Your share-of-voice percentage includes the brand mentions you counted within a named competitor set, source set, query definition and period. It excludes everything outside those boundaries. A result of 44.4% can describe your share of reviewed mentions across three brands in a collected dataset. It cannot establish your share of the whole market's conversation.
The social share of voice calculation is:
Share of voice = eligible mentions of your brand / eligible mentions of all included brands × 100.
Sprout Social's explanation uses the same brand-to-total structure and distinguishes conversation share from sales-based market share. The reporting task is to explain what went into that total.
Put the denominator beside the percentage
Before collecting data, write a measurement specification. The following is a hypothetical specification for three fictional brands. These are recommended reporting choices, not platform rules.
| Decision | Hypothetical specification |
|---|---|
| Competitors | Brand A, Brand B and Brand C; no category-wide denominator |
| Period | September 1 through September 7, 2026, UTC; end boundary September 8 at 00:00, excluded |
| Sources | Collected X posts and YouTube video records; source results also reported separately |
| Search fields | Brand terms returned by each search method, then reviewed in X post text or YouTube title and description |
| Query version | Saved brand names, aliases, handles and exclusion terms, version 1 |
| Language | English text confirmed during review; no geographic inference |
| Counting unit | One eligible post-brand pair, counted once per brand per platform ID |
| Exclusions | Brand-owned accounts, reposts, irrelevant matches and duplicate IDs |
| Sentiment | All eligible mentions, regardless of sentiment |
A post mentioning both A and B contributes one pair to each brand. Repeating A five times in one post still contributes one A pair. The denominator therefore counts brand references across posts, rather than unique posts or people.
Other counting rules can work. Fractional allocation could split a multi-brand post between brands. Choose one rule before comparing results and preserve it across periods. For record-level rules, use the guide to removing duplicate posts from a listening sample.
Declare what each source can supply
Naming two platforms does not describe two complete collections.
X's Search Posts documentation distinguishes recent search, covering the last seven days, from full-archive search. It lists full-archive access for pay-per-use and Enterprise customers. It also lists developer access and app credentials as prerequisites. Record which endpoint and access you used. A historical report needs a collection method that can reach its stated period.
YouTube's search documentation says results can identify videos, channels or playlists. Specify the resource type. Its pageInfo.totalResults value is approximate, so do not insert that number into an exact mention-count denominator. Count the records you retrieved, deduplicated and reviewed. Follow pagination and record where collection stopped.
YouTube also says relevanceLanguage can return other languages. Its regionCode describes where videos can be viewed. Neither setting establishes that every author belongs to a language group or lives in a particular country.
For the hypothetical specification above, private messages, comments, spoken references inside videos and image-only logos are outside the count. So are posts the collection never returned. These are boundaries of this measurement design; they do not describe every possible listening method.
Document missing dates, failed requests and capped exports. If coverage differs between brands, resolve the difference or label the comparison incomplete. The coverage-gap reporting guide helps turn those limits into a report readers can interpret.
Calculate the reviewed baseline
All numbers below are hypothetical. Suppose the declared searches return 360 candidate post-brand pairs. Review every candidate using the same relevance rule: the text must refer to the intended brand. A reviewer rejects an unrelated use of a similar name.
| Brand | Candidate pairs | Relevant pairs | Share of relevant pairs |
|---|---|---|---|
| A | 150 | 120 | 44.4% |
| B | 120 | 90 | 33.3% |
| C | 90 | 60 | 22.2% |
| Total | 360 | 270 | 100% before rounding |
Brand A's calculation is 120 / 270 × 100 = 44.4%, rounded to one decimal place. The rejected 90 pairs do not belong in the denominator. Small rounding differences explain why the displayed brand percentages sum to 99.9%.
Report the count and percentage together. With no eligible pairs for any included brand, the percentage is undefined. Label that period as having no eligible mentions rather than reporting 0% share.
Pooling sources also creates a choice. Summing counts gives each eligible pair equal weight, so a source with more collected mentions contributes more to the result. Averaging platform percentages gives platforms equal weight instead. Name the method and show source-level results before presenting a pooled number.
Test how the query changes the answer
A stricter query may remove false matches and genuine mentions together. Test that tradeoff before accepting a cleaner-looking percentage.
In this hypothetical test, apply an additional category-word requirement to the same saved broad collection for all three brands. Every narrower result is a member of the broader set. Both sets are fully reviewed; they are not separate samples or new live searches.
| Brand | Narrow candidates | Narrow relevant pairs | Relevant pairs lost |
|---|---|---|---|
| A | 110 | 100 | 20 |
| B | 90 | 80 | 10 |
| C | 60 | 50 | 10 |
| Total | 260 | 230 | 40 |
The retrieval evaluation method in Introduction to Information Retrieval defines precision as the proportion of retrieved items that are relevant. Applied here:
- Broad-query precision is 270 / 360 = 75.0%.
- Narrow-query precision is 230 / 260 = 88.5%, rounded.
- The narrower query removes 100 candidates, including 60 irrelevant pairs and 40 relevant pairs.
- Brand A's narrower share is 100 / 230 = 43.5%, rounded.
The filter gains about 13.5 percentage points of precision while losing 40 of the 270 known relevant pairs, or 14.8%. Brand A's share falls by about 1.0 percentage point, calculated from unrounded values, solely because the definition changed.
For a report about all brand references, the added category word may exclude valid references unnecessarily. For a report specifically about category-word conversations, that restriction may fit the question. Decide which question you mean before choosing the query. The guide to excluding irrelevant brand-query matches covers query refinement.
This test measures sensitivity within the collected data. It does not reveal relevant posts that the broad search missed. Do not label the retained fraction as recall across the platform, or treat the collection as a representative sample of consumers.
Keep definition changes out of the trend line
A competitor change alone can move the percentage. In the hypothetical baseline, dropping C raises A's share to 120 / 210 = 57.1%. A has gained no mentions.
When adding a competitor, changing an alias or replacing a source, either recalculate earlier periods under the new definition or mark a break in the series. Keep the earlier definition available. A missing source should remain a coverage gap, rather than silently becoming zero mentions.
Attach a sentence like this to the hypothetical result:
Brand A accounted for 120 of 270 reviewed eligible post-brand pairs, or 44.4%, among A, B and C in the declared X and YouTube collection for September 1-7, 2026, UTC. The count excludes brand-owned posts, reposts, irrelevant matches and duplicate IDs. It does not estimate market-wide conversation or sales share.
Before sending the next report, ask a colleague to reconstruct its numerator and denominator from the saved specification. If they cannot, add the missing rule before sharing the percentage.
