Why share of voice can rise while your brand mentions stay flat

Share of voice can rise while your brand mentions stay flat because it is a ratio, not a count. In DiscoveredBy it is the answers that name your brand family, divided by the sum of the answer counts of every active tracked brand family, yours and your competitors'. If a rival is named less often, or you pause or remove one, the denominator shrinks and your share grows even though the number of answers naming you has not moved. Brand mentions, read as brand visibility, answer a different question: how often are you named in the answers analysed, regardless of who else is named.

In short

• Share of voice is relative. It moves when your count moves, when a tracked competitor's count moves, and when you change which competitors are tracked.
• Brand visibility is absolute in the sense that it does not depend on your competitor roster. It is the share of analysed answers that name your brand.
• A rising share with flat visibility often points to a change in the field, not in your own count. That can be good news, but it is not evidence that you did anything.
• Report both numbers side by side, with the roster and the number of answers behind them, so a reader can see which one moved.
• Adding, pausing or removing a competitor changes share of voice with no change in any answer. Pausing or removing takes effect from the next load.

What is the difference between share of voice and brand mentions?

Brand visibility asks "in what share of analysed answers were we named?" Share of voice asks "of all the brand naming that happened among the brands we track, how much was ours?"

A mention is your brand named in an AI answer (see Key terms). The metric built from mentions is brand visibility: analysed answers naming your brand, divided by analysed answers. An analysed answer is a completed answer whose text has been read for the brands it names. Your competitors are not in that formula at all.

Share of voice is defined as distinct answers naming your brand family, divided by the sum, over every active tracked brand family, of that family's own distinct-answer count. A brand family is a top-level brand plus its sub-brands, counted once per answer.

The practical consequence: visibility has one moving part (your count against the analysed total), while share of voice has as many moving parts as there are families in your roster. The glossary entries for share of voice and brand mention give looser, general-purpose definitions; the formulas above are the ones the product uses.

Enlarge image

Why can share of voice rise when nothing changed for you?

There are four common ways, and only some of them say anything about your brand. The first three change the denominator or the competitors' counts and leave your own count where it was; the fourth changes everyone's.

1. A competitor was named in fewer answers. Their count falls, the sum falls, your share of it rises. This is a real change in the answers, but it is a change about them.
2. You paused or removed a competitor. The metrics page states that a paused or removed competitor's mentions stop counting at once: the mentions already recorded are kept, but they are not counted while the competitor is paused. Pausing can raise your share with no change in any answer.
3. A name was added as an alias or sub-brand of a competitor. The platform moves the project's past untracked mentions that used exactly that name onto the family. That raises the competitor's count and changes the sum, with no new answer.
4. The answers you collect changed. Different prompts, engines or days in the window change everyone's counts. This one usually moves your visibility too, so it is easier to spot.

The reverse also happens. Adding a competitor can lower your share of voice with no change in the answers at all, because the sum now has one more family in it. The metrics reference says this plainly: the roster you build sets the denominator.

Is brand visibility immune to all of this?

No. Brand visibility ignores your roster, but it still depends on the population of analysed answers and on when mention extraction has run.

Its denominator is analysed answers. Mention extraction runs after collection, so the analysed population for a window is a subset of the collected one and can lag it. A week that looks flat may simply be a week where extraction had not caught up when you looked. If you change your prompt list or engines, visibility can move for reasons that have nothing to do with how the engines describe you; a fixed set of prompts keeps it easier to compare over time.

Volume matters too. A metric that reports an observation count is treated as provisional below 30 observations. A one-point move on a small sample is not a trend. For a related diagnosis, see Why your AI visibility score changed when your website did not.

There is also a subtle point about "flat". If your mention count is flat but the number of analysed answers grew, visibility fell. If the analysed total shrank, visibility rose. When you say mentions are flat, be explicit whether you mean the count or the rate, and label your table accordingly.

Which number moved, and what should you conclude?

Read the two metrics together and use the pair to decide what to look at next. The table is a set of hypotheses to check, not verdicts.

The Explorer can break these metrics down by engine, prompt, country and brand, among other dimensions, so you can see which slice moved instead of guessing.

A worked example

Illustrative example: Quillstone and its competitors are fictional, and the numbers are made up to show the method.

Quillstone sells document-review software to legal and compliance teams. It tracks three competitors: Brieflane, Clausewise and Docket North. In both weeks below, 200 answers were analysed. Counts are the number of distinct answers naming each brand family. An answer naming two brands appears in both counts, which is why the family counts can add up to more or less than the answer total.

Quillstone's brand visibility is 40 of 200 in both weeks, which is 20%. Its share of voice is 40 of 180 in week 1 (22.2%) and 40 of 140 in week 2 (28.6%). Share of voice rose by about 6 points. Visibility did not move at all.

The correct report line is not "our AI share of voice grew". It is "we were named in the same number of answers; Brieflane and Clausewise were named in fewer, so our slice of a smaller pie is bigger". That leads to a useful question (why did two rivals drop?) instead of a false claim of progress.

Now hold week 1 fixed and change only the roster:

Same 200 answers, same 40 for Quillstone, three different shares. Only the roster changed (in the last row, assume Pleadwell's 30 week-1 answers are counted). That is why every share of voice figure you publish needs its roster printed next to it.

Enlarge image

The two-metric reporting pattern

Report share of voice and brand visibility as a pair, and print the context that lets a reader interpret them. This is the deliverable: a block you can paste into a monthly report.

AI VISIBILITY: TWO-METRIC LINE
Period: [dates]            Engines: [list]           Prompt set: [name / size, fixed or changed]

1. Brand visibility (absolute)
   Analysed answers:            [n]
   Answers naming us:           [n]
   Brand visibility:            [x%]     Previous period: [x%]     Change: [+/- pts]
   Provisional (under 30)?      [yes / no]

2. Share of voice (relative)
   Roster this period:          [list of active tracked competitors]
   Roster last period:          [list]
   Roster changes:              [added / paused / removed / alias or sub-brand added, with dates, or "none"]
   Our answers:                 [n]
   Sum across active families:  [n]
   Share of voice:              [x%]     Previous period: [x%]     Change: [+/- pts]

3. Whose count moved
   Our count:                   [prev -> now]
   Competitor A:                [prev -> now]
   Competitor B:                [prev -> now]

4. Reading
   [One sentence per metric. State if the share moved because of our count, a rival's count, or the roster.]
   Claim strength: [observed / suggestive / not established]

5. Caveats
   Untracked brands noticed but not on the roster: [names or "none noticed"]
   Analysed vs collected answers this period:      [n vs n]

Three habits make this hold up:

• Recompute the previous period on the current roster when you want a like-for-like share. Otherwise the change mixes a real movement with a roster edit.
• Keep a roster change log with dates. Shares are calculated from the current active roster, so a paused competitor's recorded mentions stop counting on your next load. Write down what the number was when you reported it.
• Never show share of voice alone in a headline. A headline number without its count invites the reader to assume you are being named more.

Do not confuse this with the other share of voice on the Competitors screen

The Competitors screen shows a second figure with the same name. The hero figure and bar chart there size each domain by how many times it was cited, not how many answers named its brand. Domain-citation share of voice is one domain's citation count over the citation counts of every active competitor row plus your own. It is relative to your roster in the same way, but it is a citation measure, not a mention measure.

So a report that quotes "share of voice" needs to say which one. In this post, share of voice means the mention-based figure, which is the one the "By brand mentioned" list on Competitors uses. For related reading on mentions and citations, see AI visibility, brand mentions, and citations: what each number tells you and Why two teams can calculate different citation rates from the same answers.

Sub-brands and the once-per-answer rule

If you track a parent brand and its products, share of voice counts the family once per answer, so an answer naming both a brand and its own sub-brand is one answer for that family. Without that rule, an answer naming a company and its product would count twice. Sub-brand rows on Competitors overlap their parent rather than adding up to it. For more, see Track a parent brand and its products without double-counting mentions and the brands and sub-brands page.

Common mistakes and what this cannot tell you

• Reading a rising share as a rising reputation. A ratio going up says your slice of the tracked naming grew. It does not say engines think better of you.
• Comparing shares across roster changes. A share computed on three competitors and one computed on five are different numbers.
• Comparing "observations" across metrics. The metrics page warns that observation counts on brand visibility, brand position and share of voice are not the same denominator. Share of voice sums across families, so one answer naming two families is counted twice there.
• Assuming a stated cause. Neither metric tells you why an engine produced an answer. Both are observations of what answers contained. Treat any explanation as a hypothesis to test.
• Ignoring untracked names. A market leader missing from your roster makes your share look better than it is.
• Reporting small samples. Below 30 observations a metric is provisional.
• Assuming a no-data period is zero. A metric with no data reads as no data, not 0%.

Frequently asked questions

Can share of voice go up while brand visibility goes down?

Yes. If your mentions fell but your rivals' fell faster, your share of the tracked total rises while your visibility falls. The pairing "Down / Up" in the table above is exactly this case, and it is a reason to read both.

Is share of voice out of 100% across all brands in AI answers?

No. It is out of 100% across the active brand families on your roster only. Brands not tracked are excluded, and paused competitors are left out. Because one answer can name several families, the sum of family counts can also exceed the number of answers.

Which metric should I put in an executive summary?

Lead with brand visibility, because it does not shift when you edit your competitor list, and show share of voice next to it with the roster stated. If the two disagree, explain which moved and why in one sentence.

Does pausing a competitor delete their data?

No. The mentions already recorded are kept. They are just not counted while the competitor is paused, and future answers naming a paused competitor are recorded unresolved rather than credited to it. That is why the roster change log matters.

How do I tell whether a rival dropped or the roster changed?

Compare each competitor's answer count for both periods on the same roster, and check your roster change log for the dates. The "By brand mentioned" list on Competitors shows one row per tracked brand family, and the Explorer can break brand metrics down by brand. If a family left the roster, its count did not necessarily fall; it stopped being counted.

Next step

Both metrics are defined in the metrics reference, and Overview shows them for your own brand. The Competitors screen is where you manage the roster that sets the denominator. To see which brand names are appearing in your answers without being tracked, open the "Brands appearing in your answers" list and decide deliberately who belongs in the comparison set. Sign in to DiscoveredBy to check both numbers for your own project.