What Is Share of Voice?

Share of voice is a brand's proportion of a defined competitive measure, such as counted mentions, within a specified source set and period. Calculate it as the brand's units divided by all included brands' units, multiplied by 100. The unit and denominator matter: mention share, publishing activity and AI-answer presence measure different things.

Dupple publishes newsletters and sells advertising and sponsored placements, including through Toolradar. We benefit commercially from visibility. Our archived AI-answer counts below describe our own small research panel; they are not an independent market audit or proof that buying a placement improves recommendations.

Define the measure before calculating the share

The phrase share of voice does not specify a complete measurement method. You must choose what counts, which competitors are included, which sources can be observed, and how duplicates are handled. A carefully calculated fraction can still mislead if its label implies coverage of an entire market that the collection method never observed.

Table: metrics sometimes grouped under visibility, with different denominators.

As read on 6 November 2026, Brandwatch's SOV glossary describes a ratio whose inputs vary by channel. Its separate Benchmark FAQ defines that product's SOV using brands' publishing activity, rather than mentions or engagement. A dashboard label therefore does not establish that two reports count the same thing.

For a media report, record whether a syndicated story appearing on several sites counts once or several times. For an answer panel, record whether repeating a brand within one response creates additional units. Preserve the rule with the export. Changing it later can alter the share even when the underlying content stays identical.

A fictional competitive calculation

Suppose a fictional media review identifies 100 included brand-story pairs in a week: 48 for Brand A, 32 for Brand B and 20 for Brand C. Under that declared rule, Brand A has 48% of the selected brand-story pairs. The result describes these three brands in these sources for that week, not all attention in the industry.

An article naming both A and B contributes two pairs in this example. There may therefore be fewer than 100 unique articles. If the report instead divides Brand A's included articles by all unique articles, it becomes a different metric. Both can be useful, but they should not share a label without explaining the denominator.

Adding Brand D to the competitive set can reduce A's share without removing a single A mention. Likewise, excluding duplicated press-release copies can change the total. Keep absolute counts alongside percentages so a reader can distinguish a brand losing mentions from the denominator expanding. If there are no eligible units, the share is undefined rather than automatically zero.

Dupple data: recounting a dated AI-answer panel

Our archived services-visibility check from 4 November 2026 contains 35 answers, one per selected question. Saved annotations mark a Dupple-family brand as present in 25 answers. We checked these annotations on 6 November 2026. The public aggregate and provenance file records the source hash, original prompt-file commit, cohort definitions and limits.

The documented run used fresh Claude Sonnet web-search answers without project context; it restricted searching for a Dupple brand unless the question named one. The archive does not separately record an exact model version. These are saved model annotations, not an independent re-audit of every factual claim, and this recount did not run a new experiment.

Question selection matters. Seven of the original questions explicitly contain a Dupple-family brand name. A case-insensitive whole-word filter for Dupple, Techpresso, AIpresso, Devshot, Cyberpresso, Finpresso, MarketingShot or Toolradar leaves 28 questions without those names, with brand presence annotated in 18 answers. The seven named-brand questions all have presence annotated.

Table: observed counts from the 4 November services panel, recounted 6 November 2026. Percentages are deliberately omitted for this small selected sample.

The original report also used a category called brand, containing six questions. Excluding that category gives 29 answers and 19 brand-present answers. It is a different filter: another question, ID 42, explicitly names Techpresso while belonging to the newsletter category. Category exclusion and literal brand-name exclusion should not be described as the same unbranded cohort.

The historical prompt file was checked at its original commit before applying the literal-name filter. Later questions added to the research set are excluded here because they have no answer in this baseline. This makes the count reproducible without presenting a larger current prompt list as if every question had been tested on 4 November.

Why those counts are not AI market share

The 25 brand-present answers are a presence count across questions, not Dupple's share of all competitor mentions. Several brands can appear in one response. To calculate mention share across brands, we would need a consistently defined and complete set of included brand-answer pairs. The presence denominator of 35 answers does not supply that competitive total.

Brand presence also differs from citation, recommendation and favorable evaluation. All 25 brand-present answers were annotated as treating our figures as self-published or unverified. A system can mention a publisher while warning that its numbers lack independent confirmation. Counting the name alone would conceal that qualification.

This is one model workflow, one pass per question and a deliberately selected small panel. Re-running can produce different sources and wording. It does not represent every assistant, every language or every buyer query. No comparison in this article measures a post-publication lift, and no observed count establishes that a sponsored page caused additional business.

Keep answer visibility separate from referral traffic

An assistant may include a brand without sending a visit. A reader may also visit later without a referrer that identifies the assistant. Our AI-referral traffic definition distinguishes observed referrals from self-reported discovery. These are related signals with different collection methods, not interchangeable shares of one funnel.

An AEO program can monitor answer retrieval and brand treatment while separately recording inquiries. For purchases of content or listings, our sponsored-listing visibility guide explains why publication alone cannot guarantee a citation or recommendation. An accepted publication is a deliverable; a measured change in answers needs its own evidence.

Connect inquiries to campaigns only under the stated attribution window and available identity data. A report can say an inquiry named an assistant as a discovery source. It cannot automatically say which specific answer, article or placement caused the inquiry if that connection was never observed.

Make the next comparison reproducible

Freeze a stable question set and preserve its exact text, language and category labels. Record the model or service, run time, browsing access and repetition count. Save the response before coding it. Apply the same brand and citation rules to both dates, and keep named-brand questions separate from generic discovery questions.

When adding new questions, report the old cohort and the additions separately. Otherwise, an apparent improvement can come from easier new questions rather than changed visibility. Repeated observations can show variability; they still do not turn a selected panel into a representative market sample. Include absolute counts and note any missing or failed answers.

For buyer-facing claims in a media kit, disclose whether the metric comes from first-party research, a platform report or an independent study. Our first-party data explanation covers the limitations of directly collected evidence. The public audience-data methodology serves a different purpose from this answer panel. Use the advertising glossary to keep exposure, interaction and outcome labels distinct when comparing advertising formats.

FAQ