By maxaeo.ai ｜ Published 2026-10-02 ｜ Updated 2026-10-02
To calculate LLM share of voice, measure how often your brand appears in a fixed set of buyer prompts, compare those mentions with competitors, and then apply consistent weights for position, engine, sentiment, and citations. The basic percentage is useful, but a defensible measurement model must separate visibility from recommendation quality.
What is LLM share of voice?
LLM share of voice is the percentage of competitive brand mentions owned by one brand across a defined set of AI-generated answers. Unlike traditional search share of voice, one answer can mention several brands, recommend one brand first, cite another brand’s website, and describe a third brand negatively.
A simple formula is:
Mention Share of Voice =
Your Brand Mentions ÷ Total Tracked Brand Mentions × 100
For example, if a prompt set produces 240 total mentions across your brand and four competitors, and your brand appears 48 times:
48 ÷ 240 × 100 = 20% mention share
This is a zero-sum share of mentions model. It differs from mention rate, which measures your appearances against the number of valid answers:
Mention Rate =
Answers Mentioning Your Brand ÷ Valid Answers × 100
The distinction matters. If an answer mentions five companies, each company can increase its mention rate, but the share of total mentions still has to be allocated among the brands being tracked. Current measurement guides commonly treat these as separate metrics, although many platforms use “share of voice” for only one of them. (llmpulse.ai)
What data do you need before calculating it?
A reliable calculation starts with a controlled measurement set. Record the following fields for every prompt and answer:
| Field | What to record |
|---|---|
| Prompt | The exact buyer question used |
| AI engine | ChatGPT, Perplexity, Gemini, Claude, or another target engine |
| Date | When the answer was collected |
| Brand mentions | Every tracked brand named in the answer |
| Position | First, second, third, or another clearly defined position |
| Citation | Whether the brand or its domain was cited |
| Sentiment | Positive, neutral, negative, or inaccurate |
| Prompt intent | Category, comparison, alternatives, use case, or pricing |
Use the same competitor set throughout a reporting period. Changing competitors halfway through the month can create a false increase or decrease in share.
The prompt set should represent real buying situations rather than only branded queries. A SaaS team might track prompts such as:
- “What are the best tools for monitoring AI search visibility?”
- “Compare platforms for tracking ChatGPT brand mentions.”
- “What is a good alternative to [competitor] for a SaaS marketing team?”
- “Which AI visibility tools support competitor benchmarking?”
- “What should an enterprise evaluate before buying an AEO platform?”
For a practical measurement framework, separate prompts into discovery, comparison, problem-solution, and brand-specific groups. This reveals whether a brand is visible only when users already know its name or also appears during category discovery.
How do you calculate cross-engine mention share?
Calculate each engine separately before producing a combined score. This prevents one platform with unusually long answers or more frequent brand mentions from dominating the result.
Step 1: Calculate engine-level mention rate
Engine Mention Rate =
Brand-Mentioning Answers on Engine ÷ Valid Answers on Engine × 100
Example:
- ChatGPT: 18 brand mentions in 40 valid answers = 45%
- Perplexity: 12 brand mentions in 40 valid answers = 30%
- Gemini: 10 brand mentions in 40 valid answers = 25%
Step 2: Calculate engine-level share of mentions
If the same three engines generate 80 total tracked mentions, and your brand receives 40 of them:
Engine-Aggregated Mention Share =
40 ÷ 80 × 100 = 50%
Step 3: Weight engines only when the business case justifies it
An unweighted average gives every engine equal importance:
Unweighted SOV =
(ChatGPT SOV + Perplexity SOV + Gemini SOV) ÷ 3
A weighted average is more appropriate when your audience or market is concentrated on specific platforms:
Weighted SOV =
Σ (Engine SOV × Engine Weight)
The engine weights must add up to 1.00. For example:
- ChatGPT: 0.40
- Perplexity: 0.25
- Gemini: 0.20
- Claude: 0.15
Do not present weighted SOV as a universal market truth. It is a decision model based on your audience, distribution, or measurement policy.
How should answer position affect the score?
Mention count alone treats a passing reference and a first recommendation as equal. A position-weighted score provides a better view of competitive prominence.
One practical weighting model is:
| Position | Suggested weight |
|---|---|
| First recommendation | 1.00 |
| Second recommendation | 0.75 |
| Third recommendation | 0.50 |
| Other named mention | 0.25 |
| Negative or inaccurate mention | 0.00 or tracked separately |
For each answer, calculate:
Position Score =
Σ Brand Position Weights ÷ Maximum Possible Position Score
Suppose your brand appears in four answers:
- First position once: 1.00
- Second position once: 0.75
- Third position once: 0.50
- Unranked mention once: 0.25
Your total position score is:
1.00 + 0.75 + 0.50 + 0.25 = 2.50
The important reporting rule is to publish the weighting policy beside the result. A score based on first-position recommendations cannot be compared fairly with a score based on simple brand mentions.
A better model: visibility, prominence, and evidence
The most useful improvement is to avoid forcing every signal into one number. Report three related metrics instead.
1. Visibility Share
This answers: How often is the brand present?
Visibility Share =
Your Brand Mentions ÷ Total Tracked Brand Mentions × 100
2. Prominence Share
This answers: How much of the answer’s recommendation space does the brand occupy?
Prominence Share =
Your Weighted Position Points ÷ All Brand Position Points × 100
3. Evidence Share
This answers: How often is the brand supported by a source?
Evidence Share =
Your Cited Mentions ÷ Total Cited Brand Mentions × 100
A brand can have high visibility but low prominence if it is repeatedly mentioned as an alternative. It can also have high prominence but low evidence share if AI answers recommend it without citing authoritative sources.
This three-part model is an original reporting improvement because it distinguishes being named, being preferred, and being supported. A single blended score can hide those differences.
How should sentiment and accuracy be handled?
Sentiment and factual accuracy should not be silently mixed into the basic share-of-voice formula. Track them as quality modifiers or separate diagnostic metrics.
For example:
Qualified Mention Rate =
Accurate Positive or Neutral Mentions ÷ Total Valid Answers × 100
You can also create an accuracy-adjusted score:
Accuracy-Adjusted SOV =
Mention Share × Accuracy Rate
If a brand owns 30% mention share but only 80% of its descriptions are accurate:
30% × 0.80 = 24% accuracy-adjusted SOV
Keep negative mentions visible rather than deleting them. A brand that appears frequently because AI systems describe it as unreliable should not receive the same interpretation as a brand frequently recommended for the right use case.
MaxAEO supports sentiment analysis and factual accuracy checks for AI answers, alongside mention tracking, ranking, citation analysis, and competitor comparisons across eight AI engines. Its daily monitoring records original answers so teams can trace the sentence behind a metric.
How often should LLM share of voice be measured?
Daily collection is useful for detecting changes, but weekly or monthly reporting is usually easier to interpret. AI answers can vary between runs, so avoid treating one response as a market movement.
Use this operating rhythm:
- Daily: Collect the fixed prompt set and store raw answers.
- Weekly: Review engine-level changes, new citations, and unusual competitor movement.
- Monthly: Report visibility share, prominence share, evidence share, sentiment, and accuracy.
- After a content change: Compare the same prompts before and after publication.
Keep the prompt wording, competitor list, geography, language, and engine mix stable during a comparison period. If any of these change, label the result as a new benchmark rather than a direct trend.
For more context, see MaxAEO’s enterprise LLM share of voice reporting framework and AI engine competitive analysis framework.
Common questions
Is LLM share of voice the same as AI mention rate?
No. Mention rate measures your brand’s appearances against valid answers. LLM share of voice compares your brand’s mentions with the total mentions captured by your tracked competitors.
Should citations count as mentions?
Only if the brand is also clearly mentioned in the answer. Track citations separately because a cited domain can support an answer without the brand being recommended prominently.
Should every AI engine have the same weight?
Not always. Equal weighting is easier to audit, while audience-based weighting can better reflect your market. Publish the weighting policy so stakeholders understand the result.
How many prompts are enough?
There is no universal number. Start with a stable set covering your highest-value buyer intents, then expand it when you identify meaningful gaps. Consistency is more important than using a large, constantly changing list.
Can a SaaS company monitor this without technical integration?
Yes. MaxAEO provides a free AI visibility diagnosis using a brand name, website, and competitor information. Its paid monitoring plans track daily visibility across eight AI engines, including mention rate, competitive position, sentiment, and citation sources.
Final takeaway
The simplest way to calculate LLM share of voice is to divide your brand’s tracked mentions by all tracked competitor mentions. For decision-grade reporting, add position weights, engine segmentation, citation evidence, sentiment, and factual accuracy.
Start with a transparent baseline, preserve the raw answers, and report the three numbers that matter most: visibility share, prominence share, and evidence share. A free MaxAEO diagnosis can help establish that baseline across major AI search platforms.