What is AI share of voice, and how do you calculate it?

This guide describes a measurement method. Automated AI prompt, mention and citation monitoring is not available in the current public BeKnow product. Collect the observations with an external service or a documented manual sample.

AI share of voice measures a brand's relative presence among a defined group of competitors in observed AI answers. It does not measure the whole market or every model conversation. It describes one sample: selected prompts, models, languages, markets, dates and counting rules.

A simple mention-based formula is:

Brand mentions ÷ mentions of all monitored brands × 100

If your brand receives 30 qualifying mentions and the complete competitor set receives 120, its share of voice is 25%. The calculation is easy; designing a fair denominator is the difficult part.

Decide what one “voice” means

Before calculating, choose a unit:

• answer presence: each brand counts at most once per answer;
• mention frequency: every qualifying occurrence counts;
• prominent presence: only recommendations or defined top positions count;
• citation share: the unit is an owned or attributed source, not a brand name.

Do not switch definitions between periods. Answer presence is usually the clearest baseline because repeated use of a name in one response cannot dominate the score. Citation share should be reported separately from mention share. Read mention rate versus citation rate before combining metrics.

Build the comparison set

List direct competitors that can realistically answer the same buyer needs. Freeze the set for the comparison period. Adding a weak brand next month can inflate everyone's historical-looking share even though no answer changed.

Keep an “other brands” bucket for unexpected names, then review whether recurring entrants belong in the monitored set. Do not silently exclude a new competitor because it makes your number smaller.

Use a balanced prompt sample

The denominator is also shaped by prompts. A sample full of your brand name guarantees biased presence. Include unbranded category, problem, comparison, alternative and trust questions. Use branded prompts only for accuracy diagnostics and report them separately.

Keep languages and markets distinct. A Spanish sample for Spain may have a different competitive set from Mexico. English results should not be used as a proxy for Italian demand. The prompt-selection guide provides a repeatable structure.

Calculate AI share of voice step by step

1. Select valid responses only; keep provider errors outside the denominator.
2. Detect monitored brands using names, domains and approved aliases.
3. Review ambiguous matches and false positives.
4. Apply the chosen unit once to every brand.
5. Sum all qualifying brand units.
6. Divide each brand's count by the total and multiply by 100.
7. Preserve raw counts beside the percentage.

When an answer contains three monitored brands, answer-presence scoring gives one unit to each. The total share still sums to 100% across brand units, not across answers.

Segment before interpreting

An overall figure can hide the useful pattern. Break the result down by:

Segment	Question it answers
Intent	Are we absent from discovery, comparison or implementation?
Model	Is the gap isolated to ChatGPT, Gemini, Claude or Perplexity?
Market and language	Where does local presence differ?
Mention role	Are we recommended or merely listed?
Source type	Do owned pages or third parties support the answer?

Avoid false progress

Compare the stable cohort first. If you add prompts or competitors, calculate the old cohort and the expanded cohort separately. Keep model and retrieval settings compatible. Display sample size and valid-run count.

A rise from one of four brand units to two of four is a large percentage change but weak evidence. Repeated observations and larger balanced samples improve stability without making the result universal.

Turn a competitor lead into a diagnosis

Open the answers and cited sources behind the leading competitor. Group the gap by intent. A competitor may win because it has a clear definition, independent reviews, current documentation or a page format your site lacks. It may also be named from model memory without a cited page.

Do not copy the winner. Decide whether the useful action is to update an existing page, strengthen product facts, obtain independent corroboration or create a genuinely missing asset. Use the AI-cited source analysis for evidence.

Frequently asked questions

Is AI share of voice the same as mention rate?

No. Mention rate measures how often your brand appears in valid answers. Share of voice compares your qualifying presence with the presence of a defined competitor set.

Must all shares add up to 100%?

They do when every qualifying brand unit is included in the denominator. A dashboard using another denominator must state its formula.

Can I combine models?

Keep model-level results first. A combined number needs transparent weights so a larger or cheaper model sample does not dominate silently.

Does a higher share prove more sales?

No. It is a visibility diagnostic. Connect referral traffic, conversions and pipeline separately.

Make the denominator inspectable

AI share of voice becomes useful when the competitor set, prompts, valid responses and formula remain visible. The goal is not a flattering percentage; it is knowing where another brand occupies a decision your content should genuinely support.

What BeKnow keeps

Keep interventions, hypotheses, assets, observations and decisions in a searchable workspace. Record what is known and what still needs checking.

A change in performance after an intervention does not prove that the intervention caused it. Cross-platform attribution and a complete analytics dashboard are not available today.

Project memory and data imports do not require an AI model key. A compatible external AI client may have its own costs. BYOK applies only to available functions that actually call an external provider.

Next step

Start with one project, one documented change and the evidence needed to review it. Source connections.

Record your first intervention. How it works.