Your brand receives the same 120 mentions in two periods, yet its reported share falls from 20% to 15%. If the monitoring tool added competitors or sources, that decline may describe a changed denominator rather than weaker visibility.
Measuring share of voice begins before the division. The difficult work is choosing what belongs in the dataset and collecting it consistently enough that a later comparison means something.
Write the measurement specification
Define the brand, aliases, competitor set, topic, language, geography, channels, observation period, and unit being counted. Decide whether a mention in a repost is a new event and how duplicate coverage will be handled.
Record excluded cases, including unrelated uses of a brand name. A short name that is also a common word can create a large false-positive problem. Review ambiguous matches manually before trusting the total.
Use one stable core competitor set for trends. Track exploratory competitors separately until you deliberately update the baseline.
Count comparable events
For mention-based competitive share, divide the brand's qualifying mentions by the total qualifying mentions across the defined brand set. Use the same matching, deduplication, and source rules for every brand.
In the fictional example, Northstar has 120 mentions, Beacon 210, Atlas 150, and Cedar 120. The total is 600, so Northstar's share is 20%.
Illustrative report with fictional data and workflow examples. This is not a customer result or a preview of a specific product screen.
If the set expands to 800 mentions while Northstar remains at 120, its share becomes 15%. The report should identify the scope change rather than presenting a clean five-point performance decline.
Keep quality alongside volume
Review relevance and sentiment where they matter to the decision. A burst of negative coverage can increase share while creating a problem for the business. Preserve the source text so an automated label can be checked.
Inspect a sample of included and excluded records. Check duplicated articles, syndication, spam, brand-name collisions, and missing sources. Record collection failures separately from a genuine absence of mentions.
When source coverage changes, annotate the report and consider maintaining an old-source comparison alongside the expanded dataset.
Handle each channel deliberately
Organic visibility estimates depend on a keyword set and weighting model. Paid impression share uses estimated eligible impressions, as explained in Google Ads documentation. Neither should be silently substituted for share of all competitive mentions.
For AI answers, distinguish response appearance rate from citation-event share. An answer can contain several brands, so response rates can overlap. If you want an additive competitive share, count comparable events across the selected brands and name the metric accordingly.
The calculation guide gives formulas for these distinctions.
Preserve enough detail to reproduce the result
Store the raw records, matching rules, normalized brand names, source dates, and exclusion reasons. Keep the calculation separate from the presentation layer so a revised chart does not overwrite the evidence.
Have a second person recompute a small sample where practical. A simple denominator or deduplication error can affect every subsequent decision while leaving the dashboard visually convincing.
Report the change and its explanation separately
Show current share, prior share, percentage-point change, event counts, and scope notes. Add the hypothesis behind the movement only after inspecting the underlying records.
A report should end with a decision: maintain the program, investigate a new source, improve a particular page, or revise the sample. Use the SOV report template for that handoff and the calculator for a small, transparent starting dataset.
