To measure Share of Voice in AI answers accurately, calculate three metrics separately over a fixed unbranded prompt panel and locked competitor list: Mention-Based Share of Voice (brand mentions divided by total tracked brand mentions), Citation-Based Share of Voice (your domain links divided by all cited links), and Position-Weighted Share of Voice averaged across repeated runs.
1. Three Formulas Behind AI Share of Voice
Classic SEO Share of Voice multiplied keyword search volume by CTR-weighted organic rankings. AI answers have no static ten-blue-link list, so teams measure probabilistic presence using three distinct formulas:
Mention-based SoV is clearest for executive reporting, citation-based SoV guides content and PR strategy, and position-weighted SoV captures prominence when assistants list multiple brands.
2. Worked Calculation Across 100 Prompts and 5 Brands
Using the same underlying 100-prompt answer log across five competing software brands produces completely different percentages depending on which unit you count:
| Formula | Worked Math | Result | Best Use |
|---|---|---|---|
| Mention-Based SoV | 60 brand mentions / 300 total rival mentions | 20.0% | Executive category presence |
| Citation-Based SoV | 25 domain citations / 1,200 total source links | 2.1% | Content & publisher gap analysis |
| Position-Weighted SoV | Harmonic weights (1, 0.5, 0.33, 0.25) vs. rivals | 26.4% | Top-of-list shortlist prominence |
3. Four Measurement Rules to Eliminate Sampling Noise
Because LLM outputs vary across runs and model updates, follow four measurement rules before presenting month-over-month deltas:
- Run high-priority competitive prompts 3 to 5 times per cycle or use rolling 4-week averages to smooth re-ask variance.
- Freeze your competitor denominator for the full quarter and annotate any list changes, alongside raw Answer Presence rate.
- Separate unbranded discovery prompts from branded prompts so brand-name queries do not inflate category SoV.
- Break out ChatGPT, Perplexity, Gemini, and Google AI Overviews individually instead of hiding engine divergence inside one blended index.
Compare platforms in our best Share of Voice tools guide and review our AI visibility metrics reference.
Frequently asked questions
Why is Citation Share of Voice usually much lower than Mention Share of Voice?
Mention SoV divides by mentions of tracked competitors, whereas Citation SoV often divides by all cited URLs across publishers, Reddit, Wikipedia, and review sites, creating a much larger denominator.
How does Position-Weighted AI Share of Voice work?
Most tools apply harmonic decay weights to ordered brand lists (1.0 for first named, 0.50 for second, 0.33 for third, 0.25 for fourth) and divide your brand's weighted sum by the category total.
Can I measure AI Share of Voice accurately with one prompt run per month?
No. Single runs suffer from probabilistic sampling noise, especially on ordered lists. Weekly runs or multi-repeat sampling per cycle produce far more reliable trend lines.
Next steps & related pages
Want a baseline visibility check for your brand?
Save the CSV checklist above for manual audits, or request a baseline check on your target US buyer prompt while our automated tracking workspace rolls out.