Calculate AI share of voice

Choose the brands and the counting unit

For this guide, an appearance means one tracked brand is present in one valid answer. A brand can appear at most once per answer for this calculation.

The comparison set is Northstar, Example B and Example C, all fictional. If an answer contains Northstar and Example B, it contributes two appearances to the set. An answer that names none of the three contributes zero appearances while remaining a valid answer for the mention-rate denominator.

Use the measurement worksheet if you need the complete twelve-row ledger.

Add appearances before dividing

The constructed sample contains ten valid answers out of twelve planned runs. Across those ten answers:

Northstar SOV = 6 / (6 + 4 + 2) = 6 / 12 = 50%

The denominator is twelve brand appearances, not twelve planned runs. The matching numbers are a coincidence in this worksheet.

Why the mention rate is 60% instead

Northstar appears in six of the ten valid answers:

Northstar mention rate = 6 / 10 = 60%
Northstar sampled SOV  = 6 / 12 appearances = 50%

Mention rate answers, “How often did our target appear in usable answers?” SOV answers, “What portion of this tracked set’s appearances belonged to our target?”

Q01 contains Northstar and Example B. It adds one valid answer to the mention denominator and two appearances to the SOV denominator. Counting the paragraph’s repeated names would create a different metric, so do not mix that method into this one.

Put citation results in their own column

The same sample has eight answers with complete citation evidence. Three explicitly cite an approved owned URL, so the owned citation rate is 3/8 = 37.5%.

That figure neither replaces 50% brand SOV nor uses the ten-answer mention denominator. Q09 and Q10 have readable answers but incomplete citation lists; assigning them zero owned citations would conceal missing evidence.

A source-domain share would require its own counting unit, such as explicit citations per domain. Define that separately before comparing it with the brand-appearance figures. See the classification examples.

Check the denominator when a share moves

A change in relative share can come from the target, the competitors or the comparison design. Inspect the underlying counts before assigning work:

The baseline has 12 planned runs, 10 valid answers and 12 appearances. Preserve those units in the report even when two totals happen to be equal.

Follow one changed question back to its answer

For the next review, start with the questions where a brand’s classification changed. Open the saved answer, confirm the identity match and inspect any explicit sources. Then decide whether the difference points to a factual error, a missing explanation or ordinary sample variation.

Use this report sentence for the constructed example: “Northstar accounted for 6 of 12 tracked-brand appearances, giving it 50% sampled share of voice; it appeared in 6 of 10 valid answers, a 60% mention rate. Collection coverage was 10 of 12 planned runs.”

Our counting methodology and answer-level workflow make the denominator inspectable. The calculation describes the selected sample and comparison set; revenue and market-wide demand require different evidence.