Why Your AI Share of Voice Depends on Who Is on Your Competitor List

Share of voice is the AI visibility number most likely to end up in a board deck. Genezio's documentation says it "is usually the number that goes into board decks and quarterly reviews," and Peec AI describes it as how much of the conversation you own.

But share of voice is a fraction, and the bottom half of that fraction is a list of brands. Change the list and the number changes, even if not a single AI answer changed. This article explains what the number divides by, how the list behind it goes wrong in real tracking data, and what to check before you report it.

What AI share of voice actually measures

AI share of voice compares how often AI answers mention your brand with how often they mention other brands. The basic formula is the same across the tools we looked at:

Share of voice = mentions of your brand ÷ mentions of all brands × 100

Peec AI's documentation gives a simple example. If your brand is mentioned 4 times and your competitor 12 times, your share of voice is 25%, because 4 ÷ (4 + 12) = 25%.

The top of that fraction is easy to understand: it is your mentions. The bottom of the fraction, the total it divides by (the denominator), is where the decisions hide. "All brands" only means the brands the tool counts. So the first question to ask about any share-of-voice figure is: which brands are in that total?

Which brands count depends on the tool

The three platforms whose documentation we read answer that question in two different ways.

Peec AI counts your brand against your tracked competitors. Its documentation says share of voice "tells you how often your brand is mentioned compared to all of your tracked competitors." A brand you don't track doesn't enter the total.

Genezio works the same way. Its KPI documentation defines the denominator as "mentions of all brands (you and the competitors)," and its share-of-voice chart shows how mentions divide between your brand and each monitored competitor. If there is no set of competitors, Genezio doesn't show share of voice at all, because a share of nothing doesn't apply.

Profound starts from a wider set. Its metrics documentation divides by "mentions of all brands," and says that for its visibility score "any brand named in the answer counts, not only your configured competitors, unless you have excluded it." Excluding a brand under its Configure Competitors setting changes share of voice for every brand.

Neither approach is simply better. A tracked list keeps the total clean, but it is blind to rivals you never added. A count of every detected brand misses no rival, but it can also count names that aren't competitors, such as search engines or review sites, until you exclude them.

What this means: two teams looking at the same AI answers can report different share-of-voice figures because of how their tools and lists are set up. Compare share of voice only within one tool and one stable configuration, never across tools.

In the tracked-list model, that makes the competitor list itself the main source of risk. So it is worth looking at how lists get built, and where they break.

How competitor lists are built, and three ways they go wrong

No tool asks you to type the whole list from memory. Peec AI suggests brands that appear alongside yours at least twice, and you accept or reject each one. Genezio extracts competitors from the brands that answer engines name together with yours, and you can add others. In both cases, what ends up in the share-of-voice total is a mix of detection and human choices.

To see what that mix looks like in practice, we examined one real Genezio tracking project. It covers the AI visibility software category, so the brand being tracked is Genezio itself, with 11 tracked rivals. It monitors ChatGPT, Microsoft Copilot, Google AI Overviews and Google AI Mode, and tracking started in August 2026. The figures below are the tool's mention counts since then. Three problems showed up.

1. Rivals the answers name often are not on the list

The project had detected 674 brand entries it was not tracking. Many were not competitors at all: Google, OpenAI, Reddit and G2 appear because AI answers name the engines and review sites they draw on. That alone shows a detected-brands list can't be used without review.

But some untracked entries were clear rivals. AI answers named BrightEdge 351 times, more often than four of the tracked competitors. Conductor and Authoritas also appeared repeatedly. None of these mentions entered the share-of-voice total.

2. One brand appears under several names

AI answers don't always use a company's official name. In this project, "Scrunch" appeared as three untracked entries next to the tracked "Scrunch AI." "Bluefish AI" sat next to the tracked "Bluefish," and "Goodie" next to "Goodie AI." Answers also named products rather than companies, such as "BrightEdge Generative Parser" and "BrightEdge AI Catalyst." Unless these variants are merged, each one counts as a separate brand or isn't counted at all.

This is not specific to one tool. Peec AI's documentation advises reviewing recent chats "to see how AI models refer to the brand" and adding name variations as aliases, which is a sign that the problem is common.

3. One rival is split in two, and the tracked half reads zero

The clearest case was SE Ranking. The project held two entries with that exact name. The tracked one had 33 mentions in total. The untracked one had 174.

For the month from 14 September to 14 October 2026, the tracked SE Ranking showed 0% presence and zero share of voice in all four AI engines. Yet in that same month, Google AI Overviews and AI Mode answers named it several times. One AI Overview answer on 1 October described "SE Ranking (AI Visibility Tracker)" and its ChatGPT and AI Overview modules. Those mentions were most likely being counted under the untracked duplicate.

What this means: a rival can look absent from AI answers when it is in fact being mentioned. Anyone reading that report would conclude SE Ranking had disappeared from AI search, and the other tracked brands' shares would be a little higher than they should be.

To see how much a missing rival can matter, go back to Peec AI's simple example, where you have 4 mentions and one competitor has 12, for a 25% share. Now imagine, purely as an illustration, an untracked rival with 8 mentions. Add it to the list and your share falls to 4 ÷ 24, about 17%. Nothing in the answers changed; only the list did.

Fixing the list can rewrite your past numbers

The natural reaction is to clean up the list. That is the right move, but it has a side effect worth knowing before your next quarterly review.

Profound documents this openly. Its page on why past numbers change explains that brand attribution "is not stored with the answer." It is read from your current configuration each time a report loads. Excluding a brand removes its past mentions on every date, so share of voice changes for every brand, back through your history. Profound also notes that an older configuration cannot be viewed.

Peec AI behaves similarly for name changes. Its documentation says that saving changes to a brand's name or aliases triggers a data update, as it searches your existing chats again and recalculates metrics.

What this means: if your share of voice moved between two quarters, the cause may be a change to the setup rather than a change in what AI assistants say. Profound's own advice is direct: if a number must not change, export it at the time of reporting.

How to check the list before you report share of voice

None of this makes share of voice useless. It makes it a number that needs a short audit before it goes in front of leadership. The steps below are suggested practice drawn from the problems above, not a tested method.

1. Read your tool's definition of the denominator. Find out whether it counts tracked competitors only or every detected brand, and what exclusions do.
2. Review untracked or suggested brands by mention count. Promote real rivals, like BrightEdge in the example above, and exclude names that aren't competitors.
3. Merge name variants and duplicates. Use aliases or your tool's equivalent so "Scrunch" and "Scrunch AI" count as one brand.
4. Treat a tracked rival at 0% as a warning sign. Spot-check a few raw answers to see whether the brand is really absent or counted somewhere else.
5. Log every configuration change with a date. When the number moves, you can tell a setup change from a real shift.
6. Snapshot the figures you report. Export them on the day, so later edits can't silently change them.
7. Report the list with the number. State how many competitors the share is calculated against, and show share of voice next to visibility and recommendation rates rather than on its own.

Two related checks also apply. Share of voice can differ sharply from one AI engine to another, so it is best measured engine by engine. And like any AI visibility figure, it needs enough prompt runs to be stable. If you are choosing a platform, add one question to your evaluation: what exactly does your share of voice divide by?

The bottom line

AI share of voice is your share of a list, not of the whole market. When the list is complete, merged and stable, it is a useful way to show leadership where you stand against rivals. When it misses a frequent rival, splits one brand in two or changes between reports, the number moves without anything changing in AI answers. So report the list along with the number.