per engine and average, since a single run is unreliable.

Ground the panel in how buyers actually phrase things, not internal jargon. Pairing it with keyword and question research keeps it honest, and our work on People Also Ask questions is a useful source of real phrasings.

How often should you measure AI share of voice?

Run your full panel monthly at a minimum, and a smaller 10 to 15 prompt subset weekly, so you catch movement without waiting a full cycle. Fast-moving categories justify a weekly full run.

The reason is drift. The set of domains and brands the engines cite shifts substantially month over month in active categories, so a quarterly snapshot misses most of what happened in between. Longitudinal data is also the only way to separate signal from noise, since a single measurement cannot tell you whether a jump is real or a lucky sample. Treat AI share of voice like a rank-tracking metric, not an annual audit, and read it alongside your other AI signals from tracking AI traffic in GA4 and general rank tracking.

What counts as a good AI share of voice score?

There is no universal target, because the metric is relative to the competitors you chose. A strong score beats your named rivals and rises over time. In a focused category with two or three serious players, leadership might mean 30 to 50 percent of the mention pool. In a fragmented market with ten alternatives, 15 percent can already be category leadership.

Two guardrails keep the number honest. First, judge it against competitors and against your own trend, never as an absolute. Second, remember how much room there is to move: a study by DareAISearch reported by CIOL found that 52 percent of brands ranking on Google's first page failed to appear in AI-generated recommendations at all. A visible brand on classic search can still hold near-zero AI share of voice, which is exactly the gap this metric exists to expose.

Mentions, citations, and share of voice: how do they differ?

These three metrics answer different questions, and tracking only one leaves blind spots. Use them together.

Citations matter because a large majority of the sources the engines cite are third-party pages, not the brand's own site, so earned mentions on directories, reviews, and independent media do a lot of the work. That is the strategic link between measurement and action: you cannot lift share of voice only by editing your homepage. Our guide to brand mentions and linkless SEO covers how those unlinked references feed AI visibility, and how to get cited by AI covers the on-page citability side.

Why does AI share of voice matter for revenue, not just vanity?

Because AI referral traffic behaves like high-intent traffic, not idle browsing. Similarweb's tracking found that ChatGPT referral traffic converts at roughly 7.1 percent, second only to paid search and ahead of organic, social, and email. A visitor who arrives after an AI engine named your brand has already been pre-qualified by the answer.

That reframes share of voice from a marketing scoreboard into a leading indicator of pipeline. If the engines increasingly name a competitor for your highest-intent buyer questions, you lose the referral before the click ever happens, and no amount of classic ranking recovers it. Connecting the metric to outcomes is the same discipline as any other channel, which our SEO ROI measurement guide applies to search generally. Disclosure: this article is written by the team behind Sorank, which is one of the tools we use to track AI visibility.

What actually moves AI share of voice up?

Once you can measure it, the levers are consistent across engines. In rough priority order:

• Third-party authority. Since most AI citations point to independent sources, earned mentions on directories, reviews, and reputable media move the number more than on-site edits.
• Extractable, answer-first content. Self-contained passages that state a claim and support it are easier for the engines to lift and attribute.
• Entity consistency. A clear, consistent brand entity across the web helps the engines associate you with your category. See our entity SEO guide.
• Original data. Proprietary numbers and studies get cited because they cannot be found elsewhere, as covered in original research content.

Measure first, then attribute movement to specific actions. Without a stable panel and cadence, you cannot tell whether a change worked or the sample simply drifted.

Frequently asked questions

Related guides