AI Share of Voice: Formula, Worked Example, and How to Grow It

Share of voice has long been a planning metric in advertising: your slice of the conversation in your category, measured against competitors. AI answers created a new version of that conversation. When a buyer asks ChatGPT or Google for the best tool in your category, the answer names a handful of brands, and the brands it names over and over are the ones that get considered.

AI share of voice measures how much of that conversation is yours. This guide gives the formula, the counting rules that make it reliable, a fully worked example, the sample sizes you need, and the levers that move it.

What is AI share of voice?#

AI share of voice is the percentage of all brand mentions in AI-generated answers, across a defined set of prompts and competitors, that belong to your brand. It answers one question: when AI engines talk about your category, how often is it about you rather than your competitors?

Three things make it different from classic search metrics:

• It is competitive by definition. Your share can fall even if your own mentions stay flat, because a competitor gained.
• It is measured across repeated runs. AI answers change between runs of the same prompt, so share of voice is a frequency, not a position.
• It covers several engines. ChatGPT, Perplexity, Claude, Gemini, Google AI Overviews, and AI Mode each retrieve from different sources and produce different numbers.

AI share of voice is the core measurement in any answer engine optimization program, because it is the metric closest to what buyers actually see.

How do you calculate AI share of voice?#

Divide your brand's mentions by the total mentions of every tracked brand, then multiply by 100. Count across the same prompts, engines, and number of runs every cycle.

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

The formula is simple. The definitions behind each number are where most measurements go wrong. Before you count anything, fix four inputs:

1. The prompt panel. A fixed list of 30-100 questions your buyers actually ask, weighted toward evaluative ones ("best X for Y," "X vs Y," "alternatives to X," "is X worth it").
2. The engine set. Which AI surfaces you measure, for example ChatGPT, Perplexity, Claude, Gemini, Google AI Overviews, and AI Mode. If your buyers also use Grok or Meta AI, our guides on how to rank in Grok and how to rank in Meta AI cover how to test them.
3. The run count. How many times each prompt runs on each engine per cycle. Three is a practical minimum.
4. The competitor set. The brands you count in the denominator. Include the ones buyers actually compare you with, plus any brand the engines mention often that you did not expect.

Change any of these and the number changes, so version them. A share of voice figure without its panel, engines, runs, and competitor list is not interpretable.

What counts as a mention?#

A mention is your brand name appearing in the answer text as an option, recommendation, or subject. Set counting rules once, write them down, and apply them the same way every cycle.

The per-answer cap matters most. Without it, a single long answer that repeats one brand's name ten times can swing your share for the whole cycle.

The branded-prompt rule matters second. "Is [your brand] good for agencies?" will mention you every time. Keep those prompts in a separate panel that measures how you are described, not how often you are chosen.

A worked example (hypothetical numbers)#

The numbers below are an illustrative example, not real data. They show the arithmetic end to end for a fictional brand ("You") tracked against three competitors (A, B, and C).

Panel setup: 40 prompts x 5 engines x 3 runs = 600 answers.

Step 1: Count mentions (max one per brand per answer).

Step 2: Calculate presence rate. You appeared in 156 of 600 answers: 156 / 600 = 26.0%. Competitor A's presence rate is 252 / 600 = 42.0%.

Step 3: Calculate citation share. Across all 600 answers the engines produced 1,200 source links, and 42 pointed to your domain: 42 / 1,200 = 3.5% citation share.

Step 4: Weight for prominence (optional). Brands named first in an answer are more likely to be read and chosen. One simple scheme gives a first-named mention a weight of 1.0 and any later mention 0.5:

Your weighted share (21.1%) is lower than your raw share (23.6%), which tells you something specific: when you are named, you are usually not named first. Competitor A shows the opposite pattern.

Step 5: Break it down by engine. Each engine produced 120 answers.

What this example tells the team: Competitor A owns the category conversation and is usually named first. "You" does best on Perplexity, which suggests your cited pages are being used there, and worst on Claude. A 3.5% citation share against a 26% presence rate suggests the engines know the brand but rarely use its pages as evidence, which points toward the evidence and structure fixes in the GEO checklist.

Which share of voice variants should you track?#

Track one headline number and three supporting numbers. Mention-based share of voice is the headline; presence rate, citation share, and sentiment explain why it moved.

Two warnings about weighting. First, keep the scheme simple and fixed; complicated weights make it hard to explain changes. Second, do not treat position as a rank. A SparkToro and Gumshoe.ai study of 2,961 prompt runs across ChatGPT, Claude, and Google's AI found under a 1 in 1,000 chance of two answers listing brands in the same order. Order only becomes meaningful when averaged over many runs.

How many prompts and runs do you need?#

You need enough answers that a real change is larger than random variation. A rough guide is the standard margin of error for a proportion: 1.96 x the square root of p(1 - p) / n, where p is your presence rate and n is the number of answers.

Using the example's 26% presence rate:

Treat these as best-case numbers. Repeated runs of the same prompt are not fully independent, so true uncertainty is larger. The practical rules:

• Do not report a change smaller than your margin of error as a win or a loss.
• Per-engine numbers are noisier than the total, because each engine has fewer answers.
• Repetition is not optional. SparkToro found under a 1 in 100 chance that the same prompt returns the same brand list twice. A single run per prompt is closer to an anecdote than a measurement.

The same study offers a useful reference point: across 994 responses to 142 different human-written headphone prompts, Bose, Sony, Sennheiser, and Apple each appeared in 55-77% of answers. Strong brands show up consistently even when the wording changes, which is exactly what presence rate and share of voice capture.

Why should you report AI share of voice by engine?#

Report by engine because each one retrieves from different infrastructure, and a blended number can hide a serious gap on the engine your buyers use most.

The retrieval sources differ by company. Google says its AI Overviews and AI Mode draw on its search index and may use "query fan-out" across subtopics. OpenAI's ChatGPT search uses OAI-SearchBot. Anthropic uses Claude-SearchBot, and Perplexity uses PerplexityBot. If one of those bots is blocked, or one index covers your category poorly, that engine's share collapses while the others look fine.

Platform reports do not fill this gap. Google's Search Generative AI performance reports show your own URL impressions in Google's AI features, without competitors or unlinked mentions. That is useful context, but it is not share of voice. For more on what each platform number can and cannot tell you, see AI visibility metrics.

What is a good AI share of voice?#

There is no reliable public benchmark. Share of voice depends on how many competitors you track, how crowded the category is, and which prompts you include, so a number from someone else's panel does not transfer to yours.

Better reference points:

• The leader in your own panel. In the example, Competitor A at 38.2% defines the bar.
• Your trend. Three or more cycles of the same panel show direction, which is what matters for decisions.
• Your fair share. With four tracked brands, an even split would be 25%. Above that, you are outperforming the average tracked competitor; below it, you are not.
• The gap between presence and citation. A high presence rate with low citation share means engines know you but do not rely on your content.

Be skeptical of vendor benchmarks that give a "good" range without publishing the panel size, competitor count, and engines behind it.

How do you increase AI share of voice?#

You increase AI share of voice by becoming easier to retrieve, easier to quote, and more often recommended by sources the engines trust. The work splits between your pages and other people's.

On your site:

1. Fix crawler access for every engine's search bot. A blocked bot is a zero for that engine.
2. Publish pages that answer your panel's questions directly, especially comparison, alternatives, and "best for" pages with real tables.
3. Add evidence. In the GEO study by Aggarwal et al. (KDD 2024), adding quotations raised visibility in AI answers by about 41%, statistics by about 33%, and cited sources by about 28%.

Off your site:

4. Find the sources engines cite for your panel and check whether you appear in them. Our AI competitor analysis guide shows how to build this source gap list.
5. Correct and contribute to comparison articles and review profiles that describe you inaccurately or leave you out.
6. Take part in the discussions engines cite, including Reddit, with a disclosed account and within each community's rules. Pew Research found Reddit among the most-linked sites in Google's AI summaries. Reddit's rules require authentic participation and ban content manipulation, so never use undisclosed brand posts, multiple accounts, or vote manipulation. The Reddit AEO guide covers how to do this well.

Then re-measure on the same panel and compare against your margin of error.

How should you report AI share of voice?#

Report one page per cycle with the headline number, its context, and what changed. Executives need the trend; practitioners need the breakdown.

A simple monthly template:

Doing this by hand means running hundreds of prompts a month and tagging every answer. Sleeng automates it: it generates buyer-intent prompts from your URL, runs them on ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews, and AI Mode, logs every mention and citation, and shows your AI share of voice against your competitors over time. If you are comparing options, our roundup of LLM rank trackers covers what to look for.

What mistakes distort AI share of voice?#

Most distortions come from changing the inputs between cycles or counting inconsistently. Avoid these:

• Changing the prompt panel or competitor list between cycles without restating past numbers.
• Counting branded prompts in the main panel, which inflates your share.
• Counting every repetition of a name inside one answer instead of capping at one.
• Running each prompt once and treating the result as stable.
• Blending engines and missing a collapse on one of them.
• Ignoring sentiment. A rising share driven by negative mentions is not a win.
• Reporting tiny changes that fall inside the margin of error.

Measure it before you try to move it#

AI share of voice is only useful when the method is fixed and the sample is large enough to trust. Set the panel, the engines, the runs, and the competitor list once, then let the trend tell you what is working.

Sleeng runs that panel for you across six AI surfaces and tracks it every cycle. Start a 14-day free trial to get your baseline share of voice against the competitors AI already recommends.

Sources#

• GEO: Generative Engine Optimization, Aggarwal et al., arXiv 2311.09735 (KDD 2024)
• SparkToro: AIs are highly inconsistent when recommending brands or products (January 28, 2026)
• Google Search Central: AI features and your website
• Google Search Central Blog: Introducing Search Generative AI performance reports in Search Console (June 3, 2026)
• OpenAI: Overview of OpenAI crawlers
• Anthropic: Does Anthropic crawl data from the web, and how can site owners block the crawler?
• Perplexity: Perplexity crawlers
• Pew Research Center: Google users are less likely to click on links when an AI summary appears in the results (July 22, 2025)
• Reddit Rules