Most tools are measuring AI Share of Voice wrong. It's not a rank position. It's the percentage of relevant AI responses that mention your brand. Here's the correct formula, why mention rate and SoV are different, and exactly what to do to move the number in 30 days.
Why traditional SoV formulas break in AI search
In traditional search, Share of Voice is a rank-weighted metric: a page ranking #1 contributes more to your SoV than a page ranking #5. Position is everything because users see results in order and click the top results at dramatically higher rates.
AI search doesn't work this way. There is no "position 1" and "position 5" in a ChatGPT response. There's mentioned, and there's not mentioned. A brand mentioned as the first recommendation and a brand mentioned as a caveat are both "in the answer" — but they're not the same thing.
The correct way to measure AI Share of Voice is simpler but requires different data: out of all the relevant AI responses in your category, what percentage mention your brand at all?
SoV = (Brand Mentions ÷ Total Prompt Responses) × 100
Measured across all tracked prompts, per time period, per AI platform.
Mention rate vs. Share of Voice — why the distinction matters
Mention rate is the raw number of times your brand appeared in AI responses across all your tracked prompts.
Share of Voice is the percentage of total possible mentions your brand captured — measured against the total responses across all tracked prompts.
These can diverge significantly. If you add 50 new prompts to your tracking, your mention rate might stay the same (you're still mentioned in 35 responses), but your SoV drops from 35% to 23% because the denominator (total responses) increased. Tracking mention rate alone makes your performance look stable when it's actually declining relative to your expanded category.
Hema calculates SoV as the percentage figure — not raw mentions — so you always see your performance relative to your tracked category.
Reading the Hema SoV chart
The Share of Voice chart in Hema's Dashboard shows three things simultaneously:
- •Your SoV trend line — weekly aggregate across all tracked prompts
- •Per-platform breakdown — your SoV on ChatGPT vs Perplexity vs Gemini (each platform has very different answer patterns)
- •Competitor comparison — your SoV vs each tracked competitor, with H2H win rate shown as "days you outperformed each rival"
5 actions that move SoV in 30 days
Ranked by speed of measurable impact — fastest first.
1. Fix Critical audit issues (impact in 24–72 hours)
A firewall rule blocking GPTBot or a critical JavaScript rendering failure can reduce your SoV to near-zero on a specific platform. Fixing these is the fastest possible SoV improvement — often 10–20 percentage point jumps within 48 hours of a fix being deployed.
2. Add FAQPage schema to your top 5 pages (impact in 3–7 days)
FAQ schema makes your content individually parseable at the question level. AI platforms can cite a specific FAQ answer without citing the full page — this increases the surface area of your citeable content significantly.
3. Publish one "vs." comparison article per week (impact in 2–4 weeks)
Comparison queries are the highest-value prompt type for SoV growth. A well-structured "Your Brand vs Competitor" article with Product schema, PriceSpecification, and FAQPage schema will typically begin appearing in AI comparison query responses within 2–3 weeks of publication.
4. Add PriceSpecification schema to your pricing page (impact in 1–2 weeks)
Missing pricing schema causes AI to either ignore your pricing or cite it incorrectly. Adding PriceSpecification JSON-LD to every pricing plan makes your pricing extractable as structured data.
5. Expand tracked prompts to cover all fanout topics (impact in 3–6 weeks)
If you're only tracking 10 prompts, your SoV measurement is based on a narrow slice of your category. Expanding to 50–100 prompts gives you both more accurate SoV data and more opportunities for your SoV to grow.
Case example: Stackflow (B2B SaaS, project management) went from 4% to 22% SoV across tracked prompts in 47 days. Day 1: removed GPTBot from CSP blocklist (+18 points). Days 14–28: published 4 comparison articles (+6 points). Days 28–47: added FAQPage schema to 12 pages (+4 points). Total: +18 percentage points over 47 days.