Reading time: 12 minutes
To calculate share of voice in AI, divide your brand’s mentions by the mentions of all tracked brands, then multiply by 100. The share of voice formula is simple, but the result depends on what counts as a mention, on position and on the competitors. So this guide shows how to measure share of voice, with a step-by-step example.
In short
- The share of voice formula divides the brand’s mentions by those of all tracked brands, on the same prompts.
- Each tool does the math its own way, and Ahrefs weights each answer by search volume. So compare only within the same tool.
- In our example, the same brand has a 20% simple share, 16% with position weights and 15% of recommendations.
- In advertising, brands with share of voice above market share tend to grow. In AI answers, however, we found no public study showing the same.
What is share of voice?
Share of voice is the slice a brand takes of the category conversation, compared with its competitors. In advertising, the math used media spend, while on social media it uses mentions. Finally, in AI answers, the metric shows the brand’s share among the brands cited on a fixed list of prompts.
On social media, the tool collects what people have already posted. In AI, however, the answer is born from the question. In other words, the prompt list defines the universe of the math. So teams that already measure share of voice on social media can use the same formula, as long as they fix prompts, engines and competitors. To combine the two, see the guide to brand monitoring in AI.
The metric is also not the presence rate. At Peec AI, for example, a brand cited in 4 of 10 chats has 40% visibility. But if the competitor gets 12 mentions, the brand’s share drops to 25%. Likewise, the AI visibility score and share of model have their own math.
How to calculate share of voice?
To calculate share of voice, count the answers in which each brand appears and add up the mentions of all tracked brands. Then divide your brand’s mentions by that sum and multiply by 100. The result is your brand’s percentage of all mentions.
Share of voice formula: brand mentions ÷ (brand mentions + competitor mentions) × 100.
The share of voice formula looks the same everywhere, but four choices change the number:
- Unit: count the brand once per answer, even if the name repeats, as Ahrefs and Otterly do.
- Denominator: use only the tracked brands or every brand cited. With every brand, each share shrinks.
- Adding up engines: add mentions and totals before dividing, instead of averaging the engines.
- Weights: decide whether position and recommendation count for more.
Adding up does make a difference. In the Otterly example, the brand has 25% in ChatGPT and, by our math, 40% in AI Overviews. Combined, the two engines give 30%, but a simple average would give 32.5%.
What counts as a mention?
Count every answer that names the brand, its products or variations of the name. For everything else, set rules before you collect:
- Variations: brand, products and abbreviations add up to one mention per answer. At Otterly, “Nike” and “Nike Air Jordan” in the same answer count once.
- Link without the name: a site that shows up as a source, without the name in the text, goes to the citation rate.
- Negative mention: it counts, but log the sentiment next to it.
- Prompt with the brand name: leave it out, since the answer tends to cite the brand you asked about.
How to measure share of voice in AI answers?
To measure share of voice in AI answers, run the same list of unbranded prompts, always in new chats. Use the engines your audience uses and repeat the collection throughout the month. Then log each brand cited, its position and whether the answer recommends it.
The full protocol, with country, language and personalization, is in the guide to measuring brand mentions and citations in AI. In addition, the collection spreadsheet can feed an SEO dashboard in Looker Studio, next to GA4 and Search Console.
How does each tool do the math?
According to the official documentation, the tools change the numerator, the denominator or the weight when they calculate share of voice:
| Tool | How it calculates |
|---|---|
| Profound | Answers with the brand ÷ total brand mentions |
| Peec AI | Brand mentions ÷ total brand mentions |
| Otterly | Brand mentions ÷ mentions of all brands, one per prompt, engine and day |
| Ahrefs | Share of impressions, estimated from search volume, against the tracked brands |
| Semrush | Share of mentions for each brand, with no published formula |
At Profound, 20 brand mentions out of 100 total brand mentions give 20%. At Ahrefs, on the other hand, a mention on a high-volume prompt is worth more, because impressions add up the search volume of the prompts. Also, across engines, it takes an average weighted by impressions. The Ahrefs Brand Radar review shows how to read these estimates.
Semrush, for its part, describes the metric as the market share of mentions, but it does not show the math. Therefore, compare numbers only within the same tool. To choose one, see the GEO tools comparison. NotHuman, which publishes this blog, works in the same market and stayed out of the table.
How many prompts and competitors go into the calculation?
Use 40 to 60 unbranded prompts, run 5 times a month on each engine, and 3 to 5 competitors, as in the measurement protocol. With fewer prompts, the margin of error grows. With too many competitors, however, each brand’s share shrinks.
Pick competitors from the answers, not only from the client’s opinion. Run the list once and keep the brands the AI cites most. Then freeze the list for a quarter, because Ahrefs itself warns that the metric changes a lot depending on the brands you set up.
If you swap a competitor, recalculate the series with the stored answers. Otherwise, the change in the list will look like a change in the AI. Finally, calculate share of voice every month. In the client SEO report, show share of voice with the margin and the list of competitors.
How to weight position and recommendation?
Give the brand points based on the position where it appears, and count the answers that recommend it separately. After all, a brand cited first and suggested as an option takes up more space. But set the weights before you collect and keep them across the series.
Otterly, for example, turns average position into a likelihood-to-buy index. Thus, first position is worth 100%, second 77.5% and third 55%, down to a floor of 10%. The documentation, however, cites no study that supports the curve.
To calculate share of voice with weights, use a simple scale. Give 3 points to the first brand cited, 2 to the second and 1 to the rest, counting every brand in the answer. Then divide your brand’s points by the points of all tracked brands.
For recommendation, use the protocol’s rule. It counts when the answer suggests the brand as an option, in a list or with a verb like “consider”. Then divide your brand’s recommendations by those of all brands.
Because position changes a lot between runs, keep the simple math as the main number.
Example: how to calculate share of voice step by step
Picture a cosmetics brand with four competitors in ChatGPT and Gemini. The team runs 50 unbranded prompts, 5 times a month on each engine. That makes 500 answers. The numbers are illustrative, but the math works for any category.
- Count the mentions. Your brand appears in 150 answers, which means 30% presence. The five brands add up to 750 mentions.
- Do the simple math. To calculate share of voice, divide 150 by 750 and multiply by 100. That gives 20%.
- Apply position weights. With 3, 2 and 1 points per position, your brand gets 240 of 1,500 points, or 16%.
- Separate recommendations. The AI suggests your brand in 45 of the five brands’ 300 recommendations: 15%.
- Calculate the margin of error. By our math, at 95% confidence, the 20% has a margin of ± 3 to ± 8 points.
| Brand | Mentions | Weighted points |
|---|---|---|
| Your brand | 150 | 240 |
| Competitor B | 250 | 580 |
| Competitor C | 200 | 400 |
| Competitor D | 100 | 190 |
| Competitor E | 50 | 90 |
| Total | 750 | 1,500 |
Your brand has the fair share, 20%, but drops to 16% with weights. In other words, the AI remembers it, but rarely in first place. Competitor B, by contrast, rises from 33% to 39%.
The low end of the margin treats the 750 mentions as independent. The high end, however, counts only the 100 combinations of prompt and engine. So, the following month, only treat a change above 4 to 11 points as real, and only once it is confirmed. To measure share of voice with a smaller margin, add more prompts.
Is AI share of voice linked to sales?
In advertising, it is: brands with share of voice above market share tend to grow. In AI answers, however, we found no public study with an open method that shows the same. For that reason, when you calculate share of voice, treat the number as a sign of consideration, not as a sales forecast.
According to a summary from LinkedIn, which sells ads, LinkedIn’s own B2B Institute analyzed the IPA advertising effectiveness databank with Les Binet and Peter Field. There, a share of voice 10 points above market share matched a gain of 0.7% in share per year in B2B. In B2C, the gain was 0.6%. But that share of voice formula uses media spend, not mentions in AI answers, which change with every run.
To measure share of voice against the business, compare the series month by month with branded searches, leads that mention AI and sales. The financial math follows the guide to SEO and GEO ROI. After all, AI share of voice is one GEO metric, not the only one.
What is a good share of voice?
There is no good number for every brand, because share of voice depends on how many brands compete in the category. So, after you calculate share of voice, compare the result with three benchmarks. With 5 tracked brands, for example, the fair share is 20%.
- Fair share: 100 divided by the number of tracked brands. Above it, the brand takes up more space than average.
- Market share: if the AI share beats the sales share, the AI gives the brand more space than the market does. In advertising, that is a good sign, but there is still no proof for AI.
- Your own series: the change against the baseline, minus the margin, shows whether the work is paying off.
Also look by engine and by prompt group, because the average hides where the brand disappears. To turn the number into a target, use the logic of digital marketing KPIs that prove results.
What we found when we analyzed 7 pages on share of voice
We analyzed 7 pages in Portuguese that showed up in searches about the metric. Six give the share of voice formula: the brand’s number divided by the total, times 100. In addition, four show an example with numbers.
Only three cover AI answers, and one of them presents the percentage of answers with the brand as the key metric. That is the presence rate. Also, only one gives weight to position. None of the three says how many times to run each prompt, and none of the seven mentions a margin of error.
Three repeat that a share of voice above market share signals growth. However, none warns that the rule comes from advertising data.
Frequently asked questions
SOV is share of voice, the brand’s slice of the category conversation or media. SOM, on the other hand, is share of market, the slice of sales. In advertising, brands with SOV above SOM tend to grow, but no public study has shown this for AI answers yet.
To calculate share of voice on social media, divide the brand’s mentions by the mentions of the brand and its competitors in the period. Then multiply by 100. In general, the data comes from a social listening tool. The math is the same as in AI, but in AI the prompt list defines the sample.
Yes. Use one row per answer and one column per brand, with 1 when the brand appears and 0 when it does not. Then add up each column and divide by the sum of all brands. Above a few hundred answers a month, however, collection usually moves to a GEO tool.
Want to know how your brand shows up in AI answers today?
NotHuman is opening spots for agencies and brands to track how they appear in AI answers. Join the waitlist and we’ll let you know when a spot opens up for you.
Main sources
- Profound, Answer Engine Insights metrics (2026).
- Peec AI, how Peec AI calculates share of mentions (2026).
- Otterly, definition of the brand report KPIs (2026).
- Ahrefs, Brand Radar AI visibility metrics (2026).
- Semrush, Brand Performance reports in the AI Visibility Toolkit (2026).
- LinkedIn, the SOV rule in B2B marketing (2020).
