The short answerClassic share of voice is your brand's mentions divided by all tracked brands' mentions, times 100. AI search adds visibility, citation share and average position. Across 141 B2B agency questions, 28 days to 18 October 2026 (Peec AI), visibility and mention share crowned Omniscient on ChatGPT, Perplexity and Google AI Overviews. Citation share and average position picked other leaders.
Most AI visibility tools sell you a share of voice number. Vendors publish different formulas, the number rarely says which one made it, and the formula decides who wins.
We ran four formulas on the same data: 141 B2B agency buyer questions, tracked daily for 28 days on ChatGPT, Perplexity and Google AI Overviews.
Omniscient was named in more AI answers than any other agency on all three engines, yet its own site ranked 8th, 3rd and 13th by citation share.
The four formulas, and who each one crowned
| Formula | What it counts | Divided by |
|---|---|---|
| Visibility | Answers that name the brand | All answers in the question set |
| Mention share (classic share of voice) | Times the brand is named | Times any of the 53 tracked agencies is named |
| Citation share | Citations to the brand's own domain | All citations in the same answers |
| Average position | The brand's rank in answers that name it (lower is better) | Answers that name the brand (200 or more ranked appearances required) |
The leader under each formula, over the same 28 days:
| Formula | ChatGPT | Perplexity | Google AI Overviews |
|---|---|---|---|
| Visibility | Omniscient 44.6% | Omniscient 34.1% | Omniscient 33.7% |
| Mention share | Omniscient 12.7% | Omniscient 14.9% | Omniscient 11.4% |
| Citation share | LoudFace 11.3% | LoudFace 6.6% | LoudFace 3.5% |
| Average position | Discovered Labs 3.16 | LoudFace 2.73 | LoudFace 2.45 |
Report one formula and you are reporting one of up to three different leaderboards.
What is the share of voice formula?
Sprout Social states the general form of share of voice as "Share of voice = Your brand metrics / Total market metrics". The metric on top changes by channel, and the denominator is always the whole market for that metric.
Each channel picked its own metric:
- Paid search uses impression share. Google Ads defines it as "the percentage of impressions that your ads receive compared to the total number of impressions that your ads could get."
- SEO tools use clicks. Ahrefs says its share of voice "shows the percentage of traffic that your target website gets, compared to the traffic of other websites."
The reason anyone cares goes back to Nielsen. Its analysis of 123 brands across 30 categories found that "a brand whose share of voice (SOV) is greater than its share of market (SOM) is more likely to gain market share", with about 0.5% of extra share growth for every 10 points of gap. Nielsen framed that norm for consumer goods brands. The principle carried over anyway: be heard more than your size and you tend to grow.
AI answers broke the tidy version. An answer can name a brand, link to a brand's page, do both, or do neither. Each of those is a different event, so each gets its own formula.
How is AI share of voice calculated?
It depends on whose tool you open.
| Source | Formula as published | Unit |
|---|---|---|
| Peec AI, visibility | Responses mentioning your brand divided by total responses, times 100 | Answers |
| Peec AI, share of voice | Your brand's mentions divided by all brand mentions, times 100 | Mentions |
| Semrush, blog formula | Your AI mentions divided by total AI mentions across all brands in your category, times 100 | Mentions |
| Semrush, Enterprise AIO | Mentions weighted by your position in the answer, plus search volume on ChatGPT | Weighted mentions |
| Nightwatch | Your brand mentions divided by total brand mentions across tracked prompts, times 100 | Mentions |
| LLM Pulse | Your mention count divided by total brand mentions across relevant queries | Mentions |
| Shadow | Responses that cite, mention or recommend you, divided by all responses for the query set | Answers |
| Scrunch | Mentions or citations, as a share of the same metric across the competitive set | Either |
Two main denominators hide in that table: answers and brand mentions. Scrunch also allows citations, and Semrush Enterprise AIO weights by position. Peec's own docs show how far apart they land: a brand named 4 times in 10 chats, beside a competitor named 12 times, has a 25% share of voice and, if each mention sits in a different chat, 40% visibility.
Even one vendor can be loose about it. Peec's share of voice page describes the denominator as all brands mentioned, including yours, and in another line as your tracked competitors. Its metrics overview says all tracked brands mentioned. Ahrefs calls its click-based share of voice the equivalent of a visibility score, while Peec treats visibility and share of voice as two separate metrics.
None of this is wrong. The formulas are published, but the number on a dashboard rarely says which one made it. A 25% and a 40% can describe the same week for the same brand.
Citations are a separate layer again. Peec defines sources as the URLs a model reads while answering and citations as "the sources explicitly referenced in the final answer." Its metrics overview puts the gap plainly: "your brand can be mentioned even when your website isn't used as a source."
Four formulas, one dataset: who leads on each engine?
The tables below use the top three under each formula. Each engine is read on its own, because a blended number hides which engine a brand is losing on.
ChatGPT (3,948 answers, 39,765 agency mentions, 36,086 citations)
| Formula | 1st | 2nd | 3rd |
|---|---|---|---|
| Visibility | Omniscient 44.6% | LoudFace 37.7% | Siege Media 27.5% |
| Mention share | Omniscient 12.7% | LoudFace 11.0% | Siege Media 6.6% |
| Citation share | LoudFace 11.3% | Derivatex 3.0% | Breaking B2B 2.7% |
| Average position | Discovered Labs 3.16 | Derivatex 3.18 | Omniscient 3.19 |
Perplexity (3,948 answers, 11,208 agency mentions, 18,072 citations)
| Formula | 1st | 2nd | 3rd |
|---|---|---|---|
| Visibility | Omniscient 34.1% | First Page Sage 17.8% | Siege Media 17.4% |
| Mention share | Omniscient 14.9% | Siege Media 7.4% | First Page Sage 7.3% |
| Citation share | LoudFace 6.6% | Derivatex 5.3% | Omniscient 3.3% |
| Average position | LoudFace 2.73 | Derivatex 2.85 | Omnius 3.44 |
Google AI Overviews (3,876 answers, 14,206 agency mentions, 38,203 citations)
| Formula | 1st | 2nd | 3rd |
|---|---|---|---|
| Visibility | Omniscient 33.7% | Skale 20.0% | First Page Sage 19.9% |
| Mention share | Omniscient 11.4% | Skale 6.6% | First Page Sage 6.5% |
| Citation share | LoudFace 3.5% | Revv Growth 3.4% | Optimist 3.2% |
| Average position | LoudFace 2.45 | Optimist 2.72 | AEO Engine 2.84 |
Two patterns hold on every engine. A third shows up on ChatGPT.
The two mention formulas agree. Visibility and mention share had 9 of their top 10 agencies in common on ChatGPT and all 10 on Perplexity and Google AI Overviews, with the same leader everywhere. In this category, if your tool reports one of them, you know roughly what the other says.
The citation formula disagrees with both. The top 10 by visibility and the top 10 by citation share had 3 agencies in common on ChatGPT, 5 on Perplexity and 2 on Google AI Overviews.
On ChatGPT, average position adds a third leader. It put Discovered Labs first, an agency named in 10.6% of answers. On the other two engines, position and citation share picked the same leader.
Why naming and citing split apart
Omniscient is the clearest case: first by visibility on every engine, and 8th, 3rd and 13th by citation share (2.0%, 3.3% and 1.0%). Siege Media is third by visibility on ChatGPT at 27.5% and 12th by citation share there, at 1.1%. On Google AI Overviews its citation share rounds to 0.0%, 47th. iPullRank sits 5th or 6th by visibility on all three engines and 22nd, 37th and 37th by citation share.
The movement runs the other way too. Derivatex is 11th on ChatGPT by visibility (13.6%) but 2nd by citation share (3.0%) and 2nd by average position (3.18). Revv Growth is 2nd by citation share on Google AI Overviews (3.4%) and 17th by visibility (6.5%).
So engines name some agencies they barely cite, and cite some agencies they barely name. Other studies find the same split. Semrush and Kevin Indig logged 3,981 domain appearances across 115 prompts and found 61.7% were ghost citations, sources cited without the brand being named. In their data Gemini named brands 83.7% of the time and cited them 21.4%, and ChatGPT did the reverse. Ahrefs tracked one brand across platforms and found AI Overviews linked it only 10.7% of the times it was mentioned, against 51.6% on Perplexity.
LoudFace's own numbers show the same split. loudface.co leads citation share on all three engines, while LoudFace is 2nd by visibility on ChatGPT, 5th on Perplexity and 8th on Google AI Overviews. A citation means the engine used your page as evidence. A mention means it named you in the answer. They are different jobs.
A large share of the evidence comes from sites outside the 53 tracked agencies' own domains. Tracked agencies' own domains took 45.8% of citations on ChatGPT, 44.6% on Perplexity and 33.2% on Google AI Overviews. On Google AI Overviews the two most-cited domains overall were linkedin.com and youtube.com.
Average position flatters the rarely named
Average position only counts answers where the brand appears. A brand named in two answers, first in both, gets a perfect score. That is why we set a floor of 200 ranked appearances per engine before ranking anyone.
Even with the floor, position rewards the specialist. On ChatGPT, Discovered Labs (3.16) edges Omniscient (3.19), yet Discovered Labs is named in 10.6% of answers and Omniscient in 44.6%. Position is a useful second read, "when they name us, are we near the top?", and a poor headline. Peec's win rate is the stricter cousin: answers where you rank first, divided by every answer, including the ones that never name you. It can never exceed your visibility.
Which formula should you report?
Match the formula to the question you are paying to answer.
| Your question | Report this | The catch |
|---|---|---|
| When buyers ask, do they hear our name? | Visibility, per engine | it reads as answers out of 100 and hides how many rivals share each answer |
| Are we gaining on named rivals? | Mention share, per engine | it sums to 100% across the tracked set, so adding brands lowers everyone |
| Do engines use our pages as evidence? | Citation share, per engine | a brand can lead it while its name is missing from the answer |
| When we are named, are we named early? | Average position, with a floor | it only counts answers that name you, so it flatters rare brands |
Our position: lead with visibility per engine on the buyer questions that matter to your pipeline, put mention share beside it, and treat citation share as a diagnostic for whether your pages are doing work. Never blend the engines. In our data ChatGPT produced 39,765 agency mentions against 11,208 on Perplexity, so a blend mostly reports ChatGPT.
When an agency or a tool hands you one AI share of voice number, ask what is on top, what is underneath and which engine. If the answer is one blended figure over mentions and citations together, you cannot act on it. To check where your own brand stands against your top competitors across ChatGPT, Claude, Gemini and Perplexity, run our free AI visibility audit.
What is a good share of voice percentage?
There is no universal number, because the denominator decides the scale. In this category, with 53 agencies tracked, the mention share leader held 11.4% to 14.9% depending on the engine. The visibility leader sat at 33.7% to 44.6%. A brand with 50% share of voice would be named as often as every other tracked brand combined. Nobody here came close.
So judge share of voice against your own set:
- Against the leader. Leaders here were named in a third or more of answers. Single digits on visibility means engines rarely think of you.
- Over time, same formula, same questions. A gain only counts if the question set and the engine stayed fixed.
- Against your market share, if you know it. That is the Nielsen idea, and it still works as a sanity check.
Answers also move day to day. Our answer stability study found ChatGPT repeated its exact list of agencies on consecutive days only 2.9% of the time, so read any of these formulas over weeks, never from one check. Our share of answer guide goes deeper on the visibility reading.
How we measured this
- Questions: the 141 prompts Peec tags as non-branded in the LoudFace project, out of 150 tracked. They are B2B SaaS agency buyer questions that LoudFace chose to track its own category, asked daily in US English.
- Window: 21 September to 18 October 2026, 28 days.
- Answers: we read each engine separately, with 3,948 answers on ChatGPT, 3,948 on Perplexity and 3,876 on Google AI Overviews.
- Brands: the 53 agencies tracked in the project. Mention share divides by mentions of those 53 only. Brands outside the set are not in the denominator, so a larger set would lower every figure.
- Citation share: citations to each agency's own listed domain, divided by all citations in the same answers.
- Average position: Peec's mean rank in answers naming the brand. Peec counts every brand detected in the answer when it assigns rank, tracked or not. Brands with fewer than 200 ranked appearances on an engine were left out of that ranking.
- Data: all of it comes from Peec AI brand and domain reports, read on 19 October 2026. We recomputed every figure from the raw report rows.
This is one category, and the question set is the one LoudFace tracks for it. Treat the absolute numbers as this category's numbers. The finding that travels is the reordering: change the formula and you change the winner.
Related studies on this category: whether all three engines recommend the same agencies, and who AI actually cites in the category. If you want the playbook side, start with how to get named in AI search, not just read.
