How to calculate share of voice for SEO, ads and AI search
·9 min read
To calculate share of voice, divide your brand's count by the combined count for you and every competitor you track, then multiply by 100. That is how to calculate share of voice in any channel, and only the thing you count changes. In paid search you count impressions, in SEO the clicks your rankings should earn, in social and PR the mentions, and in AI search the answers that name or cite you across a fixed set of prompts.
The division is the easy part. The result is only as good as your competitor set and your sample. For tracking AI answers in general, see AI visibility. This post is the arithmetic, with made-up numbers for each channel.
The share of voice formula
The share of voice formula is one division, and the shares of everyone in the set add up to 100%.
share of voice = your count ÷ (your count + every tracked competitor's count) × 100Add a competitor and every other share falls, even though nothing changed in the market. Pick four to six brands your buyers compare you with, and keep the set.
| Channel | What you count | Where the counts come from | Optional weighting |
|---|---|---|---|
| Paid search | Ad impressions | Google Ads impression share and auction insights | None needed |
| Organic SEO | Estimated clicks | Rankings × search volume × CTR at each position | Built in |
| Social and earned | Mentions | A social listening or media monitoring tool | Sentiment or reach |
| AI search | Answers that name you, or cite your site | A fixed prompt set run on chosen engines | Demand for each prompt |
One search shows several ads and one AI answer names several brands, so a rate like "named in 24% of answers" and a share like "15% of all mentions" are different numbers. Label which one you report.
How to calculate paid share of voice from impression share
Paid share of voice starts from impression share, which Google Ads defines as your impressions divided by the impressions you were eligible to receive. At 50%, your ads missed half the searches they could have shown on.
To split the space between rivals, take the impression shares of other advertisers from the auction insights report, multiply each by your eligible impressions, then divide each result by the total. Example numbers, made up, for a campaign eligible for 80,000 impressions in a month:
| Advertiser | Impression share | Estimated impressions | Paid share of voice |
|---|---|---|---|
| You | 50% | 40,000 | 25.0% |
| Rival A | 80% | 64,000 | 40.0% |
| Rival B | 45% | 36,000 | 22.5% |
| Rival C | 25% | 20,000 | 12.5% |
| Total | 200% | 160,000 | 100% |
The impression shares add to 200% because, on average, each eligible search showed two of these ads. Your 40,000 of 160,000 estimated impressions is 25%, the same as 50 ÷ 200.
Treat it as an estimate. Each rival's auctions only partly overlap yours, and the report doesn't show insights when impression share is under 10%, so the smallest advertisers drop out and everyone left reads a little high.
How to calculate SEO share of voice from rankings
SEO share of voice is each site's estimated organic clicks across a fixed keyword list, divided by the total for all tracked sites. A keyword's estimated clicks are its monthly searches times the click-through rate at the position the site holds.
estimated clicks = monthly searches × CTR at that position
SEO share of voice = your estimated clicks ÷ all tracked sites' estimated clicks × 100Take the CTR curve from your own Search Console data, as clicks divided by impressions for queries grouped by rounded position. It already reflects the ads and AI Overviews above your results.
Example numbers, made up, for a standing desk store and two rivals. The CTR curve here is 30% for 1st, 15% for 2nd, 10% for 3rd, 7% for 4th, 5% for 5th, 2% for 6th to 10th and nothing below that.
| Keyword | Monthly searches | You | Rival A | Rival B |
|---|---|---|---|---|
| standing desk | 40,000 | 5th, 2,000 | 1st, 12,000 | 3rd, 4,000 |
| electric standing desk | 12,000 | 2nd, 1,800 | 4th, 840 | 1st, 3,600 |
| standing desk for small spaces | 3,000 | 1st, 900 | 7th, 60 | Not ranked, 0 |
| best standing desk converter | 5,000 | Not ranked, 0 | 2nd, 750 | 3rd, 500 |
| Total | 60,000 | 4,700 | 13,650 | 8,100 |
| SEO share of voice | 17.8% | 51.6% | 30.6% |
You hold 4,700 of the 26,450 tracked clicks, or 17.8%. First place on the small-spaces keyword earns you 900 clicks, while 5th on "standing desk" earns 2,000, because that term carries two-thirds of the volume. Move from 5th to 3rd there, with Rival B slipping to 4th, and your share rises to 24.6%. A raw count of top-10 rankings would treat 7th on a 3,000-search keyword the same as 1st on a 40,000-search one.
Dividing by the list's total monthly searches instead gives 4,700 ÷ 60,000, or 7.8%, a version that doesn't move when you add a competitor. Check which denominator a tool uses before you compare its number with yours.
How to measure social and earned share of voice
Social and earned share of voice is your brand's mentions divided by the mentions of every tracked brand, over the same period and from the same sources. Example numbers, made up, for one month:
| Brand | Mentions | Share of voice | Negative mentions | Share without negatives |
|---|---|---|---|---|
| You | 340 | 20.0% | 20 | 23.9% |
| Rival A | 910 | 53.5% | 110 | 59.7% |
| Rival B | 450 | 26.5% | 230 | 16.4% |
| Total | 1,700 | 100% | 360 | 100% |
Just over half of Rival B's mentions were complaints. Raw counts reward bad news, so report sentiment next to the share or drop negatives, as the last column does.
Use one saved query per brand, with the same sources and filters for every brand, and include misspellings and product names. Count a press release syndicated to 40 sites as one mention, or a wire service will buy you share of voice.
How to calculate share of voice in AI search
AI share of voice is the share of brand mentions, or of cited sources, that go to you across a fixed prompt set, run on fixed engines a fixed number of times.
Freeze the prompt set first. Use 20 to 50 questions in your buyers' words, mostly category and problem questions such as "best standing desk for a small apartment". Keep branded prompts out, because a prompt that names a brand returns that brand. AI Question Finder lists the 25 most searched questions that Google AI Overviews or ChatGPT answer about a topic, with monthly searches.
Mention share of voice
Run each prompt on each engine more than once and record every brand each answer names. Example numbers, made up, for 25 prompts on ChatGPT, Gemini and Perplexity, run twice each, which gives 150 answers:
| Brand | Answers that name it | Mention rate | Share of voice |
|---|---|---|---|
| You | 36 | 24% | 15.0% |
| Rival A | 96 | 64% | 40.0% |
| Rival B | 66 | 44% | 27.5% |
| Rival C | 42 | 28% | 17.5% |
| Total | 240 mentions | 160% | 100% |
Mention rate divides by the 150 answers and adds to 160%, because most answers name more than one brand. Share of voice divides by the 240 mentions and adds to 100%.
Work out each engine before you blend them. Your 36 mentions split into 16 of 80 in ChatGPT, 6 of 75 in Gemini and 14 of 85 in Perplexity, which is 20%, 8% and 16.5%. The blended 15% hides that Gemini barely names you.
To weight prompts by demand, multiply each prompt's mentions by the monthly Google searches for the same question. Prompt volumes aren't public, so search volume is a rough stand-in.
Citation share of voice
Citations need their own table, because answers cite review sites, forums and publishers next to brand sites. Divide each domain's count by all citations. If the 150 answers list 410 sources and 22 link to your site, your citation share is 5.4%. Treat the domains above you as a to-do list. When a review page is cited more than any brand, getting onto it does more than another post on your own blog. More on that in how to get mentioned by AI.
Search Console can't give you this number. Its Generative AI performance report shows your own impressions in AI Overviews and AI Mode combined, and nobody else's, so there is no denominator.
Mistakes that skew share of voice
The formula rarely goes wrong. The sample and the setup do.
Samples too small to show a change
A share built from sampled answers has a margin of error. For a rate p measured on n answers, the 95% margin is roughly this:
margin = 1.96 × √(p × (1 − p) ÷ n)| Answers in the sample | Likely range for a measured mention rate of 24% |
|---|---|
| 50 | 12% and 36% |
| 150 | 17% and 31% |
| 600 | 21% and 27% |
Halving the margin takes four times the answers. A month-on-month change has to beat about 1.4 times the margin to count, because both months carry error. At 150 answers that is roughly 10 points, so a move from 24% to 30% could be noise. Repeat runs of one prompt aren't independent either, so the true margin is wider still.
Reading one AI run as the answer
The same prompt returns a different brand list on the next run. In a study by SparkToro run with Gumshoe, an AI tracking vendor, about 600 volunteers ran 12 prompts through ChatGPT, Claude and Google's AI 2,961 times. ChatGPT and Google's AI gave the same brand list on two runs of a prompt less than 1 time in 100, while how often each brand appeared held steadier than its position. Count appearances across repeat runs and treat position as a rough guide. The causes are in why ChatGPT gives different answers.
Adding mentions and citations together
Keep them as two numbers. An answer can cite your guide while recommending three rivals, and a combined count would make that look like a win.
Measuring competitors on different prompts
Measure every brand on the same prompts, engines, runs and dates. A vendor's report built on its own prompts can't be compared with yours, and neither can last quarter's number if you reworded prompts since. Track new prompts as a second set.
A monthly share of voice routine
Run it in the first week of each month, with every input frozen.
- Write down the competitor set, keyword list, prompt set, engines, runs per prompt and settings. Change them only at the start of a quarter.
- Export auction insights and convert impression shares to estimated impressions.
- Pull rankings and apply your CTR curve. Refresh search volumes quarterly.
- Rerun the saved social queries, merge syndicated copies and tag negatives.
- Run the prompt set on every engine at least twice, with memory off, within a few days. Record brands named, domains cited and the date.
- Calculate each channel and engine on its own. Never average them into one score, since an impression, a click and a mention are different units.
- Compare with last month only where the change beats the margin, and read the answers behind any move before you explain it.
Find which sites AI answers cite for your topic
Our AI Citation Finder gives you the citation side of AI share of voice without running prompts. Enter a topic and it returns the 25 pages cited in the most Google AI Overviews and ChatGPT answers, each with its answer count and the monthly searches behind it. It also lists the sites cited most, each with its share of the topic's answers. That share is a citation rate, so the shares add to more than 100%. If your rivals appear among them, divide each one's answer count by the total for your tracked sites to get share of voice.
Choose Google AI Overviews, ChatGPT or both, in any of 20 countries or all of them. A domain lookup counts the answers that cite that site, yours or a competitor's. The data provider tracks the questions people ask most, with far fewer ChatGPT answers than Google ones. The tool doesn't cover Perplexity or Gemini or read what answers say, so it counts citations, not mentions. A run costs 160 credits.