AI share of voice is the share of brand mentions your brand earns, out of all brand mentions, across a fixed set of prompts asked to AI engines such as ChatGPT, Perplexity and Google AI Mode. The basic formula is your mentions divided by total mentions, times 100. Semrush defines it as how visible your brand is compared to competitors in your market, and every tool I checked uses roughly that calculation.
As of 4 October 2026. Tools differ in how they count, so treat any single score as a trend line against your own baseline, not a universal benchmark.
Key Takeaways
- AI share of voice (AI SoV) is a ratio: your brand's mentions over all brand mentions in AI answers for a defined prompt set.
- There are two common variants: mention share of voice and citation share of voice. Report them separately.
- Every brand the AI names belongs in the denominator, not just the competitors you picked in advance.
- Sample 30 to 100 buyer-style prompts, run them across several engines, and add raw counts before dividing.
- Single snapshots mislead. Answers shift between runs, so track a trend.
- A low score is a diagnosis prompt: check entity recognition, topical association and sources, not just content volume.
AI share of voice: the definition
AI share of voice is a competitive visibility metric for AI-generated answers. It answers one question: when buyers ask AI tools about my category, what fraction of the brand conversation is mine?
HG Insights describes it as how often your brand shows up when buyers ask AI tools about your category, compared with everyone else competing for the same answer, and notes it covers mentions and citations across multiple engines. That is a good plain definition, and it separates AI SoV from classic share of voice, which counted ads, impressions or rankings.
A short glossary
| Term | Definition |
|---|---|
| AI share of voice | Your share of all brand mentions in AI answers for a tracked prompt set |
| Mention | A brand name appears in an AI answer, with or without a link |
| Citation | The AI credits your content as a source via link, footnote or attribution |
| Mention rate | Responses mentioning your brand divided by total responses |
| Citation rate | Responses citing your domain divided by total responses |
| Prompt set | The fixed list of questions you test repeatedly |
| Evaluation | One prompt run on one engine |
The formula, step by step
The core formula:
AI SoV (%) = your brand mentions ÷ total mentions of all brands × 100
LLM Pulse's version defines it as your mentions divided by mentions across your brand and selected competitors, and gives this worked example: 50 prompts across 5 AI models gives 250 evaluations. Brand X has 38 mentions, Competitor A has 52 and Competitor B has 30. So 38 ÷ (38 + 52 + 30) = 31.7%.
Notice what that example does and does not tell you. It tells you Brand X holds roughly a third of the named-brand conversation among those three. It does not tell you whether there was a fourth brand the AI mentioned that nobody tracked, which brings me to the main dispute in the field.
Open denominator vs closed denominator
Waikay's guide argues every brand the AI mentions should enter the denominator, not just the competitors you chose, because a pre-defined list is gameable and misses emerging rivals. LLM Pulse's example uses a selected competitor set. Both are defensible; they answer different questions.
| Approach | Denominator | Best for | Risk |
|---|---|---|---|
| Closed (named competitors) | You plus chosen rivals | Board reporting against known peers | Misses new entrants |
| Open (all brands named) | Every brand in the answers | Market mapping and diagnosis | Noisy; needs entity cleanup |
My practice: report the closed version for stakeholders, and keep the open version for diagnosis. State which one you are using in every report.
Mention share of voice vs citation share of voice
HG Insights separates two variants. Mention share of voice counts any time an answer names you, whatever the context. Citation share of voice counts when the model points to your content as a source, which suggests credibility and not just awareness.
Similarweb's AI search team makes the same distinction: a mention is a brand name without a link to your content, while a citation credits specific content via a clickable link, numbered footnote or inline attribution. Rankshift defines an AI citation as a visible reference to a source used by an AI search engine, and gives a citation rate formula: answers citing your domain divided by total monitored answers, times 100.
Why separate them? They move independently. You can be named constantly (high mention SoV) while a competitor's pages are the ones linked (higher citation SoV). The first builds awareness; the second can send traffic. Reporting one blended number hides which problem you have. For a deeper look, see my post on why a competitor ranks in AI answers instead of you.
AI SoV vs traditional share of voice
| Dimension | Traditional share of voice | AI share of voice |
|---|---|---|
| Unit counted | Ad impressions, rankings or media mentions | Brand mentions or citations in generated answers |
| Stability | Relatively stable week to week | Varies between runs and engines |
| Source of truth | Search console, ad platform, media monitor | Prompt sampling by a tracking tool |
| Controlled by you | Partly (bids, rankings) | Indirectly (content, entities, third-party sources) |
| Click link | Often direct | Only when cited and clicked |
How to measure AI share of voice in seven steps
- Define the category and competitors. Write down the market in one sentence and list five to ten rivals.
- Build the prompt set. LLM Pulse suggests a minimum of 30 to 100 prompts representing typical buyer questions. Include comparison and "best for" prompts, not only broad awareness questions.
- Choose engines. Cover ChatGPT, Perplexity, Gemini, Google AI Mode and AI Overviews where relevant. Single-engine tracking is flagged as unreliable by LLM Pulse and HG Insights.
- Run each prompt multiple times. Waikay stresses computing across multiple runs for stability.
- Record raw counts. Mentions and citations per brand, per engine.
- Add raw counts, then divide. LLM Pulse warns against averaging percentages across models; aggregate counts first.
- Repeat on a schedule. Weekly or fortnightly, with the same prompts, so the trend means something.
Semrush documents a tool-based route: its Brand Performance report in the AI Visibility Toolkit lets you enter a domain, set location and language, pick a platform and view a share of voice chart. Semrush notes its report factors in both mention frequency and position within responses, so its number may not match a plain mention ratio.
Worked example you can copy
Suppose you run an online course business and track 40 prompts across 4 engines, 160 evaluations.
| Brand | Mentions | Share of all named mentions |
|---|---|---|
| Your brand | 24 | 24 ÷ 96 = 25% |
| Rival A | 40 | 41.7% |
| Rival B | 20 | 20.8% |
| Rival C | 12 | 12.5% |
| Total | 96 | 100% |
These numbers are invented for illustration only. The point is the arithmetic: the denominator is total named mentions (96), not total evaluations (160). Mention rate is a different metric: if your brand appeared in 24 of 160 responses, mention rate is 15%.
Common mistakes with AI share of voice
- Confusing presence with share. Waikay's example: appearing in 30% of prompts is not owning 30% of the conversation if four other brands appear equally often.
- Position-weighting a single answer. Waikay calls position within one response probabilistic; frequency across many runs is more reliable. Semrush does weight position, so know what your tool does.
- Volume chasing. HG Insights warns against tracking only broad awareness queries rather than mid-to-deep-funnel comparison prompts where decisions happen.
- Sentiment blindness. A mention can be negative or wrong. Read a sample of answers.
- Snapshot reporting. HG Insights notes readings fluctuate with model retraining and re-crawling. One week is not a trend.
What a low score tells you (and what to do)
Waikay frames a weak score as diagnostic, not final. Check:
- Entity recognition. Does the AI know who you are, what you sell and where you operate?
- Topical association. Is your brand tied to the topics in your prompts?
- Source presence. Are the sites AI engines draw on, such as industry publications and review sites, mentioning you?
- Content coverage. Do you have pages that directly answer the comparison questions?
Google says there are no additional requirements to appear in AI Overviews or AI Mode beyond SEO fundamentals, per its AI features documentation, so there is no special trick; the work is the fundamentals plus being credibly present across the web. If you are weighing paid tools to track this, read my honest look at whether AI visibility tools are worth it.
Benchmarks: is there a "good" score?
No universal benchmark exists, according to LLM Pulse. A 20% score in a crowded category of ten brands is strong; 20% in a three-brand market is weak. Compare your trend against the same competitors and prompt set over time. For evidence on what actually moves these numbers, see my roundup of AEO case studies and evidence.
FAQ
What is AI share of voice in simple terms?
It is the percentage of brand mentions in AI answers that belong to you, out of all brands mentioned for a fixed set of prompts. If AI tools name five brands and yours appears in a quarter of those mentions, your AI SoV is 25%.
What is the AI share of voice formula?
Your brand mentions divided by total brand mentions, multiplied by 100. Semrush, HG Insights and LLM Pulse all give this basic form, differing mainly on which brands count in the denominator.
Is AI share of voice the same as citation rate?
No. Citation rate is the share of responses that cite your domain. Share of voice is your share of all brand mentions. Rankshift gives citation rate as answers citing your domain divided by total monitored answers.
How many prompts do I need?
LLM Pulse recommends at least 30 to 100 prompts representing typical buyer questions. Fewer than that makes the number jump around too much to trust.
Which AI engines should I track?
Track more than one. LLM Pulse lists ChatGPT, Perplexity, Gemini, Google AI Mode and AI Overviews, and HG Insights notes these engines often return different answers.
How often should I measure it?
Weekly is a common cadence in the guides I reviewed. What matters is keeping the prompt set and engines identical so the trend is comparable.
What is a good AI share of voice?
There is no universal good score. It depends on how many brands compete in the answers. Judge by trend and by gap to your closest rivals.
Does a higher AI SoV mean more traffic?
Not automatically. Mentions build awareness; citations are more likely to drive clicks, as Similarweb's team puts it. Check referral traffic separately.
Can I track AI share of voice without paid tools?
Yes, manually, with a spreadsheet and a fixed prompt list, though it is slow and sampling noise is higher. Paid tools automate repeated runs. See the Conductor AEO glossary for the wider term set.
Want AI visibility measured properly?
If you would like help building a prompt set, a baseline and a trend report for your brand, take a look at my work at younusfardeen.in and send me a note through the contact form. I am an organic growth and AEO marketer with 4+ years of marketing experience, and I like measurement that survives scrutiny.