How Yozigo measures AI visibility

How are prompts run?

A tracked prompt is one buyer question, such as “best AI visibility tools for agencies”. Each brand runs once a day in its own slot. Every prompt is sent to each engine in the plan (ChatGPT, Gemini, Perplexity, Google AI Overviews, Microsoft Copilot, Google AI Mode) from the markets the brand selects. Prompts can run in 12 regions.

A prompt run on six engines counts as one tracked prompt. Results are daily values shown with a 7-day moving average, so one noisy day does not hide the direction.

What is recorded for each answer?

• The full answer text, the engine, the region and the time.
• Every source the engine cited, with its URL and domain.
• Whether the brand and each tracked competitor is named, where it ranks among the brands listed, and the sentiment of the mention.
• The web searches the engine ran before answering (fan-out queries), where the engine exposes them. ChatGPT and Microsoft Copilot expose them on most answers and Perplexity on some; Gemini, Google AI Overviews and AI Mode do not.

How is a mention or citation counted?

A mention is the brand name (or a configured alias) appearing in the answer text. A citation is a link to the brand's domain in the answer's sources. They are tracked separately because AI often cites a site without naming the brand, and names a brand without citing it. Cited sources are grouped into types: Your site, Competitor site, Third-party company, Social media, News & press, Review platform, Knowledge hub, Business listing, Other.

How is each metric calculated?

• Visibility: Responses that mention your brand ÷ all responses × 100.
• Position: Median rank of your brand among brands named, counted only where you are mentioned. Top-position rate is the share at #1–3.
• Sentiment: Average of positive (+1), neutral (0), and negative (−1) responses, scaled to 0–100.
• Grounding rate: Responses citing your domain as a source ÷ all responses × 100.
• Share of voice: Your mentions ÷ (your mentions + tracked competitor mentions) × 100.
• Confidence: High at 15+ responses, medium at 5–14, low under 5.

Why are branded and unbranded prompts split?

A question that already names the brand (“Is Yozigo worth it?”) almost always gets a mention. Mixing those with category questions makes visibility look far higher than it is. Yozigo reports them separately; the unbranded view is the one that shows whether AI recommends you to people who have not heard of you. Our study of this effect is in branded vs unbranded AI visibility.

How are AI crawlers verified?

AI Traffic recognises 48 AI bots. A request that claims to be one of them is checked against the vendor's published IP list, reverse DNS and signed agent headers (Web Bot Auth). Requests that fail are reported as unverified, not counted as real crawls. See AI Traffic and our guide to verifying AI crawler traffic.

What are the known limits?

• Prompts refresh daily, not in real time — enough to see movement without chasing noise.
• Entry plans track four engines (ChatGPT, Perplexity, AI Overviews, Gemini); Momentum and Authority track all six.
• Branded prompts naturally lift your numbers. Switch to Discovery to see unbranded performance.
• Position is measured only in answers that mention you, so check it alongside visibility.
• AI answers vary between sessions and users. Every figure is a sample of daily runs, not a fixed ranking, so read trends rather than single days.

Where does Yozigo publish its own research?

We publish studies built from this data, such as which sources AI engines cite and what query fan-out is. Each one states its sample size and date range. Facts about the product itself are on the AI facts page.