Share of voice is the percentage of conversation in your category that belongs to you rather than your competitors. It's one of the few PR metrics that survives contact with a finance team, because it's comparative. It tells you whether you're gaining ground, not just whether you were busy.
It is also one of the easiest metrics to game, usually by accident, usually by the person reporting it.
The definition
Share of voice (SOV) = your mentions ÷ total mentions across you and a defined competitive set, over a defined period, in a defined source universe.
Every one of those "defined" qualifiers is load-bearing. Change the competitor list, the date range, or which publications you count, and the number moves several points without anything happening in the real world. A share of voice figure without its parameters attached is not a measurement, it's a mood.
A worked version: over Q3, in a set of 14 trade and business publications, your company appeared in 22 articles. Your four named competitors appeared in 31, 44, 12, and 9. Total is 118. Your SOV is 22 ÷ 118, or about 19%. The useful sentence is the comparison: "up from 14% in Q2, driven by the funding announcement and two bylines."
Why raw mention counts mislead
The failure mode is counting every mention as one unit. Four problems with that:
Not all mentions are the same size. Being the subject of a feature and being named in a list of eight vendors are both "one mention." A weighted count (full feature, substantial mention, passing reference) is more work and dramatically more honest. Even a crude three-tier weighting (3 / 2 / 1) changes the picture for most companies.
Not all outlets reach your buyer. A hundred syndicated republications of one press release can swamp a single article in the one trade publication your customers actually read. If your source universe includes aggregators and content farms, your SOV measures syndication mechanics, not attention.
Sentiment isn't captured. A company in the middle of a regulatory problem often has excellent share of voice. Volume and favorability are separate axes and should be reported separately. Automated sentiment scoring is still unreliable on industry jargon, sarcasm, and neutral-but-unflattering reporting, so treat it as a rough sort, not a number.
Brand-name ambiguity contaminates everything. If your company shares a name with a common word, a band, or a much larger company, your raw counts are junk until you build exclusion rules. Check a sample of 50 matched articles by hand before trusting any tool's total. This single step catches more measurement errors than anything else on this list.
How to actually calculate it
- Define the competitive set, narrowly. Three to six companies your buyers genuinely evaluate against you. Not the whole industry, and not aspirational giants whose coverage volume will permanently flatten your line.
- Define the source universe. The specific publications, trades, and podcasts that matter to your buyer. Twenty relevant outlets beat forty thousand indiscriminate ones. Write the list down so next quarter's number is comparable.
- Define the query. Company name plus known variants and misspellings, minus the exclusions you found in your sample check.
- Pick a window and keep it. Monthly for fast-moving categories, quarterly for most B2B. Changing the window mid-year makes the trend meaningless.
- Weight by prominence. At minimum, separate "article is about us" from "we're named in passing."
- Report alongside sentiment and reach, not merged into one score. Composite scores hide the thing you needed to see.
Tooling, honestly
- Media monitoring platforms (Meltwater, Cision, Brandwatch, Muck Rack and similar) do the collection and give you competitor dashboards out of the box. They're the right answer if PR is a real budget line. Expect meaningful annual contracts, and expect to spend a week tuning queries before the data is usable.
- Mid-market tools (Brand24, Mention, and peers) cost a fraction and cover most of what a small team needs. Less complete on print and paywalled trade press.
- Free scaffolding. Google Alerts plus a spreadsheet plus a quarterly hour of manual counting is a legitimate starting point, and it produces a more honest number than an untuned enterprise tool. Many small companies should start here and upgrade only when the counting time exceeds the license cost.
No tool covers everything. Podcasts, newsletters, closed communities, and paywalled trades are all systematically under-captured. Assume your measured universe is a sample, not a census, and stop treating small movements as signal.
Setting a benchmark you can defend
Your first number has no meaning on its own. It becomes meaningful in three ways:
- Against yourself over time. The trend line is the point. Two quarters of data is the minimum before you draw any conclusion.
- Against a fair share baseline. If there are five real players, an even split is 20%. Being at 8% with 20% market share tells you coverage lags commercial position, a genuinely actionable gap.
- Against effort. If you doubled PR activity and SOV moved two points, that's information about your approach, not just about the market.
Set a target that's a change, not a level: "from 19% to 25% over two quarters, primarily through bylined articles in three named trades." A target expressed as a number with a mechanism attached is one you can actually work toward.
Where the metric quietly breaks
Share of voice measures published coverage. Increasingly, a chunk of how buyers encounter your category never appears in a publication at all. It happens inside an AI assistant answering "what are the best tools for X." Whether you get named there is a related but distinct question, and it isn't captured by any media-monitoring query.
The honest position: this is early, the measurement methods are immature, and nobody has a reliable model for how these systems choose which companies to name. What you can do is check it directly. Run your category's buying questions through the major assistants on a schedule and record who gets mentioned, which is a crude but real second share-of-voice reading. There's a fuller treatment in our guide to tracking brand mentions in AI, and the strategic framing in generative engine optimization.
The one-paragraph version
Pick four real competitors and twenty real publications. Count quarterly. Weight prominence crudely. Check a sample by hand. Report volume and sentiment separately. Compare to your own last quarter, not to an industry giant. Then use the gap to decide what to pitch next, because the number's only job is to change what you do.
Coverage volume starts with having something to announce. [See the pricing](/pricing).