What Is AI Share of Voice and How Do You Measure It?

AI share of voice sounds like one simple percentage.

It is not useful until you know which prompts, providers, competitors, dates and counting rule created it.

One tool may report the percentage of answers mentioning your brand. Another may report your share of all competitor mentions. Both can be informative, but they are different metrics with different denominators.

Prerender Buddy separates them.

Quick answer

AI mention rate measures how often your brand appears across valid answers:

Mention rate = valid answers mentioning your brand ÷ all valid answers × 100

AI share of voice measures your share of observed brand appearances within a defined competitive set:

AI share of voice = your brand appearances ÷ appearances of your brand and tracked competitors × 100

Calculate both from the same controlled prompt set and reporting window. Report by prompt group and provider before showing an overall number.

AI share of voice describes the measured sample. It is not universal market share and does not guarantee traffic, recommendations or sales.

Why the definition matters

Suppose you monitor 20 valid answers.

Your brand appears in 8. Competitor A appears in 12. Competitor B appears in 10. Several answers mention more than one brand.

Your mention rate is:

8 ÷ 20 × 100 = 40%

Your competitive AI share of voice is:

8 ÷ (8 + 12 + 10) × 100 = 26.7%

Forty percent and 26.7% describe the same observation set. One measures coverage across answers; the other measures relative competitive presence.

If a report says only “AI SOV: 40%,” the client cannot tell which method was used.

Define the measurement universe

Before calculating anything, record:

• prompt list and version
• prompt intent groups
• branded or non-branded status
• providers and modes
• market and language
• competitor set
• run frequency
• reporting window
• valid-run criteria
• name aliases and entity rules
• counting method

The score changes when this universe changes.

Adding a competitor increases the denominator. Adding broad informational prompts can reduce mention coverage. Combining a new language can change the answer set. Annotate these method changes rather than describing them as brand performance.

Choose prompts that represent real decisions

AI share of voice is only as meaningful as the prompts behind it.

Use questions grounded in:

• sales and support conversations
• Search Console and paid-search language
• product capabilities and use cases
• customer objections
• comparisons and alternatives
• active markets and languages

Avoid a prompt set constructed to make the brand look strong.

For Prerender Buddy, a balanced non-branded set might include:

• crawler-readability diagnosis
• prerendering decision
• broken or stale page monitoring
• AI brand-mention tracking
• competitor and citation analysis
• improvement prioritization

Use How to Choose Prompts for AI Visibility Tracking for the complete selection method.

Define the competitor set honestly

Include brands that compete for the same user decision, not only companies the client prefers to compare against.

Sources include:

• competitors named by customers
• brands appearing repeatedly in monitored answers
• direct product alternatives
• adjacent solutions that genuinely satisfy the prompt
• regional competitors for localized groups

Keep the set stable during a comparison period. If an untracked brand appears frequently, preserve it as “other observed brand” and review whether it belongs in the next version.

Do not remove a strong competitor merely to improve the client's percentage.

Decide what counts as an appearance

Use a clear entity policy.

Recommended default:

• count a recognized brand once per valid answer
• normalize approved spelling and product aliases
• do not count unrelated words matching a short brand name
• treat parent company and product separately when the client decision distinguishes them
• preserve the surrounding context
• keep negative and incidental mentions visible

Counting each repeated use of a name within one answer can overvalue verbose answers. Counting once per brand per answer produces a cleaner presence measure.

Mention share, recommendation share and citation share

Do not force different visibility events into one number.

Mention share

Your share of all tracked brand appearances.

Recommendation share

Your share of answers or recommendation events in which tracked brands are presented as suitable options.

Owned citation share

Your owned-domain citations relative to owned-domain citations for the tracked competitive set.

External support share

External sources associated with each tracked brand.

A brand can lead in mentions and trail in recommendations or citations. Report the underlying events separately.

Read AI Mentions vs Citations vs Recommendations for the classification rules.

Calculate per prompt group first

Suppose a brand has:

• 45% share in technical diagnosis
• 30% share in small-agency use cases
• 5% share in AI visibility measurement
• 60% share in branded comparison prompts

An overall average can hide the weak commercial area.

Report by:

• intent group
• product segment
• buyer stage
• market and language
• provider

Then show an overall result only if the weighting is disclosed.

Do not mix branded and non-branded prompts

Branded prompts almost require the client name to appear.

Including them in the same share-of-voice number can inflate the result without showing non-branded discovery.

Use separate views:

• branded accuracy and support
• non-branded category discovery
• competitor and alternatives prompts

This also makes recommendations clearer. A branded accuracy problem requires a different action from weak non-branded presence.

Weight prompts only with a defensible reason

The simplest model gives every valid answer equal weight.

Weighted models may reflect:

• validated business value
• buyer stage
• reliable demand evidence
• strategic market priority

If prompts are weighted, disclose each weight and keep the unweighted result available.

Do not invent precise prompt volumes and present them as observed user demand. If volume data is estimated or modeled, label it.

Handle provider differences

Calculate provider-level share before combining providers.

A combined result can be calculated from total counts, but it should not hide where the difference comes from.

Record modes and whether web search or citations were observed. Do not treat a cited search answer as identical to a non-search response.

Repeat measurements consistently

Generated answers can vary across repeated runs and small prompt changes.

For trend reporting:

• keep reference prompts stable
• use defined clean-session conditions
• repeat at a consistent cadence
• preserve every valid answer
• separate failed runs
• report counts with percentages
• compare equivalent windows
• annotate product, content and method changes

Do not regenerate until the client appears and discard the other answers.

What is a good AI share of voice?

There is no universal good percentage.

The result depends on:

• number and strength of competitors
• prompt breadth and intent
• brand maturity
• market and language
• provider mix
• run count
• counting and weighting method

A 20% share in a five-brand competitive set can mean something different from 20% in a category where 30 brands appear.

Use the brand's controlled historical trend and the defined competitive set as the primary comparison. Treat external benchmarks cautiously unless their methodology matches yours.

Interpret common patterns

Mention rate increases, share of voice stays flat

The category may be naming more brands overall. Your coverage improved, but competitors improved proportionally.

Share of voice increases, mention rate stays flat

Competitor appearances may have declined while your own coverage remained stable.

Mention share is high, recommendation share is low

The brand is present but not strongly associated with fit. Review the context, product positioning and evidence.

Citation share is high, mention share is low

Owned content may support general answers without the brand being clearly named or recommended.

Overall share rises while a priority group falls

The aggregate hides a commercially important weakness. Prioritize by group, not the flattering total.

A transparent report block

1## AI share of voice — [period]
2 
3- Definition: competitive share of brand appearances
4- Formula: client appearances / all tracked brand appearances
5- Prompt version:
6- Providers and modes:
7- Market and language:
8- Valid answers:
9- Tracked competitors:
10- Counting rule: max one appearance per brand per answer
11 
12### Results
13- Overall AI SOV:
14- Non-branded AI SOV:
15- Mention rate:
16- Recommendation share:
17- Owned citation share:
18- Strongest prompt group:
19- Largest gap:
20 
21### Method changes and limitations
22- [List changes, sampling limits and failed runs]

The block makes the result auditable without overwhelming the executive summary.

Connect share of voice to page health

Before recommending content work for a weak group, check the mapped owned page:

• is it accessible?
• does the crawler receive important content?
• is it current?
• does it clearly serve the prompt?
• does it provide supportable evidence?

A zero share cannot diagnose the cause by itself. Use technical health, competitor context and sources to classify the gap.

Who this is for

• SaaS founders with already-shipped JavaScript websites
• React, Vite, Vue, Lovable, Bolt, or Base44 users
• SEO freelancers checking crawler-readable HTML
• Agencies maintaining client sites without rebuilding them

Where Prerender Buddy fits

Prerender Buddy can calculate AI share of voice from the same answer records used for:

• prompt groups
• brand and competitor mentions
• recommendations
• citations and sources
• provider comparisons
• historical runs
• mapped page health

The dashboard should always expose the formula, counts, prompt scope and competitor set behind the percentage.

The Improve layer can then explain which prompt group contributes to the gap and whether the likely next action is technical, editorial, evidential or external.

What AI share of voice cannot prove

AI share of voice cannot prove:

• universal category leadership
• an AI provider's private selection logic
• search ranking
• referral traffic
• leads, conversions or revenue
• that a specific page change caused movement
• that the next answer will repeat the result

It is a competitive observation within a defined measurement system.

You may not need Prerender Buddy if

• Server HTML is already complete.
• Static pages crawl correctly.
• You are already rebuilding with SSR or static generation.
• You only need an audit, not a rendering fix.

Final recap

Define the denominator before reporting AI share of voice.

Use mention rate for the share of valid answers naming the brand. Use competitive AI share of voice for the brand's share of all tracked appearances. Keep mentions, recommendations and citations separate.

Report prompt groups and providers before the overall score. Preserve counts, methods and raw evidence. Compare the brand with its own consistent baseline rather than chasing an unsupported universal benchmark.

For client programs, expose the same formula and denominator in the agency AI visibility reporting framework. If the measurement reveals repeated factual errors, follow the brand-correction workflow.

Want to see how your observed AI visibility compares with relevant competitors? Start a Prerender Buddy project and measure the prompt groups behind the percentage.

Frequently asked questions

What is AI share of voice?

In Prerender Buddy's method, it is your brand's share of all observed appearances among your brand and a defined competitor set across controlled AI prompt runs.

How do I calculate AI share of voice?

Divide your brand appearances by the appearances of your brand plus tracked competitors, then multiply by 100. Count each recognized brand no more than once per answer unless another method is clearly disclosed.

Is AI mention rate the same as share of voice?

No. Mention rate is the percentage of valid answers naming your brand. Competitive share of voice compares your appearances with all tracked brand appearances.

Should citations count as share of voice?

Track citation share separately. A citation, mention and recommendation are different events and can move independently.

What is a good AI share-of-voice score?

There is no universal benchmark. Evaluate the trend within a stable prompt, competitor, provider, market and counting method. External benchmarks are useful only when the methodology is comparable.

Why did my AI share of voice change?

The answers may have changed, or the prompt set, providers, competitors, valid-run count, market or counting method may have changed. Check methodology before interpreting performance.

Does higher AI share of voice guarantee traffic?

No. Measure AI referrals and business outcomes separately. A mention or citation does not guarantee a click or conversion.