AI share of voice benchmarking
AI share of voice is the percentage of brand mentions your company wins in AI-generated answers, compared with the named competitors in the same answers, across a fixed set of prompts. To benchmark it, you run the same prompts on the same assistants every week and score every brand by the same rules. The trend shows where you stand in ChatGPT and Google AI Overviews and which buyer questions you can still win.
Jevgenia PogadajevaCFO, Snoika · 6 min read
Benchmarking AI share of voice
To benchmark AI share of voice, keep everything the same except the brand, then repeat the measurement until the run-to-run noise evens out. If you change the prompts or the platform halfway through, the numbers stop being comparable. Changes to the scoring rules also break comparability.
- Use competitor analysis to name the 3 to 6 rivals your buyers actually shortlist. Leave out the companies you only wish you competed with.
- Write the prompt set from real buyer questions, then tag each prompt with its intent group.
- Pick the assistants and use logged-out sessions. Keep the region and language fixed.
- Run every prompt several times in each cycle and record every brand named in every answer.
- Score mentions and position with fixed rules. Apply fixed rules to citation and sentiment scores as well.
- Repeat weekly and compare trend lines instead of single readings.
Step 4 is the one people skip, and it's the one that decides whether your benchmark means anything. SparkToro found a less than 1-in-100 chance of ChatGPT or Google's AI returning the same brand list across repeated runs of one prompt. "If you ask an AI tool for brand/product recommendations a hundred times nearly every response will be unique," said Rand Fishkin, SparkToro's CEO. So a single run tells you little. Use a share averaged over dozens of runs to guide decisions.
AI share of voice defined
AI share of voice means your brand's mentions divided by all mentions of the brands you track, across one prompt set on one assistant. Say ChatGPT names your brand 40 times and your four rivals 160 times in total across a week of runs. Your share of voice is 20%. The tracked brands split 100% between them, so every point you gain comes out of a competitor's share.
People confuse it with visibility rate, which is the percentage of answers that mention you at all. You can appear in 60% of answers and still hold a small share if each answer lists eight brands and you're always last. Both numbers belong on the dashboard, but they answer different questions. AI visibility measures whether you show up, while share of voice measures how much of the conversation you hold against the brands your buyer is choosing between.
Choosing benchmark assistants
Benchmark the assistants your buyers use and begin with ChatGPT and Google AI Overviews, because that's where most AI search traffic happens. OpenAI CEO Sam Altman said in October 2025 that ChatGPT had reached 800 million weekly users. Alphabet reported in its second-quarter 2026 earnings that AI Overviews reached more than 2.5 billion users a month.
After those two, the right list depends on your audience. Referral traffic from Claude or Perplexity differs between developer tools and consumer brands. Check the AI referral sources in your Google Analytics 4 (GA4) data before you add a platform. Then score each assistant separately. If you average ChatGPT and AI Overviews into one number, you hide the gap you most need to see.
Segmenting by prompt group
Track category discovery and comparison prompts for AI share of voice. Add problem and use-case prompts and branded prompts to make four groups. Score each group on its own, because a mention inside a purchase recommendation is worth more than a mention inside a definition.
Category discovery prompts
These are the "best [category] for [audience]" questions buyers ask at the start of their research. Here the model chooses freely among brands, which makes this group the fairest test of your standing. Give it the largest share of your prompt budget.
Comparison and alternative prompts
"[Competitor] vs [competitor]" and "alternatives to [competitor]" are competitor analysis prompts that show you whether assistants place you on the shortlist at all. If you lose this group, you're losing buyers who are close to a decision.
Problem and use-case prompts
These describe the buyer's situation without naming a category, for example "how do I file VAT in six EU countries." They tell you whether the assistant connects your brand to the job it does.
Branded prompts
"Is [your brand] good" and "[your brand] pricing" check accuracy and sentiment. If they make up more than a small minority of the set, they inflate your score without measuring your share against competitors.
Setting realistic targets
A realistic AI share of voice target is set against your starting baseline and the category leader's share. A universal percentage won't work. In a five-brand set, an equal split is 20% each, so a challenger at 5% has a clear first milestone. A leader at 45% needs to hold that share when a rival publishes something better.
The pace depends on your starting point and the prompt group. Low bases can move fast. Snoika's founder Anton Vedeshin reported to Startup Reporter 1StopVAT's move from 2.7% to 28.3% ChatGPT visibility in 16 weeks. Comparison prompts that a rival already dominates move slowly, because the same brand keeps winning them. Set a separate target for each prompt group and assistant. Review them every quarter, since four weeks of data rarely separates a real shift from noise.
Metrics your dashboard needs
Your dashboard needs share of voice, plus the metrics that explain why it moves. A share figure with nothing beside it tells you that something changed but never why.
| Metric | What it counts | What to check |
|---|---|---|
| Share of voice | Your mentions out of all tracked-brand mentions | Trend by assistant and prompt group |
| Visibility rate | Answers that mention you, out of all answers | Gaps where rivals appear and you don't |
| First-mention rate | Answers where you're the first brand named | Whether you're recommended or only listed |
| Citation share | Answers that link to your pages as a source | Which of your URLs assistants quote |
| Sentiment | How the answer describes you | Outdated or wrong claims about your product |
| Run variance | Spread of your share across repeat runs | Whether a change is bigger than the noise |
Share of voice
- What it counts
- Your mentions out of all tracked-brand mentions
- What to check
- Trend by assistant and prompt group
Visibility rate
- What it counts
- Answers that mention you, out of all answers
- What to check
- Gaps where rivals appear and you don't
First-mention rate
- What it counts
- Answers where you're the first brand named
- What to check
- Whether you're recommended or only listed
Citation share
- What it counts
- Answers that link to your pages as a source
- What to check
- Which of your URLs assistants quote
Sentiment
- What it counts
- How the answer describes you
- What to check
- Outdated or wrong claims about your product
Run variance
- What it counts
- Spread of your share across repeat runs
- What to check
- Whether a change is bigger than the noise
When answers cite your pages, you can see which content the model trusts. You can then fix what it doesn't trust. Run variance keeps the team honest because a week-to-week change smaller than the normal spread means nothing has happened yet.
How Snoika runs benchmarks
Snoika runs benchmarks as weekly prompt sets in ChatGPT and Google AI Overviews, and measures mentions and share of voice for the client and its rivals. It also measures sentiment and citations. Each engagement starts with onboarding. The team learns about the category and its buyers in relation to the goals. Then it asks for access to GA4 and Google Search Console. The access request also covers Google Tag Manager so the prompt set reflects real demand and the results can be tied to traffic.
Every plan includes competitor tracking, and a person reviews the automated results before anyone acts on them. That review catches errors that scoring rules miss, such as a rival mentioned under an old product name.
Turning benchmarks into actions
Turn each benchmark gap into a specific page, and match the page to the prompt group you're losing. Address losses on comparison prompts with honest comparison content, and use pages that tie your product to the buyer's situation in plain terms to address losses on problem prompts. Wrong branded answers trace back to outdated pages the model still quotes.
Mentions count even when nobody clicks. Pew Research Center found that Google users who saw an AI summary clicked a result in 8% of visits, compared with 15% when no summary appeared. Being named in the answer is the whole visit. And the work pays off when it targets the right prompts. 3DPrinterOS reached #1 visibility in ChatGPT for industry-specific queries and turned AI search into a steady source of 3–4 qualified inbound leads/day. After each publishing cycle, rerun the same prompt group so you can see whether the new page moved it.
Next steps
Your next step is to record a baseline before you change any content. Without one, you can't tell which page moved your AI share of voice, and you can't defend the budget next quarter. Write down your competitor list and your four prompt groups first, then measure for at least two weekly cycles. Snoika tracks share of voice and mentions in ChatGPT and Google AI Overviews. We also track citations and can show you where your brand stands today. Get a free AI visibility report for a starting baseline.
FAQ
Your ai share of voice is your brand’s mentions divided by all tracked-brand mentions across a fixed prompt set on one assistant. If you receive 40 mentions out of 200, your share is 20%. It measures your share against the competitors you track.
Track category discovery prompts alongside comparison and alternative prompts. Include problem and use-case prompts, plus branded questions about your product. Give category discovery the largest share of your prompt budget, and keep branded questions to a small minority to avoid inflating the score.
A realistic target depends on your baseline and the category leader’s share. An equal split in a five-brand set is 20%, but that isn’t a universal goal. Set separate targets for each assistant and prompt group.
Changing the tracked competitors changes the denominator, so the resulting shares aren’t directly comparable with earlier results. Keep the competitor set fixed when measuring trends. If it needs to change, establish a new baseline before judging gains or losses.
Look for a sustained increase that exceeds the normal spread across repeated runs. Average results over dozens of runs rather than relying on one answer. Review trends quarterly, since four weeks of data rarely separates a real shift from noise.
