Share of voice (SOV) is the percentage of a category's visibility or conversation that your brand captures, compared with your competitors. Not long ago I saw a report slide that said, in big letters, "our share of voice in AI is 34%". I asked 34% of what, in which engines, against whom and measured how many times. Long silence (and it wasn't the presenter's fault: the tool didn't say anywhere either).
This happens a lot with this metric. SOV is an old and useful metric, as long as you keep fixed what you count and who you compare yourself against. In AI that means measuring how often you show up across a fixed panel of prompts and reading that number next to your market share, instead of treating it like a ranking.
First the classic version, then AI, which is where most of the confusion lives today.
What is share of voice?
Share of voice was born in advertising. In its original version it measured what percentage of a category's media investment a brand put in: if all the beverage brands spent 100 on TV and you spent 20, your share of voice was 20%.
With digital marketing the idea stretched to almost any channel where presence can be counted: ads, social media, press, search results and, more recently, the answers from ChatGPT, Gemini or Google's AI Overviews.
What didn't change is the structure. It is always your part divided by the total of a universe someone defined: which competitors are in, which channel, which period and which market. That's why the same SOV can be 10% or 40% for the same brand, depending on who built the list. The metric measures relative visibility, and relative depends on who you compare yourself with.
How do you calculate share of voice?
The formula is the same in every channel:
Share of voice = (your brand's metric ÷ the category's total metric) × 100
The "metric" changes by channel: mentions, impressions, clicks or spend. An illustrative example (made-up numbers): in October there were 5,000 mentions of the five car insurance brands you track, and 750 were about your brand. Your SOV is 750 ÷ 5,000 × 100 = 15%.
The part guides tend to leave out is that you choose the total. If tomorrow you drop the biggest competitor from the list, your share of voice goes up without anyone talking more about you. That's why any report should include the exact list of brands and the period. Without that data, the percentage is better for a pretty slide than for making a decision.
How is share of voice measured in each channel?
Each channel has its own way of counting. This is the summary I use with marketing teams:
| Channel | What you count | Where the data comes from |
|---|---|---|
| Search ads | Your ad impressions ÷ possible impressions | Google Ads impression share |
| Social media | Your brand's mentions ÷ the category's | Social listening tools |
| Press and media | Your brand's stories ÷ the sector total | Media monitoring |
| SEO (organic search) | Clicks on your keywords ÷ total for the keyword group | Search Console, Ahrefs or Semrush |
| AI | Times you are named ÷ panel total | Your own prompt panel or a tracking tool |
Ads are the simplest, because Google Ads gives you the impression share of your ads directly. On social media you need to gather the conversation about your brand and your competitors, because visibility on social media is counted in mentions (I went into more detail in the article about social listening).
In SEO there's a nuance I like to show with my own data. Over the last 90 days, the query "modo ia" showed my article about Google's AI Mode 64,785 times in Search Console, at an average position of 6, and it brought me 16 clicks. If I measure my organic search visibility by impressions, that topic looks great. If I measure it by clicks, I barely exist. Both readings are true: decide which one you report before you build the series.
How does share of voice relate to market share?
SOV measures how much you are seen (an input) and market share measures how much you sell (an output). And there is a measured relationship between the two.
The idea comes from John Philip Jones, who showed in Harvard Business Review in 1990 that brands investing in advertising above their share of sales tended to gain share. Later, Les Binet and Peter Field found that for every 10 points of excess share of voice, a brand gains on average about half a point of market share per year. They measured it with the IPA's database of effectiveness cases (the UK's Institute of Practitioners in Advertising).
That excess has a name: ESOV, the difference between your SOV and what you sell in the market.
An illustrative example: if you sell 15% of the category and your SOV is 25%, your ESOV is +10, and according to that rule you could expect about half a point more share in a year (it's an average across many brands, and your case may turn out differently). With a negative ESOV, the expected outcome is losing ground little by little.
The guides that repeat this figure usually leave out an important detail: that evidence comes from advertising investment. For SOV in AI there is still no published study showing the same relationship with sales. I think it's reasonable to treat it as a possible leading indicator, but I'd cross it with your own pipeline before using it to ask for budget.
What is share of voice in AI?
Share of voice in AI measures how often an AI model names your brand, compared with others, when it answers your category's questions. In practice you build a panel of prompts (the questions your buyer would ask), run them in ChatGPT, Gemini, Perplexity or AI Overviews, and note which brands appear in each answer.
The confusion starts with the math, because two different formulas circulate under the same name (and sometimes a third):
- Presence: answers where your brand appears ÷ total answers in the panel.
- Share of mentions: your mentions ÷ the mentions of all tracked brands.
- Weighted share: the same, with more weight for prompts with more demand.
Look at how the number changes with an illustrative example of 10 prompts and four brands:
| Brand | Appears in | Presence | Mentions (of 22) | Share of mentions |
|---|---|---|---|---|
| Your brand | 6 of 10 | 60% | 6 | 27% |
| Rival A | 8 of 10 | 80% | 8 | 36% |
| Rival B | 5 of 10 | 50% | 5 | 23% |
| Rival C | 3 of 10 | 30% | 3 | 14% |
Presence adds up to 220% because each answer usually names several brands (in an article I wrote for Previsible I explained that they tend to name between three and five). Share of mentions does add up to 100%.
Neither is wrong, but "60%" and "27%" describe exactly the same visibility. In fact, the AI Overview for the search "share of voice en IA" in Chile shows an example that mixes both in a single calculation. If someone gives you a percentage, the first question is which one they used.
The third version is proposed by Ahrefs' Brand Radar, which weights each prompt by an AI-adjusted volume (the Google volume of the keyword behind the prompt, multiplied by a ratio for each platform). In their example, the same result goes from 67% to 31% once weighted. It's useful to estimate exposure, although a keyword's volume on Google and the number of times someone types a similar question are different things. I'd report all three versions and always say which one is the main one.
Why does your share of voice in AI change every time you measure it?
Because models don't always give the same answer. In a SparkToro and Gumshoe study (January 2026) with 2,961 prompts across ChatGPT, Claude and Google's AI, the same list of brands repeated less than 1% of the time. The same list in the same order, less than 0.1%. What was more stable was how often a brand appears across many runs.
In other words, your brand's "position" inside an answer is mostly noise, and presence measured as a frequency is what's worth tracking. That's why I run each prompt several times and look at trends over several months, never a one-week spike.
The other cause of jumps is more homemade: you changed the measurement. If in July you added prompts where you were already doing well or dropped a big competitor, SOV goes up without your real visibility moving (Ahrefs explains this well in its guide to share of voice in AI). It happens more often than you'd think, especially when the person building the report changes.
How do you measure share of voice in AI step by step?
This is the method I use. The longer version of how I measure AI visibility is in the article about AEO.
-
Build the panel. To start, 40 to 60 prompts a month; in enterprise programs we use between 100 and 300. Split them into three layers: discovery ("what options are there for X in Chile"), comparison ("A or B for a company like Y") and decision ("how much does X cost"). Take them from what your customers actually ask.
-
Fix the competitor list. Between three and six brands, the same ones every month. If you change the list, write it down and start a new series.
-
Define country, language and engines. In Latin America this matters more than it seems. A Peec AI analysis found that 66% of prompts written in Spanish triggered at least one background search in English (I covered it in the article about ChatGPT fan-out queries).
And in a study I did for Ahrefs en Español, AI Overviews appeared in 81% of 2,998 popular searches across five countries (79% in Chile), so I'd include them next to ChatGPT. I'd measure Gemini separately, because each engine cites different sources and the combined number hides the weakest one.
-
Run each prompt several times and log five fields: whether you appear, where, how it describes you, which competitors it names and which sources it cites.
-
Calculate presence and share of mentions, by engine and by prompt group. The global number is of little use. Having 40% in comparisons and 8% in pricing tells you where the gap is, and that already defines your content strategy for the quarter.
-
Look at the sources before touching the content. In the 500-prompt study I published at Previsible, 78% of prompts returned at least one cited source, 67% of citations went to niche domains and, in discovery questions, 71% went to comparison lists. If your competitor shows up because it's in five industry comparisons and you're in none, the work is getting third parties to talk about you, more than rewriting your page.
-
Report with the scope written next to it. Something like "38% presence across 60 prompts, ChatGPT and AI Overviews, Chile, against four competitors, October". That way the number can be repeated next month.
What is a good share of voice?
There's no universal number. With three brands on the list, 33% is the average; with ten, 33% is leading comfortably. I read it three ways.
Against your competitors in the same panel, to know who beats you and in which prompt group. Against yourself, as a trend over several months with the same measurement. And against your market share: if you sell 20% of the category and your SOV is 10%, you are most likely losing ground where your buyers start researching, even if sales don't show it yet.
Also, a high share of voice with a negative conversation doesn't help your strategy. On social media that's measured with mention sentiment. In AI the equivalent is description accuracy: a brand that appears a lot, but described with old prices, has a more urgent problem than one that appears little.
Which tools can you use to measure share of voice?
It depends on the channel:
- Ads: Google Ads, with its impression share.
- SEO: Search Console (free, with your site's real impressions and clicks), and Ahrefs or Semrush to estimate how organic traffic is split with competitors.
- Social media and press: social listening tools such as Brandwatch, Meltwater, Sprout Social or Brand24, which count mentions and sentiment.
- AI: prompt tracking tools such as Brand Radar, or your own panel in a spreadsheet. For Google, the Search Console AI report shows data for those surfaces, and in GA4 you can separate visits coming from ChatGPT.
The tool automates the runs, but you define the panel, the brand list and the formula. Two tools can give you different numbers for the same brand and both be well built.
Frequently asked questions
What does a 50% share of voice mean?
That half of what you measured in your universe is yours: half of the mentions, impressions or answers, depending on the formula. With two brands on the list it's a tie; with ten, it's dominating the category.
Is a high share of voice always a good sign?
Not always. If it comes from negative comments, from a wrong description in AI or from a competitor list built in your favor, the number goes up without the business moving.
What I'd do this week
Take 20 real questions from your customers (Search Console, sales and comments all work) and run them three times in ChatGPT, Gemini and Google search, in a clean session. Note which brands appear in each answer and calculate your presence and your share of mentions. Then put them next to your market share. If the gap is big, you already have the conversation for your next budget committee, and a panel you can repeat next month with exactly the same rules.